It seems it relates to the issue of signal to noise ratio. Does the link previously referred provide any clear guildlines/solutions to improve the issue? http://www.elitetrader.com/vb/showt...&threadid=21739
yeah, totally, welcome man. nice lucky start in 87. et rules, and we can all learn from each other. i'm sure you have much to share. welcome. what was that first trade worth if i can ask?
I have the trading slips for the trade. When I come across them, I will scan them and post them here.
I would not go that deep into it, I would generate signals on 5 min and go into lower timeframe to pinpoint it, so noise would not be that relevant . Walter
My first post. Been lurking for years, but Acrary has lured me into the open with his thought-provoking posts -- and I have a question. If money flow data from TrimTabs is part of the Prod5 model's edge, does that mean you have not found liquidity to provide a useful edge? I am just curious as Biderman states that liquidity is the best leading indicator. Also, if I wanted to explore this edge myself, would it be feasible to use money flow data from AMG? They have a weekly money flow data service for $250. TrimTabs, as you know, is way expensive. I'm just a guy turning over rocks looking for things that work.
Welcome, FNG's. Here's some questions relating to backtesting. Hope I'm not straying too far off topic. I would say the number one rule of back testing, is don't look ahead. A simple, stupid example of this would be entering a trade based on something that happens the day after the trade. OK that's obviously a stupid case of foreknowledge. But foreknowledge is not always so obvious. For example, if I only backtest stocks that have a price over 10, is that a kind of foreknowledge? I already know, and in some way, perhaps my system "knows" that the price will return to 10 eventually, no matter how low it gets during the testing period. So any time it's below 10, it's a buy. But of course, that won't work with a real trade, because there's no guaruntee that it will ever return to 10. How serious of a problem do you folks think this is? Is it best to find systems that work on a large universe of stocks, randomly selected, omitting your usual screening criteria? Does every kind of screening critera = foreknowledge? Any other examples of foreknowledge that can be easily missed?
Sometimes you don't need to actually peek into future bars. Just something like: if Open < close buy on open exit on close can be coded easily, looks profitable, but cannot be traded because the close isn't know until after the open. Same example applies to many stoch() cross systems, that use bar high/low. Obviously the answer is to enter on the close or the open of the next bar, but that's easily overlooked. Then the system also is no longer "profitable". :eek:
What methods have you guys used to successfully minimize the number of whipsaws your systems give? By whipsaws I'm referring to the state where a system gets in just as the market reverses (worst possible entry), and the system then gets out with a loss (and might even reverse at this point). Then this pattern repeats itself for the "whipsaw" effect. I started discussing this subject at the beginning of the thread, but would like to hear your insights on this subject.