This might be the most intelligent comment I've read all year: "Trading systems are evaluated not based on how they were developed" Congratulations - you've won the first class idiot award!!!! Frankly, I'm glad there are people like you participating in the market. Your ego and belief systems are what cause you to write (and likely trade) uninformed and ignorant responses. You don't know if a system works untill you try it with real money you friggin' idiot. I don't know about you, but I don't apply real money unless I'm nearly 100% sure that the concept I am using has a sound fundamental basis. And it takes more than a "few trades" to verify any system. Frankly, you sound kind of dense... you missed my point entirely and picked up on my choice of terminology. Do you even know what a valid market hypothesis might be? Mike P.S. This ain't some silly consumer credit modeling exercise.
OK, if you say so. In my naivety I thought that a system with high success probability required many samples to be confirmed but only a few to be invalidated. I think I'll have to take it up with my old statistics professor. All this hard work for nothing... /Hugin
fellas i think you guys are arguing if the chicken or egg came first. In my original post, my intention was to find out a very specific case of modifying the stop loss target logic that produced much better result on a 5 year historical data set. I think i named this thread's subject poorly, but nonetheless i got the answers i wanted - that this is acceptable behavior in tuning an automated system. Whether the actual definition is "curve fitting" or not has no relevance to me. ps you guys sound just like the quants in my firm, they would argue those things during lunch break all day long.
You're right. In general, there are so many factors involved including a fair amount of subjectivity. When dealing with trading models, many of which are user concieved ideas about what markets function like, the basis for validation becomes subjective. Personally, I like to see at least 1000 backtest trades and 100 forward test trades. A grain of salt about forward testing: you need to know what market movement you're going after and whether or not that type of movement occurs with enough frequency to validate your model going foward.
You are probably a big time loser. The whole sentence by the way was: "Trading systems are evaluated not based on how they were developed but based on how they perform." You probably have lost everything looking for "constructive optimization" techniques (royal lol) to develop the sure thing. You will not get it because you don't have the knowledge of what it takes to get it based on the crap you spew contantly. Your ad hominen attacks will not make you money in the markets. If you are older than 30, there is no time for you to learn basic concepts to build the solid foundation that is required. If you don't have a math background don't even hope you will ever make money in the markets. Here is a link for you: http://en.wikipedia.org/wiki/Autism
Forget about back and forward test. Many good ideas cannot even be backtested. If you need 100 actual trades to determine whether a system is not good it is too late. Wall Street geek developers lost their jobs are are now delivering pizza because they were thinking along these stupid lines. There are way to determine with high probability whether a trading system can maintain its success rate in the future from a limited sample of trades. I am not going to teach you the techniques, you should find out and learn them yourself. They are at the core of success.
Hey - you started this BS by attacking my thought process via childish remarks. So far, your comments are either misguided or irrelevant. Simply put, your assumptions smell like you pulled em' out your ass pal. You clearly stated that trading systems are not evaluated by how they are developed but how they perform. Well, you're wrong, its both -- they are both equally important. It not NOT A ONLY B, its BOTH A AND B. Its amazing you attack my credentials simply because you childishly disagree with me. Then you go on to attack my trading which you know nothing about. People like you need STFU. There are plenty of people who read these forums who know exactly who I am and can vouch for my track record over the last 5 years. My education (in math and other things) is *very* advanced. I can prove my track record in a matter of seconds and I can do the same with my education. Can you? I doubt it. You're starting to *fit* the classical definiton of a troll, pun intended. Mike
Peace... I don't understand what the fighting is all about. You didn't get that answer from me though. If your stop tuning affects the entry points of your system then it is curve fitting. If curve fitting is involved that way, then it is highly probable the system will fail, sooner than later. This is my answer.
The point of forums is to help others. If you have no suggestions there's no reason to post. Let's try to follow up on measuring curve fitting and expand on your ideas of measuring curve fitting based on a limited sample. I assume you use GARCH. And if so, what variant do you recommend and why? What's your opinion of the claims that GARCH isn't a good predictor for volatility in returns?