#cell 0
import warnings
warnings.filterwarnings("ignore", message="numpy.dtype size changed") # another bogus warning, see https://github.com/numpy/numpy/pull/432
import numpy as np
import pandas as pd
import sys
from collections import defaultdict
from pprint import pformat
from collections import deque
import math
from functools import reduce
import pandas as pd
from copy import copy
import numpy as np
from pyqstrat.pq_utils import *
from pyqstrat.marketdata import *
from pyqstrat.orders import *
from pyqstrat.plot import *
from pyqstrat.evaluator import *
#cell 1
def _calc_pnl(open_trades, new_trades, ending_close, multiplier):
'''
>>> from collections import deque
>>> trades = deque([Trade('IBM', np.datetime64('2018-01-01 10:15:00'), 3, 51.),
... Trade('IBM', np.datetime64('2018-01-01 10:20:00'), 10, 50.),
... Trade('IBM', np.datetime64('2018-01-02 11:20:00'), -5, 45.)])
>>> print(_calc_pnl(open_trades = deque(), new_trades = trades, ending_close = 54, multiplier = 100))
(deque([IBM 2018-01-01 10:20 qty: 8 prc: 50.0 order: None]), 3200.0, -2800.0)
>>> trades = deque([Trade('IBM', np.datetime64('2018-01-01 10:15:00'), -8, 10.),
... Trade('IBM', np.datetime64('2018-01-01 10:20:00'), 9, 11.),
... Trade('IBM', np.datetime64('2018-01-02 11:20:00'), -4, 6.)])
>>> print(_calc_pnl(open_trades = deque(), new_trades = trades, ending_close = 5.8, multiplier = 100))
(deque([IBM 2018-01-02 11:20 qty: -3 prc: 6.0 order: None]), 60.00000000000006, -1300.0)
'''
realized = 0.
unrealized = 0.
trades = copy(new_trades)
while (len(trades)):
trade = trades[0]
if not len(open_trades) or (np.sign(open_trades[-1].qty) == np.sign(trade.qty)):
open_trades.append(copy(trade))
trades.popleft()
continue
if abs(trade.qty) > abs(open_trades[0].qty):
open_trade = open_trades.popleft()
realized += open_trade.qty * multiplier * (trade.price - open_trade.price)
trade.qty += open_trade.qty
else:
open_trade = open_trades[0]
realized += trade.qty * multiplier * (open_trades[-1].price - trade.price)
trades.popleft()
open_trade.qty += trade.qty
unrealized = sum([open_trade.qty * (ending_close - open_trade.price) for open_trade in open_trades]) * multiplier
return open_trades, unrealized, realized
[docs]class Trade:
[docs] def __init__(self, symbol, date, qty, price, fee = 0., commission = 0., order = None):
'''Args:
symbol: a string
date: Trade execution datetime
qty: Number of contracts or shares filled
price: Trade price
fee: Fees paid to brokers or others. Default 0
commision: Commission paid to brokers or others. Default 0
order: A reference to the order that created this trade. Default None
'''
assert(isinstance(symbol, str) and len(symbol) > 0)
assert(np.isfinite(qty))
assert(np.isfinite(price))
assert(np.isfinite(fee))
assert(np.isfinite(commission))
self.symbol = symbol
self.date = date
self.qty = qty
self.price = price
self.fee = fee
self.commission = commission
self.order = order
def __repr__(self):
return '{} {:%Y-%m-%d %H:%M} qty: {} prc: {}{}{} order: {}'.format(self.symbol, pd.Timestamp(self.date).to_pydatetime(),
self.qty, self.price,
' ' + str(self.fee) if self.fee != 0 else '',
' ' + str(self.commission) if self.commission != 0 else '',
self.order)
[docs]class Portfolio:
'''A portfolio contains one or more strategies that run concurrently so you can test running strategies that are uncorrelated together.'''
[docs] def __init__(self, name = 'main'):
'''Args:
name: String used for displaying this portfolio
'''
self.name = name
self.strategies = {}
[docs] def add_strategy(self, name, strategy):
'''
Args:
name: Name of the strategy
strategy: Strategy object
'''
self.strategies[name] = strategy
strategy.portfolio = self
strategy.name = name
[docs] def run_indicators(self, strategy_names = None):
'''Compute indicators for the strategies specified
Args:
strategy_names: A list of strategy names. By default this is set to None and we use all strategies.
'''
if strategy_names is None: strategy_names = list(self.strategies.keys())
if len(strategy_names) == 0: raise Exception('a portofolio must have at least one strategy')
for name in strategy_names: self.strategies[name].run_indicators()
[docs] def run_signals(self, strategy_names = None):
'''Compute signals for the strategies specified. Must be called after run_indicators
Args:
strategy_names: A list of strategy names. By default this is set to None and we use all strategies.
'''
if strategy_names is None: strategy_names = list(self.strategies.keys())
if len(strategy_names) == 0: raise Exception('a portofolio must have at least one strategy')
for name in strategy_names: self.strategies[name].run_signals()
[docs] def run_rules(self, strategy_names = None, start_date = None, end_date = None, run_first = False, run_last = True):
'''Run rules for the strategies specified. Must be called after run_indicators and run_signals.
