| Overall Statistics |
|
Total Orders
3
Average Win
0%
Average Loss
0%
Compounding Annual Return
0.492%
Drawdown
18.400%
Expectancy
0
Start Equity
100000
End Equity
111099.06
Net Profit
11.099%
Sharpe Ratio
-0.669
Sortino Ratio
-0.382
Probabilistic Sharpe Ratio
0.000%
Loss Rate
0%
Win Rate
0%
Profit-Loss Ratio
0
Alpha
-0.02
Beta
0.017
Annual Standard Deviation
0.029
Annual Variance
0.001
Information Ratio
-0.502
Tracking Error
0.155
Treynor Ratio
-1.15
Total Fees
$5.66
Estimated Strategy Capacity
$23000000.00
Lowest Capacity Asset
AGG SSC0EI5J2F6T
Portfolio Turnover
0.01%
|
# https://quantpedia.com/strategies/paired-switching/
#
# This strategy is very flexible. Investors could use stocks, funds, or ETFs as an investment vehicle. We show simple trading rules for a sample strategy
# from the source research paper. The investor uses two Vanguard funds as his investment vehicles – one equity fund (VFINX) and one government bond
from AlgorithmImports import *
# fund (VUSTX). These two funds have a negative correlation as they are proxies for two negatively correlated asset classes. The investor looks at the
# performance of the two funds over the prior quarter and buys the fund that has a higher return during the ranking period. The position is held for one
# quarter (the investment period). At the end of the investment period, the cycle is repeated.
class PairedSwitching(QCAlgorithm):
def Initialize(self):
self.SetStartDate(2004, 1, 1)
self.SetCash(100000)
self.first_symbol = self.AddEquity("SPY", Resolution.Daily).Symbol
self.second_symbol = self.AddEquity("AGG", Resolution.Daily).Symbol
self.recent_month = -1
def OnData(self, data):
if self.Time.month == self.recent_month:
return
self.recent_month = self.Time.month
if(self.recent_month % 3 == 0):
if self.first_symbol in data and self.second_symbol in data:
history_call = self.History([self.first_symbol, self.second_symbol], timedelta(days=90))
if not history_call.empty:
first_bars = history_call.loc[self.first_symbol.Value]
last_p1 = first_bars["close"].iloc[0]
second_bars = history_call.loc[self.second_symbol.Value]
last_p2 = second_bars["close"].iloc[0]
# Calculates performance of funds over the prior quarter.
first_performance = (float(self.Securities[self.first_symbol].Price) - float(last_p1)) / (float(self.Securities[self.first_symbol].Price))
second_performance = (float(self.Securities[self.second_symbol].Price) - float(last_p2)) / (float(self.Securities[self.second_symbol].Price))
# Buys the fund that has the higher return during the period.
if(first_performance > second_performance):
if(self.Securities[self.second_symbol].Invested):
self.Liquidate(self.second_symbol)
self.SetHoldings(self.first_symbol, 1)
else:
if(self.Securities[self.first_symbol].Invested):
self.Liquidate(self.first_symbol)
self.SetHoldings(self.second_symbol, 1)