def __init_base_data(self) -> pd.DataFrame:
# FIXME: some stock_id do not have div
self.base_data = FinData(
dataset="TaiwanStockPrice",
select=self.stock_id,
date=self.start_date,
end_date=self.end_date,
user_id=self.user_id,
password=self.password,
)
StockDividend = FinData(
dataset="StockDividend",
select=self.stock_id,
date=self.start_date,
end_date=self.end_date,
user_id=self.user_id,
password=self.password,
)
if not StockDividend.empty:
cash_div = StockDividend[
[
"stock_id",
"CashExDividendTradingDate",
"CashEarningsDistribution",
]
].rename(columns={"CashExDividendTradingDate": "date"})
stock_div = StockDividend[
[
"stock_id",
"StockExDividendTradingDate",
"StockEarningsDistribution",
]
].rename(columns={"StockExDividendTradingDate": "date"})
self.base_data = pd.merge(
self.base_data,
cash_div,
left_on=["stock_id", "date"],
right_on=["stock_id", "date"],
how="left",
).fillna(0)
self.base_data = pd.merge(
self.base_data,
stock_div,
left_on=["stock_id", "date"],
right_on=["stock_id", "date"],
how="left",
).fillna(0)
else:
self.base_data["StockEarningsDistribution"] = 0
self.base_data["CashEarningsDistribution"] = 0
def test_FinData():
dataset = "TaiwanStockPrice"
stock_id = "2330"
date = "2020-03-10"
end_date = "2020-03-15"
data = FinData(
dataset=dataset, select=stock_id, date=date, end_date=end_date,
)
assert all(
data
== pd.DataFrame(
{
"date": [
"2020-03-10",
"2020-03-11",
"2020-03-12",
"2020-03-13",
],
"stock_id": ["2330", "2330", "2330", "2330"],
"Trading_Volume": [74869130, 64923710, 114173351, 151268148],
"Trading_money": [
22727941511,
19913151529,
33544278206,
42448997546,
],
"open": [301.5, 309.0, 299.0, 275.0],
"max": [309.0, 310.5, 299.0, 294.0],
"min": [301.0, 302.0, 287.0, 272.5],
"close": [307.0, 302.0, 294.0, 290.0],
"spread": [1.5, -5.0, -8.0, -4.0],
"Trading_turnover": [30268, 27176, 56989, 71990],
}
)
)
def __compute_compare_market(self):
self._compare_market_detail = self._trade_detail[
["date", "EverytimeTotalProfit"]
].copy()
self._compare_market_detail["CumDailyRetrun"] = (
np.log(self._compare_market_detail["EverytimeTotalProfit"])
- np.log(
self._compare_market_detail["EverytimeTotalProfit"].shift(1)
)
).fillna(0)
self._compare_market_detail["CumDailyRetrun"] = round(
self._compare_market_detail["CumDailyRetrun"].cumsum(), 5
)
TAIEX = FinData(
dataset="TaiwanStockPrice",
select="TAIEX",
date=self.start_date,
end_date=self.end_date,
user_id=self.user_id,
password=self.password,
)[["date", "close"]]
TAIEX["CumTaiexDailyRetrun"] = (
np.log(TAIEX["close"]) - np.log(TAIEX["close"].shift(1))
).fillna(0)
TAIEX["CumTaiexDailyRetrun"] = round(
TAIEX["CumTaiexDailyRetrun"].cumsum(), 5
)
self._compare_market_detail = pd.merge(
self._compare_market_detail,
TAIEX[["date", "CumTaiexDailyRetrun"]],
on=["date"],
how="left",
)
self._compare_market_detail = self._compare_market_detail.dropna()
self._compare_market_stats = pd.Series()
self._compare_market_stats["AnnualTaiexReturnPer"] = (
convert_Return2Annual(
self._compare_market_detail["CumTaiexDailyRetrun"].values[-1],
self._trade_period_years,
)
* 100
)
self._compare_market_stats["AnnualReturnPer"] = self._final_stats[
"AnnualReturnPer"
]
class BackTest:
def __init__(
self,
strategy: typing.Type[Strategy],
user_id: str = "",
password: str = "",
stock_id: str = "",
start_date: str = "",
end_date: str = "",
trader_fund: float = 0,
fee: float = 0.001425,
):
if not (isinstance(strategy, type) and issubclass(strategy, Strategy)):
raise TypeError("`strategy` must be a Strategy sub-type")
self.stock_id = stock_id
self.start_date = start_date
self.end_date = end_date
self.trader_fund = trader_fund
self.fee = fee
underlying_type = get_asset_underlying_type(stock_id)
self.tax = get_underlying_trading_tax(underlying_type)
self.trader = Trader(
stock_id=stock_id,
hold_volume=0,
hold_cost=0,
trader_fund=trader_fund,
fee=self.fee,
tax=self.tax,
)
self.user_id = user_id
self.password = password
