Attached please find the system's performance. To get this result, parameters are optimized. However, look at the consistence, I am wondering is the system over optimized or am I doing something right. Please note that there are 429 trades in the period of 3 months and the market experienced ups and downs in the selected time frame. The total return without slippage/commission is about 30-35% in three months JS
Lets see... $20,000 profit, 429 trades, does that imply a gain of $50 per trade? Sounds ok to me. What size did you trade? Seems like it was quite large because there weren't too many $0.50 moves in that time frame
The results in the example are excellent but the chart showing the price action and trading signals appears to be extremely curve fitted - i.e. parameters uniquely optimized for the local sample period. The proof for this is to backtest with the same parameters for one or more totally dissimilar historical periods. If the system loses its shirt in these alternate sample periods then you know you're merely curve fitting and don't actually have a robust system. But if it makes comparable profits with the same parameters in dissimilar markets you'll know it's robust and able to survive in the real world.
You ask if the system is over optimized - though I may have opinions, I say that can't *really* be answered with the information you provide. It can be said that if the data that you show the results from is also the data used for optimization then your results are contaminated, period. Backtesting is limited in usefulness and real money should only be trusted to extensive forward testing. JB
Thank you for all the comments. I guess you missed the most important point : CONSISTENCE! It seems that this system, by selecting the proper parameters, can make money in up trend, down-tren and trading range. Trend-following system makes money in trending market but lose money when the market is flat. Swing system makes money when the price swings well but usually miss a trend. The consistence seems too good to be true. Anyway, this is the most consistant system (400+ trades! ) i have ever seen. Does anybody see similar consistence in other systems? JS
>Thank you for all the comments. I guess you >missed the most important point : CONSISTENCE! And we thinks you missed an even more important point...ours. >The consistence seems too good to be true. Without a MUCH larger data set and uncontaminated forward testing you don't have a clue as to it's consistency. Beware! JB
I was told this when I was first testing systems, so I will pass it on to you (although it has already been said here): You need more time to test on. 4 months is not enough. From personal experience, i had a system a few months back that worked well over last 4 months- then I got data for '03, and it still looked good (but declined a little in consistancy). Finally, I got data back 5 years, and this year and last was ok, but the rest were awful. THis generated that fear, of how will I know if market is similar to the way its been (last 4 months) or from periods that tested bad. I cannot predict how the market will act later today. The system must hold its own. End result is I cannot even tell you what that system did (i forgot about it and moved on). Get More data, test it, and good luck, I truly hope it holds up. If it does, that looks really good.