Is this normal, 199minutes, 200 minutes and 201 minutes gives me completely different results

Zimman

Can you share a backtest?

Hi Vladimir 

The attached backtest gives me -24%  loss on 201 minutes bars, the 200 minutes bars returns -12% loss.

I was also wondering if you can help me on this please man, am struggling to have multiple timeframes on one indicator.

For example, to buy QQQ if 201 minutes bar and 30 minutes bar and 1 minutes bar all have a Buy  Super Trend Signal.

Thank you 

Zimman

I see some problems in your code.

Your algorithm log has 7 pages of Insufficient buying power to complete order.

Try this

Zimman

Here is the code snippet of the above attached backtest.

# super trend master

from AlgorithmImports import *
# -------------------------------------------------------------------------
STOCK = "QQQ";  BAR = 199; SL = -0.015; TP = 0.005; ATR = 14; MULT = 0.15; 
# -------------------------------------------------------------------------
class SuperTrendIndicator(QCAlgorithm):
    
    def Initialize(self):
        self.SetStartDate(DateTime(2020, 6, 10, 9, 30, 0))  
        self.SetEndDate(DateTime(2022, 6, 10, 16, 0, 0)) 
        self.SetCash(200000)
        res = Resolution.Minute
        self.stock = self.AddEquity(STOCK, res).Symbol
        consolidator = TradeBarConsolidator(timedelta(minutes = BAR))
        self.Consolidate(self.stock, timedelta(minutes = BAR), self.BarHandler)
        self.st = SuperTrend(ATR, MULT, MovingAverageType.Wilders)
        self.RegisterIndicator(self.stock, self.st, consolidator)
        self.SetWarmUp(5*BAR*ATR, res)

        
    def BarHandler(self, consolidated):        
        if self.IsWarmingUp: return
        if not self.st.IsReady: return 
   
        self.Plot(STOCK, "Price", self.Securities[self.stock].Price)
        self.Plot(STOCK, "Super Trend", self.st.Current.Value)

        pnl = self.Securities[self.stock].Holdings.UnrealizedProfitPercent
        
        if not self.Portfolio[self.stock].IsLong:
            if self.Securities[self.stock].High > self.st.Current.Value:  
                self.SetHoldings(self.stock, 1, True, "Buy Signal")

        elif not self.Portfolio[self.stock].IsShort:
            if self.Securities[self.stock].Low < self.st.Current.Value: 
                self.SetHoldings(self.stock, -1, True, "Sell Signal") 

        elif self.Portfolio[self.stock].Invested: 
            if pnl < SL:
                self.Liquidate(self.stock, "Stop Loss")
            elif pnl > TP:
                self.Liquidate(self.stock, "Take Profit")

Hi ..

I am.just wondering why Drawdwon drops so low, even when fees and losses are negligible. 

This strategy is supposed to liquidate a position once the signal changes direction. I have seen drawdwon dropping to -30%, and I can't seem to understand what that happens. I understand there could be slippages, but the long position should still liquidate on the next sell signal, or the sell signal after that… 

For example lets say i enter a long position at $100, and price drops straight away (slippage) to say $98…. and it then goes up to 99.50 and drops again …In this case if would expext the long position to liquidate at 99.50 sell signal… unfortunately if the price keeps a downward trend with several buy and sell signals until it reaches $50 , my $100 long position still remains open with the drawdown getting worse….

How can i avoid this using only the buy sell signal triggers,  avoiding an arbitrary stop loss level? 

Zimman

I am.just wondering why Drawdwon drops so low, even when fees and losses are negligible. 

Did you have a chance to try what I recommended?

Compounding Annual Return 37.337%

Drawdown                           12.200%

If you set a stoploss in your algorithm, the drawdown would be reduced but the profits may also be compromised. 

Hi Vladimir , first off, thank you for your help. I did indeed implement your edits and I went on to improve it further and I was now getting over 50% compounding returns with 13% DD, sharpie 2.3 for the same test period 2020 -6 -10  to 22-6-10.

Unfortunately the strategies is not usable because if I change the test period to 2010 6 10 - 22-6-10 , for both your edited version and my newest version that's when results get very ugly.

My version compounding returns is now 3.5%,  DD is 44% and, sharpie 0.27

Your edited version above gives , compounding -11.142%  ,  DD is 78% and sharpie -0.7

Manav Trivedi , thank you .  

Hi All

I just want to point out one more thing: if you change a hyperparameter very slightly and it produces a completely different result. The “good” backtest is very prone to overfitting and drawing conclusions in the wrong flow, rather than a hypothesis → testing → conclusion. I would say that would probably not transfer its success to future trading since no hypothesis was proven to be working.

Best
Louis