self.strategy = strategy
self._results = pd.DataFrame()
self._final_stats = pd.Series()
def __init_base_data(self) -> pd.DataFrame:
# FIXME: some stock_id do not have div
self.base_data = FinData(
dataset="TaiwanStockPrice",
select=self.stock_id,
date=self.start_date,
end_date=self.end_date,
user_id=self.user_id,
password=self.password,
)
StockDividend = FinData(
dataset="StockDividend",
select=self.stock_id,
date=self.start_date,
end_date=self.end_date,
user_id=self.user_id,
password=self.password,
)
if not StockDividend.empty:
cash_div = StockDividend[
[
"stock_id",
"CashExDividendTradingDate",
"CashEarningsDistribution",
]
].rename(columns={"CashExDividendTradingDate": "date"})
stock_div = StockDividend[
[
"stock_id",
"StockExDividendTradingDate",
"StockEarningsDistribution",
]
].rename(columns={"StockExDividendTradingDate": "date"})
self.base_data = pd.merge(
self.base_data,
cash_div,
left_on=["stock_id", "date"],
right_on=["stock_id", "date"],
how="left",
).fillna(0)
self.base_data = pd.merge(
self.base_data,
stock_div,
left_on=["stock_id", "date"],
right_on=["stock_id", "date"],
how="left",
).fillna(0)
else:
self.base_data["StockEarningsDistribution"] = 0
self.base_data["CashEarningsDistribution"] = 0
def simulate(self):
self.__init_base_data()
trader = self.trader
strategy = self.strategy(trader)
self.base_data = strategy.init(base_data=self.base_data)
assert (
"signal" in self.base_data.columns
), "Must be create signal columns in base_data"
if not self.base_data.index.is_monotonic_increasing:
warnings.warn(
"Data index is not sorted in ascending order. Sorting.",
stacklevel=2,
)
self.base_data = self.base_data.sort_index()
for i in range(1, len(self.base_data)):
# use last date to decide buy or sell or nothing
last_date_index = i - 1
signal = self.base_data.loc[last_date_index, "signal"]
trade_price = self.base_data.loc[i, "open"]
strategy.trade(signal, trade_price)
cash_div = self.base_data.loc[i, "CashEarningsDistribution"]
stock_div = self.base_data.loc[i, "StockEarningsDistribution"]
self.__compute_div_income(strategy.trader, cash_div, stock_div)
dic_value = strategy.trader.__dict__
dic_value["date"] = self.base_data.loc[i, "date"]
self._results = self._results.append(dic_value, ignore_index=True)
self.__compute_final_stats()
def __compute_div_income(self, trader, cash_div: float, stock_div: float):
gain_stock_div = stock_div * trader.hold_volume / 10
gain_cash = cash_div * trader.hold_volume
origin_cost = trader.hold_cost * trader.hold_volume
trader.hold_volume += gain_stock_div
new_cost = origin_cost - gain_cash
trader.hold_cost = (
new_cost / trader.hold_volume if trader.hold_volume != 0 else 0
)
trader.UnrealizedProfit = round(
(
trader.trade_price * (1 - trader.tax - trader.fee)
- trader.hold_cost
)
* trader.hold_volume,
2,
)
trader.RealizedProfit += gain_cash
trader.EverytimeProfit = trader.RealizedProfit + trader.UnrealizedProfit
def __compute_final_stats(self):
self._final_stats["MeanProfit"] = np.mean(
self._results["EverytimeProfit"]
)
self._final_stats["MaxLoss"] = np.min(self._results["EverytimeProfit"])
self._final_stats["FinalProfit"] = self._results[
"EverytimeProfit"
].values[-1]
self._final_stats["MeanProfitPer[%]"] = round(
self._final_stats["MeanProfit"] / self.trader_fund * 100, 2
)
self._final_stats["FinalProfitPer[%]"] = round(
self._final_stats["FinalProfit"] / self.trader_fund * 100, 2
)
self._final_stats["MaxLossPer[%]"] = round(
self._final_stats["MaxLoss"] / self.trader_fund * 100, 2
)
def get_final_stats(self) -> pd.Series():
return self._final_stats
def get_results(self) -> pd.DataFrame():
return self._results
def plot(
self,
title: str = "Backtest Result",
xlabel: str = "Time",
ylabel: str = "Profit",
grid: bool = True,
):
try:
import matplotlib.pyplot as plt
import matplotlib.gridspec as gridspec
except ImportError:
raise ImportError("You must install matplotlib to plot importance")
fig = plt.figure(figsize=(12, 8))
