shortie, Nice, our holding times match in each other's models, but I'm going to have to re-evalutate my model. It appears that holding until a new signal is generated is key. As for the holding time arguement, it would intuitively seem that holding would not increase the edge as the second (third/fourth) day(s) could either be with or against the postion and COULD be a wash in the end. That appears not to be the case...so I'll have to update my model and re-evaluate. To do: 1) have Mike test C2>C1 just to confirm it should be ~+80k 2) update my model to include holding until next signal Nice work. masterjaz
Shortie: good work! Some questions: 1. What platform are you using? Is it in excel.?If yes, could I ask for a copy. 2. What is the success rate of the trades?
My understanding was that only one action is made. If you switch to one action only what would happen to hold time, and success rate?
Master: If you can share your excel, I would be interested in a copy. If you cannot share, I would understand. Excellent work!
To all: If you can generate a strategy that leads to trade outcome that form a normal distribution (holding time should be constant), I think I have the knowledge to turn it to a profitable strategy.
Current version is 12MB, so I'll have to find a host for the file. Be a few minutes... or seconds: http://www.filedropper.com/spyrand Please note I have no idea about this file sharing site. Be sure to anti-virus the file before opening.
my script is in Wealth Lab. win rate is 53.62% and 53.22% (without and with random, respectively) i will run the other test later
TJ, having had some fun with distributions over the last few weeks, a question or two. Do you mean the per trade pnl should be roughly normally distributed? (I would personally think the %pnl would be a better variable) or what statistic should be normally distributed? Having played with data from open to close, I can definitely say the total distribution is NOT normally distributed. Kurtosis is the main issue with a slight skew... I am beginning to see what you and Mike are thinking with regards to distributions upon distributions... masterjaz
Once again, just to clarify, the idea here is that the outcomes of some of the proposed problems that people are modeling here have certain a distribution, using this information there is a way of improving the given system. Is my understanding correct?
Craig, That's where my current understanding is as well. As I read Mike's statement, he is superimposing two distributions 1) the close-close distribution (what ever distribution it follows) and 2) the uniform random distribution (which is not normal!). Now, I am trying to visual what the super-position looks like, but need to get to my real job! More to come, masterjaz