Binary Options: Your Success Rate?

I would likely be happy to answer questions, if you truly studied the picture above. Took much effort to get to this.

Take a small IS sample size, 7k bars, optimize an algo (with 5-6 conditions for it, like iRSI + iBands etc.) for it, do this x amount of times using a neural network with no pre-concieved trading ideas (everything is random), ensuring at least 64% or something. Then run all algo's generated with this % (we want to combine them to get more signals - but watch it; signal bars may double up and so % could drop!) through a first OOS sample of actual broker data, of let's say 100k bars (I only have 76k bars of specific broker data so that is what I use). A number will not hold up. Fine, drop them. Then run resulting algo's through 2 million bars orso. Again drop any that do not hold up (threashold 59% for example). Then merge all algo's and again filter our those that drag down the total %. Ensure total % somewhere 58-59%. 55.5% is needed for 1 minute 80% BO's.

$1/trade, ~2.5 months of data. $193 matches 30K bars example from drawing above

Same algo, with over 5 years of data. $1 per trade, 60.77%. Green bar is correct, red bar is wrong. Red arrow shows trades beign taken. All from algo's originally compiled on 30K bars. Similar results on less bars. See above. Ideal seems ~7k bars. 7200==one week of data. If the algo has a number of good combinations (try iRSI + iMFI + Close[i] compare Close[i+x] + iADX + iBands UPPER vs LOWER width + Barsize) forget about "overfitting" - let the NN to make as accurate a formula as possible, and many of them, at a certain %, with enough occurence, then combine them and finetune. Future thougt; per-hour or per-day optimization may make sense too.

These are all 1 minute trades. 10 Minute trades have higher %, but lower payout. I prefer 1 minute. Trading on bar close (second=59). Getting timing right (millisecond timer, 100 millisec per check) seems important. Also broker oddities, like my broker disables trading if less then 7 price changes happened on this candle - this needs to be included in formula (Volume[i]>=7) and they disable trading for several hours (&& TimeHour(Time[i])!=0 etc.) All this needs to be in one's algo, otherwise you cannot optimize correctly.

Hmm. Also need to incorporate draws better. i.e. % check and balance increase/decreas on gains/losses only.

Thank you for the image. I had a lot of joy recreating the exact same graph within JavaScript which I will use in the future to better rate my systems. :-).
Only one thing, can you tell me how you defined poor systems? This seems like an arbitrary statement.

Hi roel13..
Great series of recent posts... tracking your test frame thinking ..
and the use of programming skills to churn data to pinpoint probability.
IMHO Binaries lend them selves to this sort of analysis much more easily than Spot.
Nice job.
However, my ignorance of programming skills (depended for years on an exIBM partner.. sadly no longer with us),
has left me lost about some of your ideas..
Can U post a pdf or something in greater detail... logic-wise rather than program-wise?
for C++ challenged individuals like me.. ??

C