I'll add to this. It's not enough to back test a system in Tradestation. That only reveals the system's metrics, but not it's character. You have to go through the backtest trade by trade and total up the P&L as you go along, taking care to observe the changing market conditions in which the system is performing.This will give you valuable insight about the nature of the system - where it draws its P from and where it loses its L to. You can do it easily in Tradestation too where you print out the charts with the system buy and sell orders on overlay and "backtest" it visually. Like the law of unintended consequences, systems can have hidden vulnerabilities and hidden strengths that can be quite different from what you initially thought. eg, someone trading a small portfolio of merger arb names might think he is market neutral because he is so small, but his P&L has a hidden exposure to a bear market because in a bear market deal flow dries up relative to the number of arb players, crushing merger spreads and therefore profitability for even the smallest players. eg, someone may think that he is profitably trading the closure of down gaps. But in reality he was merely buying dips in a bull market. When the bull market went away, the down gaps sometimes never closed, destroying all profits.
As it happens, I had originally rejected doing it this way because I *thought* that it wouldn't be very accurate. Of course I never tested this assumption out, but just rejected it out of hand. As a result of your comments, I will take a new look at this approach to see what merit it has to me in my backtesting. It will certainly simplify things for me, which is what I was hoping to do anyway. And just for reference, I don't think that the way I have been doing this is necessarily any more accurate or that easy so change might be good.
IMHO you don't need to fade a good system. A system is only as good in realtime trading as the individual who programmed it! If the programmer is not a successful trader then NO SYSTEM he programs will be successful. END OF DISCUSSION!
Contraian, you are correct in that sense (although I do not agree with your signature). A programmer alone is not going to create a good system - simply for the fact that it is not his bag, baby! He is a programmer, not a market guy. This is common sense. But a trader who uses the programming as an extension to his thoughts - this is a powerful idea, and, being in trading business for quite some time, this is becoming the trend. More and more traders are wanting to automate stragedies they have been using for years, with high sucesss rates. Just look at consistant growth of tradestation company and similar. Seeing my "friends" do this, and make fortunes at it, I had to try it out myself. So far, it has turned into a very rewarding experience - the programming part I really like, also. Also, you mention "Fading a good system". My original question was fading, or being contrary to a losing system.
I apologize in that I missed understood your original question. However, how do you define a "losing system"? what were the original parameters of the system? what type of system are we talking about? is the trader that is trading the system also the programmer? One must first answer the above in order to answer that question.
The original thought came to mind when I was creating simple systems with a program I built. The original post went like this: "But, there are some systems, which look absolutely horrible. They produce an equity curve that would wipe Bill Gates out in a month or two. So, you all know where I am going (if you read the header of the thread) why not reverse the signals of this system, and trade it from the other way?" Now, what i mean by systems, are just areas on a chart where I believe there is price action that forms a pattern, a pattern that is repeated constantly. Now, this of course is the basis for TA - price patterns which repeat themselves. I am not talking about head and shoulders, or wedges or the like. Charts are prices, prices are charts, just visually different. I have theories on certain price actions. About a year and a half ago, while trading, it just occurred to me that, I was trading patterns (again, not text book stuff, my own perceptions - mostly just price action), why not see if they work over long term? This led me to system development. So, recently, the question arose, if the damn thing looks really awful, which happened to a few of my ideas, why not take the other side? If I sold where that thing bought, then it would have yeilded a profit. This of course, is all assuming past data repeats itself, which I will never answer.
Let's consider that your system has a buy signal when bid/ask is 1000/1000.25 : You buy at 1000.25 Let's say that you sell with a profit when the bid/ask is 1005.25/1005.50 : You get a 5 pt profit. If you had taken the opposite trade, you would have shorted at 1000 and cover at 1000.50. That's a 5.5 pt loss. You've got a 0.5 pt difference between your system and its opposte. That is a problem due to backtesting. You've only got T&S data and in no case you can consider that you would have been filled at the same price. To be exact, you would have to have an history of bid/ask.
I am using backtesting as a lead in to what I refer to as my "forward testing" period. Currently, I back test, and yet, using bar data (1 min. OHLC) is skewing my results to a degree I cannot understand now, as you state. If the system looks good, then I plan to "foward test" it. I insist on quotations, because I get impression my definition of forward test is different from other peoples definition. The plan is if a system looks decent, and consistant, then, simulate trade it on real time data for 6 month. This will insure that that as bid / ask and size come in (real time) i can place trades (simulated) and at the end, come up with an idea of how I did (6 months). Sure, if the thing is working great in under 6 months, i may jump in. ( I will also devise way to store the bid / ask data). Oh, if systems I am evaluating (my ideas about market) suck, then all this work will never happen! Now, this is all under the assumption that I cannot obtain bid / ask data over 6 months or more period. Anyone know if this is availble on EMINI? Also, while I am at it, anyone know where I can get minute data for 1 year or more?