See run function for argument descriptions
'''
start_date, end_date = str2date(start_date), str2date(end_date)
if strategy_names is None: strategy_names = list(self.strategies.keys())
if len(strategy_names) == 0: raise Exception('a portofolio must have at least one strategy')
strategies = [self.strategies[key] for key in strategy_names]
min_date = min([strategy.dates[0] for strategy in strategies])
if start_date: min_date = max(min_date, start_date)
max_date = max([strategy.dates[-1] for strategy in strategies])
if end_date: max_date = min(max_date, end_date)
iter_list = []
for strategy in strategies:
dates, iterations = strategy._get_iteration_indices(start_date = start_date, end_date = end_date, run_first = run_first, run_last = run_last)
iter_list.append((strategy, dates, iterations))
dates_list = [tup[1] for tup in iter_list]
#for v in self.rebalance_rules.values():
# _, freq = v
# rebalance_dates = pd.date_range(min_date, max_date, freq = freq).values
# dates_list.append(rebalance_dates)
all_dates = np.array(reduce(np.union1d, dates_list))
iterations = [[] for x in range(len(all_dates))]
for tup in iter_list: # per strategy
strategy = tup[0]
dates = tup[1]
iter_tup = tup[2] # vector with list of (rule, symbol, iter_params dict)
for i, date in enumerate(dates):
idx = np.searchsorted(all_dates, date)
iterations[idx].append((Strategy._iterate, (strategy, i, iter_tup[i])))
#for name, tup in self.rebalance_rules.items():
# rule, freq = tup
# rebalance_dates = pd.date_range(all_dates[0], all_dates[-1], freq = freq).values
# rebalance_indices = np.where(np.in1d(all_dates, rebalance_dates))[0]
# iterations[idx].append(lambda : Portfolio.rebalance(rule), idx)
self.iterations = iterations # For debugging
for iter_idx, tup_list in enumerate(iterations):
for tup in tup_list:
func = tup[0]
args = tup[1]
func(*args)
[docs] def run(self, strategy_names = None, start_date = None, end_date = None, run_first = False, run_last = True):
'''
Run indicators, signals and rules.
Args:
strategy_names: A list of strategy names. By default this is set to None and we use all strategies.
start_date: Run rules starting from this date.
Sometimes we have a few strategies in a portfolio that need different lead times before they are ready to trade
so you can set this so they are all ready by this date. Default None
end_date: Don't run rules after this date. Default None
run_first: Force running rules on the first bar even if signals do not require this. Default False
run_last: Force running rules on penultimate bar even if signals do not require this.
'''
start_date, end_date = str2date(start_date), str2date(end_date)
self.run_indicators()
self.run_signals()
return self.run_rules(strategy_names, start_date, end_date, run_first, run_last)
[docs] def df_returns(self, sampling_frequency = 'D', strategy_names = None):
'''
Return dataframe containing equity and returns with a date index. Equity and returns are combined from all strategies passed in.
Args:
sampling_frequency: Date frequency for rows. Default 'D' for daily so we will have one row per day
strategy_names: A list of strategy names. By default this is set to None and we use all strategies.
'''
if strategy_names is None: strategy_names = list(self.strategies.keys())
if len(strategy_names) == 0: raise Exception('portfolio must have at least one strategy')
equity_list = []
for name in strategy_names:
equity = self.strategies[name].df_returns(sampling_frequency = sampling_frequency)[['equity']]
equity.columns = [name]
equity_list.append(equity)
df = pd.concat(equity_list, axis = 1)
df['equity'] = df.sum(axis = 1)
df['ret'] = df.equity.pct_change()
return df
[docs] def evaluate_returns(self, sampling_frequency = 'D', strategy_names = None, plot = True, float_precision = 4):
'''Returns a dictionary of common return metrics.
Args:
sampling_frequency: Date frequency. Default 'D' for daily so we downsample to daily returns before computing metrics
strategy_names: A list of strategy names. By default this is set to None and we use all strategies.
plot: If set to True, display plots of equity, drawdowns and returns. Default False
float_precision: Number of significant figures to show in returns. Default 4
'''
returns = self.df_returns(sampling_freq, strategy_names)
ev = compute_return_metrics(returns.index.values, returns.ret.values, returns.equity.values[0])
display_return_metrics(ev.metrics(), float_precision = float_precision)
if plot: plot_return_metrics(ev.metrics())
return ev.metrics()
[docs] def plot(self, sampling_frequency = 'D', strategy_names = None):
'''Display plots of equity, drawdowns and returns
Args:
sampling_frequency: Date frequency. Default 'D' for daily so we downsample to daily returns before computing metrics
strategy_names: A list of strategy names. By default this is set to None and we use all strategies.
'''
returns = self.df_returns(sampling_frequency, strategy_names)
ev = compute_return_metrics(returns.index.values, returns.ret.values, returns.equity.values[0])
plot_return_metrics(ev.metrics())
def __repr__(self):
return f'{self.name} {self.strategies.keys()}'
[docs]class ContractPNL:
'''Computes pnl for a single contract over time given trades and market data'''
def __init__(self, contract):
self.symbol = contract.symbol
self.multiplier = contract.multiplier
self.marketdata = contract.marketdata
self.dates = self.marketdata.dates
self.unrealized = np.empty(len(self.dates), dtype = np.float) * np.nan; self.unrealized[0] = 0
self.realized = np.empty(len(self.dates), dtype = np.float) * np.nan; self.realized[0] = 0
#TODO: Add commission and fee from trades
self.commission = np.zeros(len(self.dates), dtype = np.float);
self.fee = np.zeros(len(self.dates), dtype = np.float);
self.net_pnl = np.empty(len(self.dates), dtype = np.float) * np.nan; self.net_pnl[0] = 0
self.position = np.empty(len(self.dates), dtype = np.float) * np.nan; self.position[0] = 0
self.close = self.marketdata.c
self._trades = []
self.open_trades = deque()
[docs] def add_trades(self, trades):
'''Args:
trades: A list of Trade objects
'''
self._trades += trades
[docs] def calc(self, prev_i, i):
'''Compute pnl and store it internally
Args:
prev_i: Start index to compute pnl from
i: End index to compute pnl to
'''
calc_trades = deque([trade for trade in self._trades if trade.date > self.dates[prev_i] and trade.date <= self.dates[i]])
if not np.isfinite(self.close[i]):
unrealized = self.unrealized[prev_i]
realized = 0.
