PyFacts/tests/test_fincal.py

529 lines
20 KiB
Python
Raw Permalink Normal View History

2022-03-11 04:10:37 +00:00
import datetime
import math
import random
2022-04-03 09:57:07 +00:00
from typing import List
2022-03-11 04:10:37 +00:00
import pytest
from dateutil.relativedelta import relativedelta
2022-04-09 06:28:35 +00:00
from fincal import (
AllFrequencies,
FincalOptions,
Frequency,
TimeSeries,
create_date_series,
)
2022-03-11 04:10:37 +00:00
from fincal.exceptions import DateNotFoundError
def create_prices(s0: float, mu: float, sigma: float, num_prices: int) -> list:
"""Generates a price following a geometric brownian motion process based on the input of the arguments.
Since this function is used only to generate data for tests, the seed is fixed as 1234.
Many of the tests rely on exact values generated using this seed.
If the seed is changed, those tests will fail.
Parameters:
------------
s0: float
Asset inital price.
mu: float
Interest rate expressed annual terms.
sigma: float
Volatility expressed annual terms.
num_prices: int
number of prices to generate
Returns:
--------
Returns a list of values generated using GBM algorithm
2022-03-11 04:10:37 +00:00
"""
random.seed(1234) # WARNING! Changing the seed will cause most tests to fail
all_values = []
for _ in range(num_prices):
s0 *= math.exp(
(mu - 0.5 * sigma**2) * (1.0 / 365.0) + sigma * math.sqrt(1.0 / 365.0) * random.gauss(mu=0, sigma=1)
)
all_values.append(round(s0, 2))
return all_values
2022-04-03 09:57:07 +00:00
def create_test_data(
frequency: Frequency,
num: int = 1000,
skip_weekends: bool = False,
mu: float = 0.1,
sigma: float = 0.05,
eomonth: bool = False,
) -> List[tuple]:
"""Creates TimeSeries data
Parameters:
-----------
frequency: Frequency
The frequency of the time series data to be generated.
num: int
Number of date: value pairs to be generated.
2022-03-11 04:10:37 +00:00
skip_weekends: bool
Whether weekends (saturday, sunday) should be skipped.
Gets used only if the frequency is daily.
mu: float
Mean return for the values.
sigma: float
standard deviation of the values.
Returns:
--------
Returns a TimeSeries object
"""
start_date = datetime.datetime(2017, 1, 1)
timedelta_dict = {
2022-03-15 19:05:36 +00:00
frequency.freq_type: int(
frequency.value * num * (7 / 5 if frequency == AllFrequencies.D and skip_weekends else 1)
)
}
end_date = start_date + relativedelta(**timedelta_dict)
2022-04-03 09:57:07 +00:00
dates = create_date_series(start_date, end_date, frequency.symbol, skip_weekends=skip_weekends, eomonth=eomonth)
values = create_prices(1000, mu, sigma, num)
2022-04-03 09:57:07 +00:00
ts = list(zip(dates, values))
2022-03-11 04:10:37 +00:00
return ts
2022-04-03 09:57:07 +00:00
class TestDateSeries:
def test_daily(self):
start_date = datetime.datetime(2020, 1, 1)
end_date = datetime.datetime(2020, 12, 31)
d = create_date_series(start_date, end_date, frequency="D")
assert len(d) == 366
start_date = datetime.datetime(2017, 1, 1)
end_date = datetime.datetime(2017, 12, 31)
d = create_date_series(start_date, end_date, frequency="D")
assert len(d) == 365
with pytest.raises(ValueError):
create_date_series(start_date, end_date, frequency="D", eomonth=True)
def test_monthly(self):
start_date = datetime.datetime(2020, 1, 1)
end_date = datetime.datetime(2020, 12, 31)
d = create_date_series(start_date, end_date, frequency="M")
assert len(d) == 12
d = create_date_series(start_date, end_date, frequency="M", eomonth=True)
assert datetime.datetime(2020, 2, 29) in d
start_date = datetime.datetime(2020, 1, 31)
d = create_date_series(start_date, end_date, frequency="M")
assert datetime.datetime(2020, 2, 29) in d
assert datetime.datetime(2020, 8, 31) in d
assert datetime.datetime(2020, 10, 30) not in d
