Overall Statistics |
Total Trades 986 Average Win 0.30% Average Loss -0.10% Compounding Annual Return 11.321% Drawdown 17.500% Expectancy 1.224 Net Profit 80.561% Sharpe Ratio 0.875 Probabilistic Sharpe Ratio 33.659% Loss Rate 43% Win Rate 57% Profit-Loss Ratio 2.91 Alpha 0.093 Beta 0.058 Annual Standard Deviation 0.113 Annual Variance 0.013 Information Ratio -0.018 Tracking Error 0.196 Treynor Ratio 1.723 Total Fees $1171.51 |
import numpy as np class TransdimensionalOptimizedAtmosphericScrubbers(QCAlgorithm): def Initialize(self): self.SetStartDate(2015, 4, 29) # Set Start Date self.SetCash(100000) # Set Strategy Cash res = Resolution.Daily self.STOCKS = [self.AddEquity('QQQ', res).Symbol] self.BONDS = [self.AddEquity(ticker, res).Symbol for ticker in ['TLT', 'IEF']] self.XLI = self.AddEquity('XLI', res).Symbol self.XLU = self.AddEquity('XLU', res).Symbol self.UUP = self.AddEquity('UUP', res).Symbol self.MKT = self.STOCKS[0] self.VOLA = 126; self.BULL = 1; self.COUNT = 0; self.OUT_DAY = 0; self.RET_INITIAL = 80; self.LEV = 1.00; self.Schedule.On(self.DateRules.EveryDay(), self.TimeRules.AfterMarketOpen('QQQ', 140), self.daily_check) def daily_check(self): vola = self.History([self.MKT], self.VOLA + 1, Resolution.Daily).loc[self.MKT][ 'close'].pct_change().std() * np.sqrt(252) WAIT_DAYS = int(vola * self.RET_INITIAL) RET = int((1.0 - vola) * self.RET_INITIAL) P = self.History([self.XLI, self.XLU, self.UUP], RET + 2, Resolution.Daily)['close'].unstack(level=0).iloc[:-1].dropna() if (len(P.columns) < 2): return ratio = (P[self.XLI].iloc[-1] / P[self.XLI].iloc[0]) / (P[self.XLU].iloc[-1] / P[self.XLU].iloc[0]) exit = ratio < 1.0 if exit: self.BULL = 0; self.OUT_DAY = self.COUNT; elif (self.COUNT >= self.OUT_DAY + WAIT_DAYS): self.BULL = 1 self.COUNT += 1 wt_stk = self.LEV if self.BULL else 0; wt_bnd = 0 if self.BULL else self.LEV; wt = {} for sec in self.STOCKS: wt[sec] = wt_stk / len(self.STOCKS); for sec in self.BONDS: wt[sec] = wt_bnd / len(self.BONDS) for sec, weight in wt.items(): self.SetHoldings(sec, weight)