What I've done is just go to the TS analysis screen. Write down the average winning trade + 1.28 * the winning std. deviation. Then write down the average losing trade - 1.28 * the std. deviation. This gives you the 80% threshold for the largest winning trades and largest losing trades. Use the numbers in the spreadsheet to test for larger size trades. If found put the trade result in a column. In the next column over test for whether the trade is present. If so, put in the next trade result after the current trade. Do this for all the trades. Once for the winning trades. Another time for the losing trades. Then use the corellation function to check the two columns (large size trade and next trade).
I don't have enough info. to tell if it's curve fit. You can test it yourself by making the parameters you use inputs to the model. Then optimize on one parameter at a time. After each pass, check the optimization report and see how sensitive the results are as the parameter changes. The ones that are most sensitive are where you may potentially have a curve fit issue. Once you're done with the model it's always good to just let it sit in the market you're planning to trade and watch it for months before trading it. I do that with everything I trade just to make sure it's not curve fit.
I've only talked about the original model I started with in 2002. I have multiple systems and have a couple of others that I'm watching before going live. I'm also working on models to take live in 2008.The one I'm most excited about is a swing trading model for the ER2 market. I've been watching it since last August and made a few changes. If it continues to make money every month (no losing months so far) then I plan to go live with it starting Oct. 1.
Sorry I didn't respond to your pm. Tampa PM'd me into the original thread and warned me most people were following to try and copy my system. I'm still paranoid about giving away too much info on it. You've culled out some of the criteria that I use. Here's the selection criteria for taking trades the next day using daily data: If rg = 1 then trade tomorrow. if doji(25) and range < range[1] then rg = 1; if doji(20) and range = lowest(range,7) then rg = 1; if doji(20) and (high < high[1] and low > low[1]) then rg = 1; if range = lowest(range,4) and (high < high[1] and low > low[1]) then rg=1; if (avgTruerange(2) = lowest(avgtruerange(2),20)) then rg = 1; Also, if oday is not = 1 and a doji[15] exists, then trade tomorrow. if high > high[1] and low < low[1] then ODAY = 1; There's some additional next day criteria for only taking one way trades but I don't plan on giving it away. Also, there is plenty of other "stuff" on the moving envelope that I didn't post and don't plan on doing so that is mainly aimed at avoiding rangebound trading days. I don't use ATR for volatility. I use the average range (high - low), since it's a daytrading system. ATR would add in the range outside the day session. I hope you find some of this useful, but don't plan on me giving the system away in it's complete form while I'm trading it. 43yotrader
Its always been my theroy that the more people that use the same stratagy the more profitable it will be.
43yotrader, What is the process you use to find which days to trade? Obviously the goal is to find days that will suit your strategy, but how do you find those days? I am finding that this is a herculean task. There are almost infinite numbers of things/parameters to test. For example, how did you find that the day below is a day you want to trade? Did you just keep trying different things that seemd to make sense? "if doji(25) and range < range[1] then rg=1;"
Taleb would say it was luck or survivorship bias. I had bad luck with TA so I tried other ways to decide what days to trade. Since my trading timeframe is part of one day, I used daily bars to search for patterns (higher timeframe). It was mostly trial and error. When I found something profitable I just played with it to improve it as much as I could. I've recently picked up all the tick data for the SP back to 1983 and tested the basic strategy from then until now. No losing years! I guess I'm just very lucky.
would you say that picking the right days contributes the most to a system's profitability? or would you say the specifics of the entry and exit rules are more important? it seems to me that being able to find the right days is more important. i am thinking that if you already know beforehand that the day is going to trend, then you have already made it a lot easier on yourself. any trending related entry and exit rule should do well. in my testing it's a constant battle between finding day filters and entry/exit rules. i can't seem to determine if the problem is with the day filter or with the entry/exit rules.