def create_noop_environment():
oneday = timedelta(days=1)
start = datetime(2006, 1, 1, tzinfo=pytz.utc)
bm_returns = []
tr_curves = OrderedDict()
for day in date_gen(start=start, delta=oneday, count=252):
dr = DailyReturn(day, 0.01)
bm_returns.append(dr)
curve = {
'10year': 0.0799,
'1month': 0.0799,
'1year': 0.0785,
'20year': 0.0765,
'2year': 0.0794,
'30year': 0.0804,
'3month': 0.0789,
'3year': 0.0796,
'5year': 0.0792,
'6month': 0.0794,
'7year': 0.0804,
'tid': 1752
}
tr_curves[day] = curve
load_nodata = lambda x: (bm_returns, tr_curves)
return trading.TradingEnvironment(load=load_nodata)
def get_benchmark_returns(symbol, start_date=None, end_date=None):
"""
Returns a list of return percentages in chronological order.
"""
if start_date is None:
start_date = datetime(year=1950, month=1, day=3)
if end_date is None:
end_date = datetime.utcnow()
# Get the benchmark data and convert it to a list in chronological order.
data_points = list(get_benchmark_data(symbol, start_date, end_date))
data_points.reverse()
# Calculate the return percentages.
benchmark_returns = []
for i, data_point in enumerate(data_points):
if i == 0:
curr_open = data_points[i]['open']
returns = (data_points[i]['close'] - curr_open) / curr_open
else:
prev_close = data_points[i - 1]['close']
returns = (data_point['close'] - prev_close) / prev_close
daily_return = DailyReturn(date=data_point['date'], returns=returns)
benchmark_returns.append(daily_return)
return benchmark_returns
def create_returns_from_range(sim_params):
current = sim_params.first_open
end = sim_params.last_close
test_range = []
while current <= end:
r = DailyReturn(current, random.random())
test_range.append(r)
current = trading.environment.next_trading_day(current)
return test_range
def create_returns_from_range(sim_params):
current = sim_params.first_open
end = sim_params.last_close
one_day = timedelta(days=1)
test_range = []
while current <= end:
r = DailyReturn(current, random.random())
test_range.append(r)
current = get_next_trading_dt(current, one_day)
return test_range
def create_returns_from_list(returns, sim_params):
current = sim_params.first_open
test_range = []
# sometimes the range starts with a non-trading day.
if not trading.environment.is_trading_day(current):
current = trading.environment.next_trading_day(current)
for return_val in returns:
r = DailyReturn(current, return_val)
test_range.append(r)
current = trading.environment.next_trading_day(current)
return test_range
def get_benchmark_returns(symbol, start_date=None, end_date=None):
if start_date is None:
start_date = datetime(year=1950, month=1, day=3)
if end_date is None:
end_date = datetime.utcnow()
benchmark_returns = []
for data_point in get_benchmark_data(symbol, start_date, end_date):
returns = (data_point['close'] - data_point['open']) / \
data_point['open']
daily_return = DailyReturn(date=data_point['date'], returns=returns)
benchmark_returns.append(daily_return)
return benchmark_returns
def create_returns(daycount, sim_params):
"""
For the given number of calendar (not trading) days return all the trading
days between start and start + daycount.
"""
test_range = []
current = sim_params.first_open
one_day = timedelta(days=1)
for day in range(daycount):
current = current + one_day
if trading.environment.is_trading_day(current):
r = DailyReturn(current, random.random())
test_range.append(r)
return test_range
def load_market_data(self, bm_symbol='^GSPC'):
#TODO Parametric
#event_dt = datetime.today().replace(tzinfo=pytz.utc)
event_dt = datetime.now(pytz.utc)
# Getting today benchmark return
#NOTE Seems shit but later, previous days could be used to compute indicators
#last_bench_return = get_benchmark_returns(bm_symbol, start_date=(event_dt - pd.datetools.Day(self.loopback)))
#last_bench_return = last_bench_return[-1]
#print('Benchmark on {}: {}'.format(last_bench_return.date, last_bench_return.returns))
for exchange, infos in datautils.Exchange.iteritems():
if infos['index'] == bm_symbol:
code = datautils.Exchange[exchange]['code']
break
bm_returns = []
while event_dt < self.last_trading_day:
#TODO Current value to give