start_date = datetime.datetime(2020, 2, 29)
d = create_date_series(start_date, end_date, frequency="M")
assert len(d) == 11
assert datetime.datetime(2020, 2, 29) in d
assert datetime.datetime(2020, 8, 31) not in d
assert datetime.datetime(2020, 10, 29) in d
def test_quarterly(self):
start_date = datetime.datetime(2018, 1, 1)
end_date = datetime.datetime(2020, 12, 31)
d = create_date_series(start_date, end_date, frequency="Q")
assert len(d) == 12
d = create_date_series(start_date, end_date, frequency="Q", eomonth=True)
assert datetime.datetime(2020, 4, 30) in d
start_date = datetime.datetime(2020, 1, 31)
d = create_date_series(start_date, end_date, frequency="Q")
assert len(d) == 4
assert datetime.datetime(2020, 2, 29) not in d
assert max(d) == datetime.datetime(2020, 10, 31)
start_date = datetime.datetime(2020, 2, 29)
d = create_date_series(start_date, end_date, frequency="Q")
assert datetime.datetime(2020, 2, 29) in d
assert datetime.datetime(2020, 8, 31) not in d
assert datetime.datetime(2020, 11, 29) in d
d = create_date_series(start_date, end_date, frequency="Q", eomonth=True)
assert datetime.datetime(2020, 11, 30) in d
class TestTimeSeriesCreation:
def test_creation_with_list_of_tuples(self):
ts_data = create_test_data(frequency=AllFrequencies.D, num=50)
ts = TimeSeries(ts_data, frequency="D")
assert len(ts) == 50
assert isinstance(ts.frequency, Frequency)
assert ts.frequency.days == 1
def test_creation_with_string_dates(self):
ts_data = create_test_data(frequency=AllFrequencies.D, num=50)
ts_data1 = [(dt.strftime("%Y-%m-%d"), val) for dt, val in ts_data]
ts = TimeSeries(ts_data1, frequency="D")
datetime.datetime(2017, 1, 1) in ts
ts_data1 = [(dt.strftime("%d-%m-%Y"), val) for dt, val in ts_data]
ts = TimeSeries(ts_data1, frequency="D", date_format="%d-%m-%Y")
datetime.datetime(2017, 1, 1) in ts
ts_data1 = [(dt.strftime("%m-%d-%Y"), val) for dt, val in ts_data]
ts = TimeSeries(ts_data1, frequency="D", date_format="%m-%d-%Y")
datetime.datetime(2017, 1, 1) in ts
ts_data1 = [(dt.strftime("%m-%d-%Y %H:%M"), val) for dt, val in ts_data]
ts = TimeSeries(ts_data1, frequency="D", date_format="%m-%d-%Y %H:%M")
datetime.datetime(2017, 1, 1, 0, 0) in ts
def test_creation_with_list_of_dicts(self):
ts_data = create_test_data(frequency=AllFrequencies.D, num=50)
ts_data1 = [{"date": dt.strftime("%Y-%m-%d"), "value": val} for dt, val in ts_data]
ts = TimeSeries(ts_data1, frequency="D")
datetime.datetime(2017, 1, 1) in ts
def test_creation_with_list_of_lists(self):
ts_data = create_test_data(frequency=AllFrequencies.D, num=50)
ts_data1 = [[dt.strftime("%Y-%m-%d"), val] for dt, val in ts_data]
ts = TimeSeries(ts_data1, frequency="D")
datetime.datetime(2017, 1, 1) in ts
def test_creation_with_dict(self):
ts_data = create_test_data(frequency=AllFrequencies.D, num=50)
ts_data1 = [{dt.strftime("%Y-%m-%d"): val} for dt, val in ts_data]
ts = TimeSeries(ts_data1, frequency="D")
datetime.datetime(2017, 1, 1) in ts
class TestTimeSeriesBasics:
def test_fill(self):
ts_data = create_test_data(frequency=AllFrequencies.D, num=50, skip_weekends=True)
ts = TimeSeries(ts_data, frequency="D")
ffill_data = ts.ffill()
assert len(ffill_data) == 68
ffill_data = ts.ffill(inplace=True)
assert ffill_data is None
assert len(ts) == 68
ts_data = create_test_data(frequency=AllFrequencies.D, num=50, skip_weekends=True)
ts = TimeSeries(ts_data, frequency="D")
bfill_data = ts.bfill()
assert len(bfill_data) == 68
bfill_data = ts.bfill(inplace=True)
assert bfill_data is None
assert len(ts) == 68
2022-04-04 18:18:06 +00:00
data = [("2021-01-01", 220), ("2021-01-02", 230), ("2021-01-04", 240)]
2022-04-03 09:57:07 +00:00
ts = TimeSeries(data, frequency="D")
ff = ts.ffill()