else:
open_trades, unrealized, realized = _calc_pnl(self.open_trades, calc_trades, self.close[i], self.multiplier)
self.open_trades = open_trades
self.unrealized[i] = unrealized
self.realized[i] = self.realized[prev_i] + realized
trade_qty = sum([trade.qty for trade in calc_trades])
self.position[i] = self.position[prev_i] + trade_qty
self.net_pnl[i] = self.realized[i] + self.unrealized[i] - self.commission[i] - self.fee[i]
if np.isnan(self.net_pnl[i]):
raise Exception(f'net_pnl: nan i: {i} realized: {self.realized[i]} unrealized: {self.unrealized[i]} commission: {self.commission[i]} fee: {self.fee[i]}')
[docs] def trades(self, start_date = None, end_date = None):
'''Get a list of trades
Args:
start_date: A string or numpy datetime64. Trades with trade dates >= start_date will be returned. Default None
end_date: A string or numpy datetime64. Trades with trade dates <= end_date will be returned. Default None
'''
start_date, end_date = str2date(start_date), str2date(end_date)
trades = [trade for trade in self._trades if (start_date is None or trade.date >= start_date) and (end_date is None or trade.date <= end_date)]
return trades
[docs] def df(self):
'''Returns a pandas dataframe with pnl data, indexed by date'''
df = pd.DataFrame({'date' : self.dates, 'unrealized' : self.unrealized, 'realized' : self.realized,
'fee' : self.fee, 'net_pnl' : self.net_pnl, 'position' : self.position})
df.dropna(subset = ['unrealized', 'realized'], inplace = True)
df['symbol'] = self.symbol
return df[['symbol', 'date', 'unrealized', 'realized', 'fee', 'net_pnl', 'position']].set_index('date')
[docs]class Contract:
'''A Contract can be a real or virtual instrument. For example, for futures you may wish to create a single continous contract instead of
a contract for each future series
'''
[docs] def __init__(self, symbol, marketdata, multiplier = 1.):
'''
Args:
symbol: A unique string reprenting this contract. e.g IBM or WTI_FUTURE
multiplier: If you have to multiply price to get price per contract, set that multiplier there.
marketdata: A MarketData object containing prices for this contract.
'''
assert(isinstance(symbol, str) and len(symbol) > 0)
assert(multiplier > 0)
self.symbol = symbol
self.multiplier = multiplier
self.marketdata = marketdata
[docs]class Account:
'''An Account calculates pnl for a set of contracts'''
[docs] def __init__(self, contracts, starting_equity = 1.0e6, calc_frequency = 'D'):
'''
Args:
contracts: A list of Contract objects
starting_equity: Starting equity in account currency. Default 1.e6
calc_frequency: Account will calculate pnl at this frequency. Default 'D' for daily
'''
if calc_frequency != 'D': raise Exception('unknown calc frequency: {}'.format(calc_frequency))
self.calc_freq = calc_frequency
self.contract_pnls = defaultdict()
self.current_calc_index = 0
self.marketdata = {}
self.all_dates = None
self.starting_equity = starting_equity
if len(contracts) == 0:
raise Exception('must add at least one contract')
for contract in contracts:
self.add_contract(contract)
def _set_dates(self, dates):
if self.all_dates is not None and not np.array_equal(dates, all_dates):
raise Exception('all symbols in a strategy must have the same dates')
self.all_dates = dates
calc_dates = dates.astype('M8[D]')
self.calc_dates = np.unique(calc_dates)
self.calc_indices = np.searchsorted(dates, self.calc_dates, side='left') - 1
if self.calc_indices[0] == -1: self.calc_indices[0] = 0
self._equity = np.empty(len(dates), np.float) * np.nan;
self._equity[0] = self.starting_equity
[docs] def symbols(self):
return list(self.contract_pnls.keys())
[docs] def add_contract(self, contract):
if self.all_dates is None: self._set_dates(contract.marketdata.dates)
self.contract_pnls[contract.symbol] = ContractPNL(contract)
self.marketdata[contract.symbol] = contract.marketdata
def _add_trades(self, symbol, trades):
self.contract_pnls[symbol].add_trades(trades)
[docs] def calc(self, i):
'''
Computes P&L and stores it internally for all contracts.
