Hey all, I'm new to automated trading systems. I finally decided to take my discretionary trading style and apply it to a mechanical system. It isn't based on any oscillator or lagging indicator.. it is purely price action based. That said, I've heard a lot about curve fitting. Currently, I've optimized it using two inputs. Does this equity curve look curve fitted? Can one tell by looking at the equity line? What are some tell tale signs of curve fitting? Also, how does one forward test in Tradestation? Cheers, Amed
From your equity curve, I see that your average profit per trade is about 1 NQ point per trade. What were your transaction costs assumptions (in terms of bid/ask spread, slippage, and commissions)?
Just test your strategy using those exact variable values on out of sample data. What happens if you adjust those variables? If tweaking them just a bit causes your equity curve to change a lot, that's not a good sign.
Nonlinear, Re: average one point per trade.. how did you surmise this? I put in $4.60 per roundtrip as commission, did not account for slippage. As it stands right now, I enter through a market order on the very next bar.
When running the optimization.. its always in the green. The curve doesn't change much, but the execution costs go up. I'm currently working with the values that gave me the best net profit and the lowest amount of trades. There were other optimizations that made double the networth, but double the commish.. Anyhow, hrm.. I just don't know much about this. I know I'm using a "kids toy" when it comes to automation, but do you guys have any experience forward testing with Tradestation? (on a side note, the current bloodbath of MLK day has me on the right side of the market.. haven't even scaled out 1 position yet ) =)
The curve shows about 2500 trades and about $53,000 net P&L. That means average profit per trade is $21.20, which is about 1 NQ point. I don't know how TradeStation accounts for bid/ask spread and slippage, but if it doesn't, your realistic profit per trade would probably be about 0.5 NQ points per trade, because you enter and exit on market orders. I am assuming here that you are trading 1 contract for this simulation. What is the "Profit Factor" figure that TradeStation reports for this curve? It would be more informative if you post a full report.
Forward testing just means testing it on data that isn't included in your optimization. So if you had tested this on 2007, try testing it on 2006 using the same variables. I'm not familiar with the forward testing option in TS, but you can just test on another time period using the same method you're using now. The results you have now are actually meaningless until its been tested on unseen data.