assert ff["2021-01-03"][1] == 230
bf = ts.bfill()
assert bf["2021-01-03"][1] == 240
2022-04-05 05:13:31 +00:00
def test_fill_weekly(self):
ts_data = create_test_data(frequency=AllFrequencies.W, num=10)
ts_data.pop(2)
ts_data.pop(6)
ts = TimeSeries(ts_data, frequency="W")
assert len(ts) == 8
ff = ts.ffill()
assert len(ff) == 10
assert "2017-01-15" in ff
assert ff["2017-01-15"][1] == ff["2017-01-08"][1]
bf = ts.bfill()
assert len(ff) == 10
assert "2017-01-15" in bf
assert bf["2017-01-15"][1] == bf["2017-01-22"][1]
2022-04-03 09:57:07 +00:00
2022-03-11 04:10:37 +00:00
class TestReturns:
def test_returns_calc(self):
2022-04-03 09:57:07 +00:00
ts_data = create_test_data(AllFrequencies.D, skip_weekends=True)
ts = TimeSeries(ts_data, "D")
2022-03-11 04:10:37 +00:00
returns = ts.calculate_returns(
"2020-01-01", annual_compounded_returns=False, return_period_unit="years", return_period_value=1
2022-03-11 04:10:37 +00:00
)
assert round(returns[1], 6) == 0.112913
returns = ts.calculate_returns(
"2020-04-01", annual_compounded_returns=False, return_period_unit="months", return_period_value=3
2022-03-11 04:10:37 +00:00
)
assert round(returns[1], 6) == 0.015908
returns = ts.calculate_returns(
"2020-04-01", annual_compounded_returns=True, return_period_unit="months", return_period_value=3
2022-03-11 04:10:37 +00:00
)
assert round(returns[1], 6) == 0.065167
returns = ts.calculate_returns(
"2020-04-01", annual_compounded_returns=False, return_period_unit="days", return_period_value=90
2022-03-11 04:10:37 +00:00
)
assert round(returns[1], 6) == 0.017673
returns = ts.calculate_returns(
"2020-04-01", annual_compounded_returns=True, return_period_unit="days", return_period_value=90
2022-03-11 04:10:37 +00:00
)
assert round(returns[1], 6) == 0.073632
with pytest.raises(DateNotFoundError):
ts.calculate_returns("2020-04-04", return_period_unit="days", return_period_value=90, as_on_match="exact")
2022-03-11 04:10:37 +00:00
with pytest.raises(DateNotFoundError):
ts.calculate_returns("2020-04-04", return_period_unit="months", return_period_value=3, prior_match="exact")
2022-03-11 04:10:37 +00:00
def test_date_formats(self):
2022-04-03 09:57:07 +00:00
ts_data = create_test_data(AllFrequencies.D, skip_weekends=True)
ts = TimeSeries(ts_data, "D")
2022-03-11 04:10:37 +00:00
FincalOptions.date_format = "%d-%m-%Y"
with pytest.raises(ValueError):
ts.calculate_returns(
"2020-04-10", annual_compounded_returns=True, return_period_unit="days", return_period_value=90
)
2022-03-11 04:10:37 +00:00
returns1 = ts.calculate_returns(
"2020-04-01", return_period_unit="days", return_period_value=90, date_format="%Y-%m-%d"
)
returns2 = ts.calculate_returns("01-04-2020", return_period_unit="days", return_period_value=90)
2022-03-11 04:10:37 +00:00
assert round(returns1[1], 6) == round(returns2[1], 6) == 0.073632
FincalOptions.date_format = "%m-%d-%Y"
with pytest.raises(ValueError):
ts.calculate_returns(
"2020-04-01", annual_compounded_returns=True, return_period_unit="days", return_period_value=90
)
2022-03-11 04:10:37 +00:00
returns1 = ts.calculate_returns(
"2020-04-01", return_period_unit="days", return_period_value=90, date_format="%Y-%m-%d"
)
returns2 = ts.calculate_returns("04-01-2020", return_period_unit="days", return_period_value=90)
2022-03-11 04:10:37 +00:00
assert round(returns1[1], 6) == round(returns2[1], 6) == 0.073632
def test_limits(self):
FincalOptions.date_format = "%Y-%m-%d"
2022-04-03 09:57:07 +00:00
ts_data = create_test_data(AllFrequencies.D)
ts = TimeSeries(ts_data, "D")
2022-03-11 04:10:37 +00:00
with pytest.raises(DateNotFoundError):
ts.calculate_returns("2020-11-25", return_period_unit="days", return_period_value=90, closest_max_days=10)
2022-04-03 19:18:55 +00:00