Args:
i: Index to compute P&L at. Account remembers the last index it computed P&L up to and will compute P&L between these two indices
'''
calc_indices = self.calc_indices[:]
if self.current_calc_index == i: return
intermediate_calc_indices = np.ravel(np.where(np.logical_and(calc_indices > self.current_calc_index, calc_indices <= i)))
if not len(intermediate_calc_indices) or calc_indices[intermediate_calc_indices[-1]] != i:
calc_indices = np.append(calc_indices, i)
intermediate_calc_indices = np.append(intermediate_calc_indices, len(calc_indices) - 1)
for symbol, symbol_pnl in self.contract_pnls.items():
prev_calc_index = self.current_calc_index
for idx in intermediate_calc_indices:
calc_index = calc_indices[idx]
symbol_pnl.calc(prev_calc_index, calc_index)
self._equity[calc_index] = self._equity[prev_calc_index] + symbol_pnl.net_pnl[calc_index] - symbol_pnl.net_pnl[prev_calc_index]
# print(f'prev_calc_index: {prev_calc_index} calc_index: {calc_index} prev_equity: {self._equity[prev_calc_index]} net_pnl: {symbol_pnl.net_pnl[calc_index]} prev_net_pnl: {symbol_pnl.net_pnl[prev_calc_index]}')
prev_calc_index = calc_index
self.current_calc_index = i
[docs] def position(self, symbol, date):
'''Returns position for a symbol at a given date in number of contracts or shares. Will cause calculation if Account has not previously calculated
up to this date'''
i = self.find_index_before(date)
self.calc(i)
return self.contract_pnls[symbol].position[i]
[docs] def equity(self, date):
'''Returns equity in this account in Account currency. Will cause calculation if Account has not previously calculated up to this date'''
i = self.find_index_before(date)
self.calc(i)
return self._equity[i]
[docs] def trades(self, symbol = None, start_date = None, end_date = None):
'''Returns a list of trades with the given symbol and with trade date between (and including) start date and end date if they are specified.
If symbol is None trades for all symbols are returned'''
start_date, end_date = str2date(start_date), str2date(end_date)
if symbol is None:
trades = []
for symbol, sym_pnl in self.contract_pnls.items():
trades += sym_pnl.trades(start_date, end_date)
return trades
else:
return self.contract_pnls[symbol].trades(start_date, end_date)
[docs] def find_index_before(self, date):
'''Returns the market data index before or at date'''
return np.searchsorted(self.all_dates, date)
[docs] def transfer_cash(self, date, amount):
'''Move cash from one portfolio to another'''
i = self.find_index_before(date)
curr_equity = self.equity(date)
if (amount > curr_equity): amount = curr_equity # Cannot make equity negative
self._equity[i] -= amount
return amount
[docs] def df_pnl(self, symbol = None):
'''Returns a dataframe with P&L columns. If symbol is set to None (default), sums up P&L across symbols'''
if symbol:
ret = self.contract_pnls[symbol].df()
else:
dfs = []
for symbol, symbol_pnl in self.contract_pnls.items():
df = symbol_pnl.df()
dfs.append(df)
ret = pd.concat(dfs)
ret = ret.reset_index().groupby('date').sum()
df_equity = pd.DataFrame({'equity' : self._equity}, index = self.all_dates).dropna()
ret = pd.merge(ret, df_equity, left_index = True, right_index = True, how = 'outer')
ret.index.name = 'date'
return ret
[docs] def df_trades(self, symbol = None, start_date = None, end_date = None):
'''Returns a dataframe with data from trades with the given symbol and with trade date between (and including) start date and end date
if they are specified. If symbol is None, trades for all symbols are returned'''
start_date, end_date = str2date(start_date), str2date(end_date)
if symbol:
trades = self.contract_pnls[symbol].trades(start_date, end_date)
else:
trades = [v.trades(start_date, end_date) for v in self.contract_pnls.values()]
trades = [trade for sublist in trades for trade in sublist] # flatten list
df = pd.DataFrame.from_records([(trade.symbol, trade.date, trade.qty, trade.price, trade.fee, trade.commission, trade.order.date, trade.order.qty, trade.order.params()) for trade in trades],
columns = ['symbol', 'date', 'qty', 'price', 'fee', 'commission', 'order_date', 'order_qty', 'order_params'])
return df
[docs]class Strategy:
[docs] def __init__(self, contracts, starting_equity = 1.0e6, calc_frequency = 'D'):
'''
Args:
contracts: A list of contract objects
starting_equity: Starting equity in Strategy currency. Default 1.e6
calc_frequency: How often P&L is calculated. Default is 'D' for daily
'''
self.name = None
self.account = Account(contracts, starting_equity, calc_frequency)
self.symbols = [contract.symbol for contract in contracts]
self.indicators = {}
self.indicator_values = defaultdict(dict)
self.signals = {}
self.signal_values = defaultdict(dict)
self.rules = {}
self.rule_signals = {}
self.market_sims = {}
self._trades = defaultdict(list)
self._orders = []
self.dates = self.account.all_dates
[docs] def add_indicator(self, name, indicator_function):
'''
Args:
name: Name of the indicator
indicator_function: A function taking a MarketData object and returning a numpy array
containing indicator values. The return array must have the same length as the MarketData object
'''
self.indicators[name] = indicator_function
[docs] def add_signal(self, name, signal_function):
'''
Args:
name: Name of the signal
signal_function: A function taking a MarketData object and a dictionary of indicator value arrays as input and returning a numpy array
containing signal values. The return array must have the same length as the MarketData object
'''
self.signals[name] = signal_function
[docs] def add_rule(self, name, rule_function, signal_name, sig_true_values):
'''Add a trading rule
Args:
name: Name of the trading rule
rule_function: A trading rule function that returns a list of Orders
signal_name: The strategy will call the trading rule function when the signal with this name matches sig_true_values
sig_true_values: A numpy array of values. If the signal value at a bar is equal to one of these, the Strategy will call the trading rule function
'''
self.rule_signals[name] = (signal_name, sig_true_values)
self.rules[name] = rule_function
[docs] def add_market_sim(self, market_sim_function, symbols = None):
'''Add a market simulator. A market simulator takes a list of Orders as input and returns a list of Trade objects.