#TODO Append only if trading day and market hour
#bm_returns.append(DailyReturn(date=event_dt.replace(microsecond=0), returns=last_bench_return.returns))
bm_returns.append(DailyReturn(date=event_dt.replace(microsecond=0), returns=code))
#TODO Frequency control
event_dt += self.offset
bm_returns = sorted(bm_returns, key=attrgetter('date'))
tr_gen = get_treasury_data()
while True:
try:
last_tr = tr_gen.next()
except StopIteration:
break
tr_curves = {}
#tr_dt = datetime.today().replace(tzinfo=pytz.utc)
tr_dt = datetime.now(pytz.utc)
while tr_dt < self.last_trading_day:
#tr_dt = tr_dt.replace(hour=0, minute=0, second=0, tzinfo=pytz.utc)
tr_curves[tr_dt] = last_tr
tr_dt += self.offset
tr_curves = OrderedDict(sorted(
((dt, c) for dt, c in tr_curves.iteritems()),
key=lambda t: t[0]))
return bm_returns, tr_curves
def test_risk_metrics_returns(self):
trading.environment = trading.TradingEnvironment()
# Advance start date to first date in the trading calendar
if trading.environment.is_trading_day(self.start_date):
start_date = self.start_date
else:
start_date = trading.environment.next_trading_day(self.start_date)
self.all_benchmark_returns = pd.Series({
x.date: x.returns
for x in trading.environment.benchmark_returns
if x.date >= self.start_date
})
start_index = trading.environment.trading_days.searchsorted(start_date)
end_date = trading.environment.trading_days[start_index + len(RETURNS)]
sim_params = SimulationParameters(start_date, end_date)
risk_metrics_refactor = risk.RiskMetricsIterative(sim_params)
todays_date = start_date
cur_returns = []
for i, ret in enumerate(RETURNS):
todays_return_obj = DailyReturn(todays_date, ret)
cur_returns.append(todays_return_obj)
try:
risk_metrics_original = risk.RiskMetricsBatch(
start_date=start_date,
end_date=todays_date,
returns=cur_returns)
except Exception as e:
#assert that when original raises exception, same
#exception is raised by risk_metrics_refactor
np.testing.assert_raises(
type(e), risk_metrics_refactor.update, todays_date,
self.all_benchmark_returns[todays_return_obj.date])
continue
risk_metrics_refactor.update(
todays_date, ret,
self.all_benchmark_returns[todays_return_obj.date])
# Move forward day counter to next trading day
todays_date = trading.environment.next_trading_day(todays_date)
self.assertEqual(risk_metrics_original.start_date,
risk_metrics_refactor.start_date)
self.assertEqual(risk_metrics_original.end_date,
risk_metrics_refactor.algorithm_returns.index[-1])
self.assertEqual(risk_metrics_original.treasury_period_return,
risk_metrics_refactor.treasury_period_return)
np.testing.assert_allclose(risk_metrics_original.benchmark_returns,
risk_metrics_refactor.benchmark_returns,
rtol=0.001)
np.testing.assert_allclose(risk_metrics_original.algorithm_returns,
risk_metrics_refactor.algorithm_returns,
rtol=0.001)
risk_original_dict = risk_metrics_original.to_dict()
risk_refactor_dict = risk_metrics_refactor.to_dict()
self.assertEqual(set(risk_original_dict.keys()),
set(risk_refactor_dict.keys()))
err_msg_format = """\
"In update step {iter}: {measure} should be {truth} but is {returned}!"""
for measure in risk_original_dict.iterkeys():
if measure == 'max_drawdown':
np.testing.assert_almost_equal(
risk_refactor_dict[measure],
risk_original_dict[measure],
err_msg=err_msg_format.format(
iter=i,
measure=measure,
truth=risk_original_dict[measure],
returned=risk_refactor_dict[measure]))
else:
if isinstance(risk_original_dict[measure], numbers.Real):
np.testing.assert_allclose(
risk_original_dict[measure],
risk_refactor_dict[measure],
rtol=0.001,
err_msg=err_msg_format.format(
iter=i,
measure=measure,
truth=risk_original_dict[measure],
returned=risk_refactor_dict[measure]))
else:
np.testing.assert_equal(
risk_original_dict[measure],
risk_refactor_dict[measure],
err_msg=err_msg_format.format(
iter=i,
measure=measure,
truth=risk_original_dict[measure],
returned=risk_refactor_dict[measure]))
def load_market_data(bm_symbol='^GSPC'):
try:
fp_bm = get_datafile(get_benchmark_filename(bm_symbol), "rb")
except IOError:
print("""
data msgpacks aren't distributed with source.
Fetching data from Yahoo Finance.