def test_rolling_returns(self):
# Yet to be written
return True
class TestExpand:
def test_weekly_to_daily(self):
ts_data = create_test_data(AllFrequencies.W, 10)
ts = TimeSeries(ts_data, "W")
expanded_ts = ts.expand("D", "ffill")
assert len(expanded_ts) == 64
assert expanded_ts.frequency.name == "daily"
assert expanded_ts.iloc[0][1] == expanded_ts.iloc[1][1]
def test_weekly_to_daily_no_weekends(self):
ts_data = create_test_data(AllFrequencies.W, 10)
ts = TimeSeries(ts_data, "W")
expanded_ts = ts.expand("D", "ffill", skip_weekends=True)
2022-04-04 18:18:06 +00:00
assert len(expanded_ts) == 46
2022-04-03 19:18:55 +00:00
assert expanded_ts.frequency.name == "daily"
assert expanded_ts.iloc[0][1] == expanded_ts.iloc[1][1]
def test_monthly_to_daily(self):
ts_data = create_test_data(AllFrequencies.M, 6)
ts = TimeSeries(ts_data, "M")
expanded_ts = ts.expand("D", "ffill")
assert len(expanded_ts) == 152
assert expanded_ts.frequency.name == "daily"
assert expanded_ts.iloc[0][1] == expanded_ts.iloc[1][1]
def test_monthly_to_daily_no_weekends(self):
ts_data = create_test_data(AllFrequencies.M, 6)
ts = TimeSeries(ts_data, "M")
expanded_ts = ts.expand("D", "ffill", skip_weekends=True)
assert len(expanded_ts) == 109
assert expanded_ts.frequency.name == "daily"
assert expanded_ts.iloc[0][1] == expanded_ts.iloc[1][1]
def test_monthly_to_weekly(self):
ts_data = create_test_data(AllFrequencies.M, 6)
ts = TimeSeries(ts_data, "M")
expanded_ts = ts.expand("W", "ffill")
assert len(expanded_ts) == 22
assert expanded_ts.frequency.name == "weekly"
assert expanded_ts.iloc[0][1] == expanded_ts.iloc[1][1]
def test_yearly_to_monthly(self):
ts_data = create_test_data(AllFrequencies.Y, 5)
ts = TimeSeries(ts_data, "Y")
expanded_ts = ts.expand("M", "ffill")
assert len(expanded_ts) == 49
assert expanded_ts.frequency.name == "monthly"
assert expanded_ts.iloc[0][1] == expanded_ts.iloc[1][1]
2022-04-03 09:57:07 +00:00
class TestReturnsAgain:
data = [
("2020-01-01", 10),
("2020-02-01", 12),
("2020-03-01", 14),
("2020-04-01", 16),
("2020-05-01", 18),
("2020-06-01", 20),
("2020-07-01", 22),
("2020-08-01", 24),
("2020-09-01", 26),
("2020-10-01", 28),
("2020-11-01", 30),
("2020-12-01", 32),
("2021-01-01", 34),
]
def test_returns_calc(self):
ts = TimeSeries(self.data, frequency="M")
returns = ts.calculate_returns(
"2021-01-01", annual_compounded_returns=False, return_period_unit="years", return_period_value=1
)
assert returns[1] == 2.4
returns = ts.calculate_returns(
"2020-04-01", annual_compounded_returns=False, return_period_unit="months", return_period_value=3
)
assert round(returns[1], 4) == 0.6
returns = ts.calculate_returns(
"2020-04-01", annual_compounded_returns=True, return_period_unit="months", return_period_value=3
)
assert round(returns[1], 4) == 5.5536
returns = ts.calculate_returns(
"2020-04-01", annual_compounded_returns=False, return_period_unit="days", return_period_value=90
)
assert round(returns[1], 4) == 0.6
returns = ts.calculate_returns(
"2020-04-01", annual_compounded_returns=True, return_period_unit="days", return_period_value=90
)
assert round(returns[1], 4) == 5.727
returns = ts.calculate_returns(
"2020-04-10", annual_compounded_returns=True, return_period_unit="days", return_period_value=90
)
assert round(returns[1], 4) == 5.727
with pytest.raises(DateNotFoundError):
ts.calculate_returns("2020-04-10", return_period_unit="days", return_period_value=90, as_on_match="exact")
with pytest.raises(DateNotFoundError):
ts.calculate_returns("2020-04-10", return_period_unit="days", return_period_value=90, prior_match="exact")
def test_date_formats(self):
ts = TimeSeries(self.data, frequency="M")
FincalOptions.date_format = "%d-%m-%Y"
with pytest.raises(ValueError):
ts.calculate_returns(