Args:
market_sim_function: A function that takes a list of Orders and MarketData as input and returns a list of Trade objects
symbols: A list of the symbols that this market_sim_function applies to. If None (default) it will apply to all symbols
'''
if symbols is None: symbols = self.symbols
for symbol in symbols: self.market_sims[symbol] = market_sim_function
[docs] def run_indicators(self, indicator_names = None, symbols = None):
'''Calculate values of the indicators specified and store them.
Args:
indicator_names: List of indicator names. If None (default) run all indicators
symbols: List of symbols to run these indicators for. If None (default) use all symbols
'''
if indicator_names is None: indicator_names = self.indicators.keys()
if symbols is None: symbols = self.symbols
for indicator_name in indicator_names:
indicator_function = self.indicators[indicator_name]
for symbol in symbols:
marketdata = self.account.marketdata[symbol]
self.indicator_values[symbol][indicator_name] = series_to_array(indicator_function(marketdata))
[docs] def run_signals(self, signal_names = None, symbols = None):
'''Calculate values of the signals specified and store them.
Args:
signal_names: List of signal names. If None (default) run all signals
symbols: List of symbols to run these signals for. If None (default) use all symbols
'''
if signal_names is None: signal_names = self.signals.keys()
if symbols is None: symbols = self.symbols
for signal_name in signal_names:
signal_function = self.signals[signal_name]
for symbol in symbols:
marketdata = self.account.marketdata[symbol]
self.signal_values[symbol][signal_name] = series_to_array(signal_function(marketdata, self.indicator_values[symbol]))
[docs] def run_rules(self, rule_names = None, symbols = None, start_date = None, end_date = None, run_first = False, run_last = True):
'''Run trading rules.
Args:
rule_names: List of rule names. If None (default) run all rules
symbols: List of symbols to run these signals for. If None (default) use all symbols
start_date: Run rules starting from this date. Default None
end_date: Don't run rules after this date. Default None
run_first: Force running rules on the first bar even if signals do not require this. Default False
run_last: Force running rules on penultimate bar even if signals do not require this.
'''
start_date, end_date = str2date(start_date), str2date(end_date)
dates, iterations = self._get_iteration_indices(rule_names, symbols, start_date, end_date, run_first, run_last)
# Now we know which rules, symbols need to be applied for each iteration, go through each iteration and apply them
# in the same order they were added to the strategy
for i, tup_list in enumerate(iterations):
self._iterate(i, tup_list)
def _get_iteration_indices(self, rule_names = None, symbols = None, start_date = None, end_date = None,
run_first = False, run_last = True):
start_date, end_date = str2date(start_date), str2date(end_date)
if rule_names is None: rule_names = self.rules.keys()
if symbols is None: symbols = self.symbols
dates = self.dates
num_dates = len(dates)
iterations = [[] for x in range(num_dates)]
self.orders_iter = [[] for x in range(num_dates)]
for rule_name in rule_names:
rule_function = self.rules[rule_name]
for symbol in symbols:
marketdata = self.account.marketdata[symbol]
market_sim = self.market_sims[symbol]
signal_name = self.rule_signals[rule_name][0]
sig_true_values = self.rule_signals[rule_name][1]
sig_values = self.signal_values[symbol][signal_name]
dates = marketdata.dates
null_value = False if sig_values.dtype == np.dtype('bool') else np.nan
if start_date: sig_values[0:np.searchsorted(dates, start_date)] = null_value
if end_date: sig_values[np.searchsorted(dates, end_date):] = null_value
indices = np.nonzero(np.isin(sig_values, sig_true_values))[0]
if indices[-1] == len(sig_values) -1: indices = indices[:-1] # Don't run rules on last index since we cannot fill any orders
if run_first and indices[0] != 0: indices = np.insert(indices, 0, 0)
if run_last and indices[-1] != len(sig_values) - 2: indices = np.append(indices, len(sig_values) - 2)
indicator_values = self.indicator_values[symbol]
iteration_params = {'market_sim' : market_sim, 'indicator_values' : indicator_values, 'signal_values' : sig_values, 'marketdata' : marketdata}
for idx in indices: iterations[idx].append((rule_function, symbol, iteration_params))
self.iterations = iterations # For debugging
return self.dates, iterations
def _iterate(self, i, tup_list):
for tup in self.orders_iter[i]:
try:
open_orders, symbol, params = tup
open_orders = self._sim_market(i, open_orders, symbol, params)
if len(open_orders): self.orders_iter[i + 1].append((open_orders, symbol, params))
except Exception as e:
raise type(e)(f'Exception: {str(e)} at rule: {type(tup[0])} symbol: {tup[1]} index: {i}').with_traceback(sys.exc_info()[2])
for tup in tup_list:
try:
rule_function, symbol, params = tup
open_orders = self._get_orders(i, rule_function, symbol, params)
self._orders += open_orders
if len(open_orders): self.orders_iter[i + 1].append((open_orders, symbol, params))
except Exception as e:
raise type(e)(f'Exception: {str(e)} at rule: {type(tup[0])} symbol: {tup[1]} index: {i}').with_traceback(sys.exc_info()[2])
def _get_orders(self, idx, rule_function, symbol, params):
indicator_values, signal_values, marketdata = (params['indicator_values'], params['signal_values'], params['marketdata'])
open_orders = rule_function(self, symbol, idx, self.dates[idx], marketdata, indicator_values, signal_values, self.account)
return open_orders
def _sim_market(self, idx, open_orders, symbol, params):
'''
Keep iterating while we have open orders since they may get filled
TODO: For limit orders and trigger orders we can be smarter here and reduce indices like quantstrat does
'''
market_sim_function = params['market_sim']
trades = market_sim_function(self, open_orders, idx, self.dates[idx], self.account.marketdata[symbol])
if len(trades) == 0: return []
self._trades[symbol] += trades
self.account._add_trades(symbol, trades)
self.account.calc(idx)
open_orders = [order for order in open_orders if order.status == 'open']
return open_orders
[docs] def df_data(self, symbols = None, add_pnl = True, start_date = None, end_date = None):
'''
Add indicators and signals to end of market data and return as a pandas dataframe.