gs = gridspec.GridSpec(4, 1, figure=fig)
ax = fig.add_subplot(gs[:2, :])
xpos = self._results.index
ax.plot("UnrealizedProfit", data=self._results, marker="", alpha=0.8)
ax.plot("RealizedProfit", data=self._results, marker="", alpha=0.8)
ax.plot("EverytimeProfit", data=self._results, marker="", alpha=0.8)
ax.grid(grid)
ax.legend(loc=2)
axx = ax.twinx()
axx.bar(
xpos,
self._results["hold_volume"],
alpha=0.2,
label="hold_volume",
color="pink",
)
axx.legend(loc=3)
ax2 = fig.add_subplot(gs[2:, :], sharex=ax)
ax2.plot(
"trade_price",
data=self._results,
marker="",
label="open",
alpha=0.8,
)
ax2.plot(
"hold_cost",
data=self._results,
marker="",
label="hold_cost",
alpha=0.8,
)
# TODO: add signal plot
ax2.legend(loc=2)
ax2.grid(grid)
if title is not None:
ax.set_title(title)
if xlabel is not None:
ax.set_xlabel(xlabel)
if ylabel is not None:
ax.set_ylabel(ylabel)
plt.show()
class BackTest:
def __init__(
self,
user_id: str = "",
password: str = "",
stock_id: str = "",
start_date: str = "",
end_date: str = "",
trader_fund: float = 0,
fee: float = 0.001425,
strategy: Strategy = None,
):
self.stock_id = stock_id
self.start_date = start_date
self.end_date = end_date
self.trader_fund = trader_fund
self.fee = fee
underlying_type = get_asset_underlying_type(stock_id)
self.tax = get_underlying_trading_tax(underlying_type)
self.trader = Trader(
stock_id=stock_id,
hold_volume=0,
hold_cost=0,
trader_fund=trader_fund,
fee=self.fee,
tax=self.tax,
)
self.user_id = user_id
self.password = password
self.strategy = strategy
self._trade_detail = pd.DataFrame()
self._final_stats = pd.Series()
self.__init_base_data()
def add_strategy(self, strategy: Strategy):
self.strategy = strategy
def __init_base_data(self) -> pd.DataFrame:
# FIXME: some stock_id do not have div
self.stock_price = FinData(
dataset="TaiwanStockPrice",
select=self.stock_id,
date=self.start_date,
end_date=self.end_date,
user_id=self.user_id,
password=self.password,
)
StockDividend = FinData(
dataset="StockDividend",
select=self.stock_id,
date=self.start_date,
end_date=self.end_date,
user_id=self.user_id,
password=self.password,
)
if not StockDividend.empty:
cash_div = StockDividend[
[
"stock_id",
"CashExDividendTradingDate",
"CashEarningsDistribution",
]
].rename(columns={"CashExDividendTradingDate": "date"})
stock_div = StockDividend[
[
"stock_id",
"StockExDividendTradingDate",
"StockEarningsDistribution",
]
].rename(columns={"StockExDividendTradingDate": "date"})
self.stock_price = pd.merge(
self.stock_price,
cash_div,
left_on=["stock_id", "date"],
right_on=["stock_id", "date"],
how="left",
).fillna(0)
self.stock_price = pd.merge(
self.stock_price,
stock_div,
left_on=["stock_id", "date"],
right_on=["stock_id", "date"],
how="left",
).fillna(0)
else:
self.stock_price["StockEarningsDistribution"] = 0
self.stock_price["CashEarningsDistribution"] = 0
def simulate(self):
trader = self.trader
strategy = self.strategy(
trader, self.stock_id, self.start_date, self.end_date
)
self.stock_price = strategy.create_trade_sign(
stock_price=self.stock_price
)
assert (
"signal" in self.stock_price.columns
), "Must be create signal columns in stock_price"
if not self.stock_price.index.is_monotonic_increasing:
warnings.warn(
"Data index is not sorted in ascending order. Sorting.",
stacklevel=2,
)
self.stock_price = self.stock_price.sort_index()
for i in range(1, len(self.stock_price)):
# use last date to decide buy or sell or nothing
last_date_index = i - 1
signal = self.stock_price.loc[last_date_index, "signal"]
trade_price = self.stock_price.loc[i, "open"]
strategy.trade(signal, trade_price)
cash_div = self.stock_price.loc[i, "CashEarningsDistribution"]
stock_div = self.stock_price.loc[i, "StockEarningsDistribution"]
self.__compute_div_income(strategy.trader, cash_div, stock_div)
dic_value = strategy.trader.__dict__
dic_value["date"] = self.stock_price.loc[i, "date"]
dic_value["signal"] = self.stock_price.loc[i, "signal"]