"2020-04-10", annual_compounded_returns=True, return_period_unit="days", return_period_value=90
)
returns1 = ts.calculate_returns(
"2020-04-10", return_period_unit="days", return_period_value=90, date_format="%Y-%m-%d"
)
returns2 = ts.calculate_returns("10-04-2020", return_period_unit="days", return_period_value=90)
assert round(returns1[1], 4) == round(returns2[1], 4) == 5.727
FincalOptions.date_format = "%m-%d-%Y"
with pytest.raises(ValueError):
ts.calculate_returns(
"2020-04-10", annual_compounded_returns=True, return_period_unit="days", return_period_value=90
)
returns1 = ts.calculate_returns(
"2020-04-10", return_period_unit="days", return_period_value=90, date_format="%Y-%m-%d"
)
returns2 = ts.calculate_returns("04-10-2020", return_period_unit="days", return_period_value=90)
assert round(returns1[1], 4) == round(returns2[1], 4) == 5.727
def test_limits(self):
ts = TimeSeries(self.data, frequency="M")
FincalOptions.date_format = "%Y-%m-%d"
with pytest.raises(DateNotFoundError):
ts.calculate_returns("2020-04-25", return_period_unit="days", return_period_value=90, closest_max_days=10)
class TestVolatility:
def test_daily_ts(self):
2022-04-03 09:57:07 +00:00
ts_data = create_test_data(AllFrequencies.D)
ts = TimeSeries(ts_data, "D")
assert len(ts) == 1000
sd = ts.volatility(annualize_volatility=False)
assert round(sd, 6) == 0.002622
sd = ts.volatility()
assert round(sd, 6) == 0.050098
sd = ts.volatility(annual_compounded_returns=True)
assert round(sd, 4) == 37.9329
sd = ts.volatility(return_period_unit="months", annual_compounded_returns=True)
assert round(sd, 4) == 0.6778
sd = ts.volatility(return_period_unit="years")
assert round(sd, 6) == 0.023164
sd = ts.volatility(from_date="2017-10-01", to_date="2019-08-31", annualize_volatility=True)
assert round(sd, 6) == 0.050559
2022-03-13 17:22:41 +00:00
sd = ts.volatility(from_date="2017-02-01", frequency="M", return_period_unit="months")
assert round(sd, 6) == 0.050884
sd = ts.volatility(
frequency="M",
return_period_unit="months",
return_period_value=3,
annualize_volatility=False,
)
assert round(sd, 6) == 0.020547
2022-03-15 19:05:36 +00:00
class TestDrawdown:
def test_daily_ts(self):
2022-04-03 09:57:07 +00:00
ts_data = create_test_data(AllFrequencies.D, skip_weekends=True)
ts = TimeSeries(ts_data, "D")
2022-03-15 19:05:36 +00:00
mdd = ts.max_drawdown()
assert isinstance(mdd, dict)
assert len(mdd) == 3
assert all(i in mdd for i in ["start_date", "end_date", "drawdown"])
expeced_response = {
"start_date": datetime.datetime(2017, 6, 6, 0, 0),
"end_date": datetime.datetime(2017, 7, 31, 0, 0),
"drawdown": -0.028293686030751997,
}
assert mdd == expeced_response
def test_weekly_ts(self):
2022-04-03 09:57:07 +00:00
ts_data = create_test_data(AllFrequencies.W, mu=1, sigma=0.5)
ts = TimeSeries(ts_data, "W")
2022-03-15 19:05:36 +00:00
mdd = ts.max_drawdown()
assert isinstance(mdd, dict)
assert len(mdd) == 3
assert all(i in mdd for i in ["start_date", "end_date", "drawdown"])
expeced_response = {
"start_date": datetime.datetime(2019, 2, 17, 0, 0),
"end_date": datetime.datetime(2019, 11, 17, 0, 0),
"drawdown": -0.2584760499552089,
}
assert mdd == expeced_response
2022-04-09 06:28:35 +00:00
class TestSync:
def test_weekly_to_daily(self):
daily_data = create_test_data(AllFrequencies.D, num=15)
weekly_data = create_test_data(AllFrequencies.W, num=3)
daily_ts = TimeSeries(daily_data, frequency="D")
weekly_ts = TimeSeries(weekly_data, frequency="W")
synced_weekly_ts = daily_ts.sync(weekly_ts)
assert len(daily_ts) == len(synced_weekly_ts)
assert synced_weekly_ts.frequency == AllFrequencies.D
assert "2017-01-02" in synced_weekly_ts
assert synced_weekly_ts["2017-01-02"][1] == synced_weekly_ts["2017-01-01"][1]