Args:
symbols: list of symbols to include. All if set to None (default)
add_pnl: If True (default), include P&L columns in dataframe
start_date: string or numpy datetime64. Default None
end_date: string or numpy datetime64: Default None
'''
start_date, end_date = str2date(start_date), str2date(end_date)
if symbols is None: symbols = self.symbols
if not isinstance(symbols, list): symbols = [symbols]
mds = []
for symbol in symbols:
md = self.account.marketdata[symbol].df(start_date, end_date)
md.insert(0, 'symbol', symbol)
if add_pnl:
df_pnl = self.account.df_pnl(symbol)
del df_pnl['symbol']
indicator_values = self.indicator_values[symbol]
for k in sorted(indicator_values.keys()):
name = k
if name in md.columns: name = name + '.ind' # if we have a market data column with the same name as the indicator
md.insert(len(md.columns), name, indicator_values[k])
signal_values = self.signal_values[symbol]
for k in sorted(signal_values.keys()):
name = k
if name in md.columns: name = name + '.sig'
md.insert(len(md.columns), name, signal_values[k])
if add_pnl: md = pd.merge(md, df_pnl, left_index = True, right_index = True, how = 'left')
# Add counter column for debugging
md.insert(len(md.columns), 'i', np.arange(len(md)))
mds.append(md)
return pd.concat(mds)
[docs] def marketdata(self, symbol):
'''Return MarketData object for this symbol'''
return self.account.marketdata[symbol]
[docs] def trades(self, symbol = None, start_date = None, end_date = None):
'''Returns a list of trades with the given symbol and with trade date between (and including) start date and end date if they are specified.
If symbol is None trades for all symbols are returned'''
start_date, end_date = str2date(start_date), str2date(end_date)
return self.account.trades(symbol, start_date, end_date)
[docs] def df_trades(self, symbol = None, start_date = None, end_date = None):
'''Returns a dataframe with data from trades with the given symbol and with trade date between (and including) start date and end date
if they are specified. If symbol is None, trades for all symbols are returned'''
start_date, end_date = str2date(start_date), str2date(end_date)
return self.account.df_trades(symbol, start_date, end_date)
[docs] def orders(self, symbol = None, start_date = None, end_date = None):
'''Returns a list of orders with the given symbol and with order date between (and including) start date and end date if they are specified.
If symbol is None orders for all symbols are returned'''
start_date, end_date = str2date(start_date), str2date(end_date)
return [order for order in self._orders if (symbol is None or order.symbol == symbol) and (
start_date is None or order.date >= start_date) and (end_date is None or order.date <= end_date)]
[docs] def df_orders(self, symbol = None, start_date = None, end_date = None):
'''Returns a dataframe with data from orders with the given symbol and with order date between (and including) start date and end date
if they are specified. If symbol is None, orders for all symbols are returned'''
start_date, end_date = str2date(start_date), str2date(end_date)
orders = self.orders(symbol, start_date, end_date)
df_orders = pd.DataFrame.from_records([(order.symbol, type(order).__name__, order.date, order.qty, order.params())
for order in orders], columns = ['symbol', 'type', 'date', 'qty', 'params'])
return df_orders
[docs] def df_pnl(self, symbol = None):
'''Returns a dataframe with P&L columns. If symbol is set to None (default), sums up P&L across symbols'''
return self.account.df_pnl(symbol)
[docs] def df_returns(self, symbol = None, sampling_frequency = 'D'):
'''Return a dataframe of returns and equity indexed by date.