self._trade_detail = self._trade_detail.append(
dic_value, ignore_index=True
)
self.__compute_final_stats()
def __compute_div_income(self, trader, cash_div: float, stock_div: float):
gain_stock_div = stock_div * trader.hold_volume / 10
gain_cash = cash_div * trader.hold_volume
origin_cost = trader.hold_cost * trader.hold_volume
trader.hold_volume += gain_stock_div
new_cost = origin_cost - gain_cash
trader.hold_cost = (
new_cost / trader.hold_volume if trader.hold_volume != 0 else 0
)
trader.UnrealizedProfit = round(
(
trader.trade_price * (1 - trader.tax - trader.fee)
- trader.hold_cost
)
* trader.hold_volume,
2,
)
trader.RealizedProfit += gain_cash
trader.EverytimeProfit = trader.RealizedProfit + trader.UnrealizedProfit
def __compute_final_stats(self):
self._final_stats["MeanProfit"] = np.mean(
self._trade_detail["EverytimeProfit"]
)
self._final_stats["MaxLoss"] = np.min(
self._trade_detail["EverytimeProfit"]
)
self._final_stats["FinalProfit"] = self._trade_detail[
"EverytimeProfit"
].values[-1]
self._final_stats["MeanProfitPer"] = round(
self._final_stats["MeanProfit"] / self.trader_fund * 100, 2
)
self._final_stats["FinalProfitPer"] = round(
self._final_stats["FinalProfit"] / self.trader_fund * 100, 2
)
self._final_stats["MaxLossPer"] = round(
self._final_stats["MaxLoss"] / self.trader_fund * 100, 2
)
# +1, calculate_Datenbr not include last day
trade_days = (
calculate_Datenbr(
self._trade_detail["date"].min(),
self._trade_detail["date"].max(),
)
+ 1
)
trade_years = (trade_days + 1) / 365
# +1, self._trade_detail wihtout contain first day
self._final_stats["AnnualReturnPer"] = round(
(
(self._final_stats["FinalProfitPer"] / 100 + 1)
** (1 / trade_years)
- 1
)
* 100,
2,
)
timestep_returns = (
self._trade_detail["EverytimeProfit"]
- self._trade_detail["EverytimeProfit"].shift(1)
) / (self._trade_detail["EverytimeProfit"].shift(1) + self.trader_fund)
stratagy_return = np.mean(timestep_returns)
stratagy_std = np.std(timestep_returns)
self._final_stats["AnnualSharpRatio"] = calculate_sharp_ratio(
stratagy_return, stratagy_std
)
@property
def final_stats(self) -> pd.Series():
self._final_stats = pd.Series(
output.final_stats(**self._final_stats.to_dict()).dict()
)
return self._final_stats
@property
def trade_detail(self) -> pd.DataFrame():
self._trade_detail = pd.DataFrame(
[
output.trade_detail(**row_dict).dict()
for row_dict in self._trade_detail.to_dict("records")
]
)
return self._trade_detail
def plot(
self,
title: str = "Backtest Result",
xlabel: str = "Time",
ylabel: str = "Profit",
grid: bool = True,
):
try:
import matplotlib.pyplot as plt
import matplotlib.gridspec as gridspec
except ImportError:
raise ImportError("You must install matplotlib to plot importance")
fig = plt.figure(figsize=(12, 8))
gs = gridspec.GridSpec(4, 1, figure=fig)
ax = fig.add_subplot(gs[:2, :])
xpos = self._trade_detail.index
ax.plot(
"UnrealizedProfit", data=self._trade_detail, marker="", alpha=0.8
)
ax.plot("RealizedProfit", data=self._trade_detail, marker="", alpha=0.8)
ax.plot(
"EverytimeProfit", data=self._trade_detail, marker="", alpha=0.8
)
ax.grid(grid)
ax.legend(loc=2)
axx = ax.twinx()
axx.bar(
xpos,
self._trade_detail["hold_volume"],
alpha=0.2,
label="hold_volume",
color="pink",
)
axx.legend(loc=3)
ax2 = fig.add_subplot(gs[2:, :], sharex=ax)
ax2.plot(
"trade_price",
data=self._trade_detail,
marker="",
label="open",
alpha=0.8,
)
ax2.plot(
"hold_cost",
data=self._trade_detail,
marker="",
label="hold_cost",
alpha=0.8,
)
# TODO: add signal plot
ax2.legend(loc=2)
ax2.grid(grid)
if title is not None:
ax.set_title(title)
if xlabel is not None:
ax.set_xlabel(xlabel)
if ylabel is not None:
ax.set_ylabel(ylabel)
plt.show()
def get_asset_underlying_type(stock_id: str) -> str:
TaiwanStockInfo = FinData("TaiwanStockInfo", date="")
underlying_type = TaiwanStockInfo[TaiwanStockInfo["stock_id"] == stock_id][
"industry_category"
].values[0]
return underlying_type
