I appreciate your posts, comments and counsel. Hope you hang around so some of us can learn from you. Regards,
a VERY quick primer: Normal/Gauss: symmetric, two parameters logNormal: BIG positive (right-hand) leg, so that $0 can close the other (left-hand) side Pareto: in shape, the opposite of logNormal Weibull: "Anyway you want it, that's the way you need it....!" 4 parameters... and with those parameters, you can get pretty close to a perfectly symmetrical Normal, to a skewed logNormal, to a skewed-the-other-way Pareto, all to your (or your underlying's) taste.....
Hey, Not offended. I just think you missed the point of what I was saying. There is a difference between building a trading model with underlying assumptions on price distribution and checking to see if two distributions are different. I know a lot gets made of the fact that return distributions aren't log normal. I get why. But for some things using the assumption that they're log normal simplifies the math and is good enough. That being said, if someone is worried about it there are ways to test that don't require a normal underlying distribution and they don't take any extra time to run if you're using most stats libraries. -J
I like to run a system on each stock individually, (same patameters) rather than 1000 stocks and see how the equity curve of each stock over time compares to the stock itself over time. Filter the good and bad in general. Random seems to work in trend. I can prove this to myself by selecting each bar for entry and testing a wide range of time stops relative to entry bar with a wide price stop loss. Everything seems to work ok until applied counter trend. This is Geico. Everyone knows that! I do wonder that discretionary market selection is underrated. Seems to me to be a real skill.
Interesting article Truth_. What is amazing is random trading produced the least volatility meaning higher risk adjusted returns? Of course all methods produced positive outcomes because of the upward bias of the indices but random trading producing the least volatile returns! I have to do some backtest myself to believe it.
So far I am using machine learning to test hundreds or thousands or feature combinations which can result in tens of thousands of different strategies. I do this on just a few tickers. From there I retest the strategies I am interested in based on performance metrics on a greater universe of stocks. This is all automated. I might start with 18,000 strategies and end up with about 20 or 30 that look interesting. This whole process takes about 15 seconds. Now I am at the stage where I want to quantify the merits of the remaining strategies. Good stuff in the thread gents. I will post what I end up implementing. fan27
You must be a programming wizard! How are you able to test so many different strategies and combinations? What programming/coding are you using? I use Excel/VBA and it takes a lot of effort for me to test a strategy each time. My current homework assignment is testing the random entry strategy to see if it indeed beats TA for the equities I am trading. Regards,