""").strip()
dump_benchmarks(bm_symbol)
fp_bm = get_datafile(get_benchmark_filename(bm_symbol), "rb")
bm_list = msgpack.loads(fp_bm.read())
# Find the offset of the last date for which we have trading data in our
# list of valid trading days
last_bm_date = tuple_to_date(bm_list[-1][0])
last_bm_date_offset = trading_days.searchsorted(
last_bm_date.strftime('%Y/%m/%d'))
# If more than 1 trading days has elapsed since the last day where
# we have data,then we need to update
if len(trading_days) - last_bm_date_offset > 1:
update_benchmarks(bm_symbol, last_bm_date)
fp_bm = get_datafile(get_benchmark_filename(bm_symbol), "rb")
bm_list = msgpack.loads(fp_bm.read())
bm_returns = []
for packed_date, returns in bm_list:
event_dt = tuple_to_date(packed_date)
daily_return = DailyReturn(date=event_dt, returns=returns)
bm_returns.append(daily_return)
fp_bm.close()
bm_returns = sorted(bm_returns, key=attrgetter('date'))
try:
fp_tr = get_datafile('treasury_curves.msgpack', "rb")
except IOError:
print("""
data msgpacks aren't distributed with source.
Fetching data from data.treasury.gov
""").strip()
dump_treasury_curves()
fp_tr = get_datafile('treasury_curves.msgpack', "rb")
tr_list = msgpack.loads(fp_tr.read())
# Find the offset of the last date for which we have trading data in our
# list of valid trading days
last_tr_date = tuple_to_date(tr_list[-1][0])
last_tr_date_offset = trading_days.searchsorted(
last_tr_date.strftime('%Y/%m/%d'))
# If more than 1 trading days has elapsed since the last day where
# we have data,then we need to update
if len(trading_days) - last_tr_date_offset > 1:
update_treasury_curves(last_tr_date)
fp_tr = get_datafile('treasury_curves.msgpack', "rb")
tr_list = msgpack.loads(fp_tr.read())
tr_curves = {}
for packed_date, curve in tr_list:
tr_dt = tuple_to_date(packed_date)
# tr_dt = tr_dt.replace(hour=0, minute=0, second=0, microsecond=0,
# tzinfo=pytz.utc)
tr_curves[tr_dt] = curve
fp_tr.close()
tr_curves = OrderedDict(sorted(
((dt, c) for dt, c in tr_curves.iteritems()),
key=lambda t: t[0]))
return bm_returns, tr_curves
def test_risk_metrics_returns(self):
risk_metrics_refactor = risk.RiskMetricsIterative(self.start_date)
todays_date = self.start_date
cur_returns = []
for i, ret in enumerate(RETURNS):
todays_return_obj = DailyReturn(todays_date, ret)
cur_returns.append(todays_return_obj)
# Move forward day counter to next trading day
todays_date += self.oneday
while not trading.environment.is_trading_day(todays_date):
todays_date += self.oneday
try:
risk_metrics_original = risk.RiskMetricsBatch(
start_date=self.start_date,
end_date=todays_date,
returns=cur_returns)
except Exception as e:
#assert that when original raises exception, same
#exception is raised by risk_metrics_refactor
np.testing.assert_raises(type(e), risk_metrics_refactor.update,
todays_date, ret)
continue
risk_metrics_refactor.update(todays_date, ret)
self.assertEqual(risk_metrics_original.start_date,
risk_metrics_refactor.start_date)
self.assertEqual(risk_metrics_original.end_date,
risk_metrics_refactor.end_date)
self.assertEqual(risk_metrics_original.treasury_duration,
risk_metrics_refactor.treasury_duration)
self.assertEqual(risk_metrics_original.treasury_curve,
risk_metrics_refactor.treasury_curve)
self.assertEqual(risk_metrics_original.treasury_period_return,
risk_metrics_refactor.treasury_period_return)
self.assertEqual(risk_metrics_original.benchmark_returns,
risk_metrics_refactor.benchmark_returns)
self.assertEqual(risk_metrics_original.algorithm_returns,
risk_metrics_refactor.algorithm_returns)
risk_original_dict = risk_metrics_original.to_dict()
risk_refactor_dict = risk_metrics_refactor.to_dict()
self.assertEqual(set(risk_original_dict.keys()),
set(risk_refactor_dict.keys()))
err_msg_format = """\
"In update step {iter}: {measure} should be {truth} but is {returned}!"""
for measure in risk_original_dict.iterkeys():
if measure == 'max_drawdown':
np.testing.assert_almost_equal(
risk_refactor_dict[measure],
risk_original_dict[measure],
err_msg=err_msg_format.format(
iter=i,
measure=measure,
truth=risk_original_dict[measure],
returned=risk_refactor_dict[measure]))
else:
np.testing.assert_equal(
risk_original_dict[measure],
risk_refactor_dict[measure],
err_msg_format.format(
iter=i,
measure=measure,
truth=risk_original_dict[measure],
returned=risk_refactor_dict[measure]))