Args:
symbol: The symbol to get returns for. If set to None (default), this returns the sum of PNL for all symbols
sampling_frequency: Downsampling frequency. Default is None. See pandas frequency strings for possible values
'''
pnl = self.df_pnl(symbol)[['equity']]
pnl.equity = pnl.equity.ffill()
pnl = pnl.resample(sampling_frequency).last()
pnl['ret'] = pnl.equity.pct_change()
return pnl
[docs] def plot(self, symbols = None, md_columns = 'c', pnl_columns = 'equity', title = None, figsize = (20, 15), date_range = None,
date_format = None, sampling_frequency = None, trade_marker_properties = None, hspace = 0.15):
'''Plot indicators, signals, trades, position, pnl
Args:
symbols: List of symbols or None (default) for all symbols
md_columns: List of columns of market data to plot. Default is 'c' for close price. You can set this to 'ohlcv' if you want to plot
a candlestick of OHLCV data
pnl_columns: List of P&L columns to plot. Default is 'equity'
title: Title of plot (None)
figsize: Figure size. Default is (20, 15)
date_range: Tuple of strings or datetime64, e.g. ("2018-01-01", "2018-04-18 15:00") to restrict the graph. Default None
date_format: Date format for tick labels on x axis. If set to None (default), will be selected based on date range. See matplotlib date format strings
sampling_frequency: Downsampling frequency. Default is None. The graph may get too busy if you have too many bars of data, in which case you may want to
downsample before plotting. See pandas frequency strings for possible values
trade_marker_properties: A dictionary of order reason code -> marker shape, marker size, marker color for plotting trades with different reason codes.
Default is None in which case the dictionary from the ReasonCode class is used
hspace: Height (vertical) space between subplots. Default is 0.15
'''
date_range = strtup2date(date_range)
if symbols is None: symbols = self.symbols
if not isinstance(symbols, list): symbols = [symbols]
if not isinstance(md_columns, list): md_columns = [md_columns]
if not isinstance(pnl_columns, list): pnl_columns = [pnl_columns]
for symbol in symbols:
md = self.marketdata(symbol)
md_dates = md.dates
if md_columns == ['ohlcv']:
md_list = [OHLC('price', dates = md_dates, o = md.o, h = md.h, l = md.l, c = md.c, v = md.v)]
else:
md_list = [TimeSeries(md_column, dates = md_dates, values = getattr(md, md_column)) for md_column in md_columns]
indicator_list = [TimeSeries(indicator_name, dates = md_dates, values = self.indicator_values[symbol][indicator_name], line_type = '--'
) for indicator_name in self.indicators.keys() if indicator_name in self.indicator_values[symbol]]
signal_list = [TimeSeries(signal_name, dates = md_dates, values = self.signal_values[symbol][signal_name]
) for signal_name in self.signals.keys() if signal_name in self.signal_values[symbol]]
df_pnl_ = self.df_pnl(symbol)
pnl_list = [TimeSeries(pnl_column, dates = df_pnl_.index.values, values = df_pnl_[pnl_column].values) for pnl_column in pnl_columns]
if trade_marker_properties:
trade_sets = trade_sets_by_reason_code(self._trades[symbol], trade_marker_properties)
else:
trade_sets = trade_sets_by_reason_code(self._trades[symbol])
main_subplot = Subplot(indicator_list + md_list + trade_sets, height_ratio = 0.5, ylabel = 'Indicators')
signal_subplot = Subplot(signal_list, ylabel = 'Signals', height_ratio = 0.167)
pnl_subplot = Subplot(pnl_list, ylabel = 'Equity', height_ratio = 0.167, log_y = True, y_tick_format = '${x:,.0f}')
position = df_pnl_.position.values
pos_subplot = Subplot([TimeSeries('position', dates = df_pnl_.index.values, values = position, plot_type = 'filled_line')],
ylabel = 'Position', height_ratio = 0.167)
plot = Plot([main_subplot, signal_subplot, pos_subplot, pnl_subplot], figsize = figsize,
date_range = date_range, date_format = date_format, sampling_frequency = sampling_frequency, title = title, hspace = hspace)
plot.draw()
[docs] def evaluate_returns(self, symbol = None, plot = True, float_precision = 4):
'''Returns a dictionary of common return metrics.
Args:
sampling_frequency: Date frequency. Default 'D' for daily so we downsample to daily returns before computing metrics
strategy_names: A list of strategy names. By default this is set to None and we use all strategies.
