optimal f sizing test on executed trades.

K

kenten

i am running a few trading systems automatically. in fact more than a few , 40+ systems running with their allocated capitals.
the sizing is determined by the optimal f i had run on the backtest results some time ago . all systems use the same ratio on their respective capital . Naturally i added and removed some systems, some systems shat the bed (capital dropped below the minimum level)etc.
today i had an idea while rereading ralph vince book. i run an optimal f test for the executed trade results , all the systems combined.
the result shows %20 percent is optimal while i was using %70. (on respective capital allocated to the system, divide that with 40 to get overall risk)
new result shows if i use %70 , i should blowup. according to the test i have blownup already. But in reality for 2017 18 acc shows a somewhat healty profit.
i think i am doing some logical mistake somewhere

do you have some idea
 
They have run multiple studies on how much risk you should take on each individual trade. Most have come to the conclusion that it is 2% risk per trade to avoid risk of ruin or blowing up your capital. There is also, a suggestion that if you are a newbie getting into the stockmarket, to limit your risk per trade to 1% per trade. How much are you risking per trade? You never mentioned that. I think that would determine the chances of you blowing up your account. I am only guessing you are sticking to 2% risk per trade when you size your positions. Also, you might have a trading edge. What is the size of your gains vs losses in dollar amounts on average?
 
i dont use 1-2 percent heuristics . i am trying to find the logical mistake i am probably doing.
the other aspects of the systems are not important
 
I'm just guessing here, but could it be that correlation between your various trading systems influence your actual and calculation results?
 
It is not entirely clear from your post what your 70% allocation actually means with regard to your live trading. Can you be more specific and give an example.
 
lets say i have 100k capital and 10 systems running , so every system starts with 10k and risks %70 of 10k.
you may say every trade risk %7 of overall capital ,(at least at the beginning).
these systems run for some time (overall profitably)then i decide to see what was the actual "optimal f" . when i get all trades like they are trades of a single system , and find the optimal f , it shows a lower f level.
when i do that by individual systems they are mostly around optimal f , but of course some systems made loss, some went below ruin level. so their optimal f is close to 0. i think combining these is not mathamatically sound idea.
everytime i read ralph vince's books i learn something new but they are boring as f . it increases my sleep hours
 
It sounds to me as if you are adding up the results from each system as if they were statistically correlated to each other. If all systems perform well at the same point in time, or perform not well at the same point in time.
If the systems are not correlated (or their correlation is low) then this would not be realistic. Poor performance of one system could then be offset by good performance of some other systems.
 
when i see the results of these seemingly not so correlated systems , there are times i would not predict correlation collapsing, it does dramatically. there 15-20 days intervals ,correlation goes upwards with no apparent reason. there are times at macro you can say trending , no profit from trend following and mean reverting. then comes a time with apparent s/r levels. all make money.
i say using correlation on any kind of risk management is doomed.
i think this kind of times when correlation of returns moves simultaneously makes standard deviation of results greater so the lower f .
i am now thinking about whether this is a real effect which must be taken into account or just a calculation mistake.
 
i think i solved the problem.
every system has a final result which must be added to every other one .
when combining all together i was multiplying each system ,

i calculated that way, it works as it should now
 
You must log in or register to reply here.
Back
Top