This is my first post in the non-elite section of BMT in quite a long time, but I thought this may be helpful to some of you. . .
As my final database begins to take shape, I have increasingly large data pools to play with, and quite a few interesting/informative/valuable findings.
The following data was created by considering millions (yes) of trades, going both long and short, across a variety of instruments, without slippage or commissions, using completely random logic in both entry and exit, with average in-market times ranging randomly between 30minutes and 300minutes before exiting (exits also using completely random logic). The data set used was from Jan 1st 2009, to Jan 1st 2016.
Trades are considered based upon the entry time, not the exit time. . meaning, a trade initiated at 1:10 am and ending at 3:30 am has been counted as hour '1', not hour '3'.
A positive value indicates a 'long' bias, a negative value indicates a 'short' bias.
Enjoy. . if you have any questions, feel free to ask.
Please register on futures.io to view futures trading content such as post attachment(s), image(s), and screenshot(s).
That might be handy if you were only using 1hr Bar Increments, and only trading for precisely one hour/bar in every trade. . . but that's not how I intraday trade, so I wanted a much wider and more representative sample.
Also, I'm not so interested as to whether or not the trade closed up or down, as to the overall average EXTENT of this 'up' or 'down'.
Of course the data used to create this table all roughly coalesce around the mid-point of my avg-time-in-market extremes (30min, 300min), so one could use the midpoint of these two and run their own tests and get very similar data. . but it would still be different, as it leaves out the tail ends of the bell curve of time-in-market ranges.
The completely random entry logic is just to assure that the results are not biased towards (i.e. oriented around) any single type of happening, such as a moving avg crossover. . . I wanted to mix all 'happenings'. The results are almost precisely similar to what one would see if they were to enter hundreds of thousands of trades at a randomly chosen minute within the hour
I try to make all of my tests emulate the form/structure of my actual live trading, insofar as possible.
To be frank though, this wasn't a specific test I set out to perform to produce this data set, so much as data I needed for another task (normalizing 'profit factor', according to instrument/hourofday combo), that I thought might be useful to someone here . . it seems to show some interesting historical patterns.
Last edited by Dionysus; January 10th, 2016 at 07:53 PM.