plot: If set to True, display plots of equity, drawdowns and returns. Default False
float_precision: Number of significant figures to show in returns. Default 4
'''
returns = self.df_returns(symbol)
ev = compute_return_metrics(returns.index.values, returns.ret.values, self.account.starting_equity)
display_return_metrics(ev.metrics(), float_precision = float_precision)
if plot: plot_return_metrics(ev.metrics())
return ev.metrics()
[docs] def plot_returns(self, symbol = None):
'''Display plots of equity, drawdowns and returns for the given symbol or for all symbols if symbol is None (default)'''
if symbol is None:
symbols = self.symbols()
else:
symbols = [symbol]
df_list = []
for symbol in symbols:
df_list.append(self.df_returns(symbol))
df = pd.concat(df_list, axis = 1)
ev = compute_return_metrics(returns.index.values, returns.ret.values, self.account.starting_equity)
plot_return_metrics(ev.metrics())
def __repr__(self):
return f'{pformat(self.indicators)} {pformat(self.rules)} {pformat(self.account)}'
[docs]def test_strategy():
from datetime import datetime, timedelta
set_defaults()
def sim_order(order, i, date, md):
symbol = order.symbol
trade_price = np.nan
skid_fraction = 0.5
if not md.valid_row(i): return None
if isinstance(order, MarketOrder):
if order.qty > 0:
trade_price = 0.5 * (md.c[i] + md.h[i])
else:
trade_price = 0.5 * (md.c[i] + md.l[i])
elif order.qty > 0 and md.h[i] > order.trigger_price:
trade_price = skid_fraction * max(md.o[i], order.trigger_price, md.l[i]) + (1 - skid_fraction) * md.h[i]
elif order.qty < 0 and md.l[i] < order.trigger_price:
trade_price = skid_fraction * min(md.o[i], order.trigger_price, md.h[i]) + (1 - skid_fraction) * md.l[i]
else:
return None
sim_trade = Trade(symbol, date, order.qty, trade_price, order = order)
return sim_trade
def market_simulator(strategy, orders, i, date, marketdata):
trades = []
for order in orders:
sim_trade = sim_order(order, i, date, marketdata)
if sim_trade is None: continue
if math.isclose(sim_trade.qty, order.qty): order.status = 'filled'
trades.append(sim_trade)
return trades
def trade_rule(strategy, symbol, i, date, marketdata, indicator_values, signal_values, account):
heat = 0.05
reason_code = None
if not marketdata.valid_row(i): return []
curr_pos = account.position(symbol, date)
if i == len(marketdata.dates) - 2: # Last date so get out of position
if not math.isclose(curr_pos, 0):
return [MarketOrder(symbol, date, -curr_pos, reason_code = ReasonCode.BACKTEST_END)]
else:
return []
trend = signal_values[i]
fast_resistance, fast_support, slow_resistance, slow_support = (indicator_values['fast_resistance'][i],
indicator_values['fast_support'][i], indicator_values['slow_resistance'][i], indicator_values['slow_support'][i])
if trend == 1:
entry_limit = fast_resistance
stop_limit = fast_support
elif trend == -1:
entry_limit = fast_support
stop_limit = fast_resistance
else:
return []
if math.isclose(curr_pos, 0): # We got a trade in the previous bar so put in a stop limit order
if math.isclose(entry_limit, stop_limit): return []
curr_equity = account.equity(date)
order_qty = curr_equity * heat / (entry_limit - stop_limit)
trigger_price = entry_limit
reason_code = ReasonCode.ENTER_LONG if order_qty > 0 else ReasonCode.ENTER_SHORT
else:
order_qty = -curr_pos
trigger_price = stop_limit
reason_code = ReasonCode.EXIT_LONG if order_qty < 0 else ReasonCode.EXIT_SHORT
order_qty = round(order_qty)
if np.isnan(order_qty):
raise Exception(f'Got nan order qty date: {date} i: {i} curr_pos: {curr_pos} curr_equity: {curr_equity} entry_limit: {entry_limit} stop_limit: {stop_limit}')
if math.isclose(order_qty, 0): return []
order = StopLimitOrder(symbol, date, order_qty, trigger_price, reason_code = reason_code)
return [order]
def get_support(lows, n): return pd.Series(lows).rolling(window = n, min_periods = 1).min().values
def get_resistance(highs, n): return pd.Series(highs).rolling(window = n, min_periods = 1).max().values
def get_trend(md, ind):
trend = pd.Series(np.where(pd.Series(md.h) > shift_np(ind['slow_resistance'], 1), 1,
np.where(pd.Series(md.l) < shift_np(ind['slow_support'], 1), -1,
np.nan)))
trend.fillna(method = 'ffill', inplace = True)
return trend.values
def build_strategy(contract, fast_interval, slow_interval):
strategy = Strategy([contract])
strategy.add_indicator('slow_resistance', lambda md : get_resistance(md.h, slow_interval))
strategy.add_indicator('slow_support', lambda md : get_support(md.l, slow_interval))
strategy.add_indicator('fast_resistance', lambda md : get_resistance(md.h, fast_interval))
strategy.add_indicator('fast_support', lambda md : get_support(md.l, fast_interval))
strategy.add_signal('trend', get_trend)
strategy.add_market_sim(market_simulator)
strategy.add_rule('trade_rule', trade_rule, 'trend', np.array([-1, 1]))
return strategy
np.random.seed(0)
dates = np.arange(datetime(2018, 1, 1, 9, 0, 0), datetime(2018, 3, 1, 16, 0, 0), timedelta(minutes = 5))
dates = np.array([dt for dt in dates.astype(object) if dt.hour >= 9 and dt.hour <= 16]).astype('M8[m]')
rets = np.random.normal(size = len(dates)) / 1000
c_0 = 100
c = np.round(c_0 * np.cumprod(1 + rets), 2)
l = np.round(c * (1. - np.abs(np.random.random(size = len(dates)) / 1000.)), 2)
h = np.round(c * (1. + np.abs(np.random.random(size = len(dates)) / 1000.)), 2)
o = np.round(l + (h - l) * np.random.random(size = len(dates)), 2)
v = np.round(np.random.normal(size = len(dates)) * 100)
portfolio = Portfolio()
slow_interval = 0
for days in [0.5, 1, 2]:
# 1 day from 9 - 4 pm has 7 hours which translate to 7 x 12 = 84 5 minute periods
fast_interval = round(days * 84)
slow_interval = round(5 * fast_interval)
contract = Contract('IBM', MarketData(dates, c, o, h, l, v))
portfolio.add_strategy(f'strat_{days}', build_strategy(contract, fast_interval, slow_interval))
# Start at max slow days so all strategies start at the same time
print('running')
portfolio.run(start_date = dates[slow_interval])
print('done')
strat1 = portfolio.strategies['strat_0.5']
portfolio.plot();
strat1.plot();
if __name__ == "__main__":
test_strategy()