Since it's an obvious variation on T-stat, this automatically assumes some degree of trading frequency. I'd say for a strategy with occasional presence in the market, you should do it per trade instance (using the same T-stat logic). Also, if the strategy has a dominant direction (e.g. you keep buying SPX in a predominantly bull market), some form of dual-population test (like a student T-test) is smart to do to establish if you are just up-playing a selective bias.
Thanks nonlinear5, I did not account for slippage and commission. I now account for slippage and commission, and I get more recent trading data from 1/1/2017 to 11/1/2017. Please see attached. /GC instrument looks good and ready for some real money. It has low drawdown as well. My only worry is if 318 enough trades to make decision or should I test for more trades? But if I do this, drawdown will go up.
2017 has been pretty steady bull market. Take your data from 2007 all the way through to now. Compare your equity chart to SnP 500 chart and see if it provided a better risk/reward then a simple buy and hold. That’s a good first start to figuring out if it’s a strategy that you’d be willing to put your money in.
Thanks Metamega, That's a good comment by first comparing strategy to Vanguard S&P 500 Index Fund. I can compare profits of strategy to profits of index fund, beginning each with $10K capital.
That's GC (gold) and he's mostly going long, I don't see any bias here in that sense. 2017 alone doesn't mean anything though. It's so easy to come up with something that performs astronomically for 4-5 years but before that, completely fails. The profit factor is on the low side as well.
Really need much more years of testing than 11 months, am surprised it lost in crude oil as this seems to be tailored for intermittent running markets? Gold very tough to consistently scalp, it does have several months a system can do well then... Gold has slippage as I have had as much as 5 ticks in recent time, so real time you might find it no where near back testing.
Thank you Handle123, @d08 and @Metamega I now run back test from 2007 til Present time. Below are the results. This is an Intraday System, all trades close at end of market session and 1 contract. I believe this strategy with /GC is ready for real time and make the money now. 1. I do not understand when people say "the system will not last long". If that's the case why back test for over 10 years of data if no system will last long? 2. If starting with 1 contract and using /GC results as example, when do I scale up to 2 contracts?
Handle123 makes a great point as usual. To the OP, you MUST factor in if you were filled with a Market Order or a Limit Order in your backtest. I have had countless systems die and never make it to live trading after I factor in the effects of how I got filled. You can't just take the Close of a Bar and call that your fill. It will absolutely destroy most intraday systems. I am not saying that you cannot be a successful intraday trader but if you want to be an intraday system trader then you must factor this in. How do you factor it in?: To assume a Limit Order Fill you must see price exceed where you took profit by the width of the spread. For example for CL it would be price+1 tick (-1 tick for shorts). If you don't want to do that then subtract the bid/ask spread from each entry and exit price. For example, on you CL backtest you have 974 trades. You will need to subtract (974*10)=-$9740.(Number_Of_Market_Orders*(1 Tick Dollar Value*(Ask-Bid))) If that was 974 round turns then multiply that by 2. This is still an assumption but it is better than assuming you got a fill at the price of each bar close. People often wonder what makes intraday trading so hard. The main challenge is the artificial time stop that we place on ourselves. To be a daytrader it means that you will be flat by the end of the day. You can't truly "let winners run" because you will have to close them out by the end of the day. The costs (commissions & bid/ask spread) combined with this time-stop make it a rough game to play. When I look back at my old trading stats from when I traded Stocks with a prop firm I can see that most of my money came from ECN Rebates (unintentionally). I switched to Futures (rebates not possible, at least for retail) and what I can say without a doubt is that you must factor in the cost of Hitting Bids and Lifting Offers. If this has already been accounted for then congrats. Also congrats on advancing beyond pen & paper backtesting.
OK let's look at this your largest winner and largest loser are steady average loser does not fall far from largest loser,but average winner does differ substantially from largest winner you have 16 consecutive losses if they were same as largest loser would be 16 X 404.28 =$ 6468.5,but your max drawdown is twice that amount plus change with that your profit chart looks smooth which makes me believe the results were tweaked with imho and you would run out of margin to trade this system from your earlier post "That's a good comment by first comparing strategy to Vanguard S&P 500 Index Fund. I can compare profits of strategy to profits of index fund, beginning each with $10K capital." OK so you have 16 consecutive days of losses You state that you trade one contract per session,but you have total number of trades 3767 and with given number of trading days as i looked at 252(2018) that gives me on average 1.5 contracts a trading day,not 1 contract per day It may not be enough in real market conditions ,bad fills ,slippage etc.in general it looks near random system with some money mgmt keeping it just above loss(that part could be curve fitted here as your earlier posts indicate different results in profit chart) there are some things that i can not discuss that tell me straight away at first glance that there is no legit edge in this system
He also needs maxdd Peak to Valley otherwise the dd risk is curve fitted by definition. The (consecutive losses*largest loss) could be looked at as a worst case short cut of doing monte carlo analysis. The P2V DrawDown is based on the actual historic data.