Overall Statistics |
Total Orders
626
Average Win
3.00%
Average Loss
-2.95%
Compounding Annual Return
-1.227%
Drawdown
52.000%
Expectancy
0.045
Start Equity
100000
End Equity
73573.90
Net Profit
-26.426%
Sharpe Ratio
-0.256
Sortino Ratio
-0.293
Probabilistic Sharpe Ratio
0.000%
Loss Rate
48%
Win Rate
52%
Profit-Loss Ratio
1.02
Alpha
-0.024
Beta
-0.045
Annual Standard Deviation
0.1
Annual Variance
0.01
Information Ratio
-0.353
Tracking Error
0.194
Treynor Ratio
0.565
Total Fees
$1377.39
Estimated Strategy Capacity
$1800000.00
Lowest Capacity Asset
IWP S6I9NW8U0D45
Portfolio Turnover
3.10%
|
#region imports from AlgorithmImports import * #endregion # https://quantpedia.com/strategies/momentum-factor-and-style-rotation-effect/ # # Russell’s ETFs for six equity styles are used # (small-cap value, mid-cap value, large-cap value, small-cap growth, mid-cap growth, large-cap growth). # Each month, the investor calculates 12-month momentum for each style and goes long on the winner and short on the loser. # The portfolio is rebalanced each month. # # QC Implementation: # - Trading IVW in 10/2020 is skipped due to data error. class MomentumFactorAndStyleRotationEffect(QCAlgorithm): def Initialize(self): self.SetStartDate(2000, 1, 1) self.SetCash(100000) self.tickers = [ 'IWS', # iShares Russell Midcap Value ETF 'IWP', # iShares Russell Midcap Growth ETF 'IWN', # iShares Russell 2000 Value ETF 'IWO', # iShares Russell 2000 Growth ETF 'IVE', # iShares S&P 500 Value ETF 'IVW' # iShares S&P 500 Growth ETF ] self.mom = {} self.period = 12 * 21 self.SetWarmUp(self.period) for ticker in self.tickers: security = self.AddEquity(ticker, Resolution.Daily) security.SetFeeModel(CustomFeeModel()) security.SetLeverage(10) self.mom[security.Symbol] = self.MOM(security.Symbol, self.period) self.recent_month = -1 def OnData(self, data): if self.recent_month == self.Time.month: return self.recent_month = self.Time.month mom_ready = [ s for s in self.mom if self.mom[s].IsReady and s in data] if mom_ready: sorted_mom = sorted(mom_ready, key = lambda x: self.mom[x].Current.Value, reverse=True) for symbol in sorted_mom[1:-1]: if self.Portfolio[symbol].Invested: self.Liquidate(symbol) winner = sorted_mom[0] loser = sorted_mom[-1] if self.Securities[winner].Price != 0 and self.Securities[winner].IsTradable: if (self.Time.month == 10 and self.Time.year == 2020) and winner.Value == 'IVW': # prevent data error self.Liquidate(winner) else: self.SetHoldings(winner, 1) if self.Securities[loser].Price != 0 and self.Securities[loser].IsTradable: if (self.Time.month == 10 and self.Time.year == 2020) and loser.Value == 'IVW': # prevent data error self.Liquidate(loser) else: self.SetHoldings(loser, -1) # Custom fee model. class CustomFeeModel(FeeModel): def GetOrderFee(self, parameters): fee = parameters.Security.Price * parameters.Order.AbsoluteQuantity * 0.00005 return OrderFee(CashAmount(fee, "USD"))