so i am using ai to perfect my trading systems and explore trading ideas that i didn't even have or would have even thought possible. here is an interesting study i been working on its' a 10 point range bar so there are 40 ticks within each bar. i am tracking the time it takes to make a 10 point bar appears at B: also tracking how many contracts per second being traded appears at V: i divide the range of the bar into 5 segments above and below the open and then i assign a pattern to it appears as 1:xx, 2:xx etc. the bottom panel detrends the the market and centers everything around the open or the bar. yellow bars are choppy market and blue and orange bars are trend start. i also have the vwap, elapse time of bar formation, total volume, current range then using grok3 ai i run the patterns along with all that data into a optimization that will look for combinations of patterns that will make money. it's crazy what you can find when you detail. i send all this data out to a .csv file and then load up to ai and it will run analysis based on any criteria i wish. the limitation is your capability to open up from previously constrained preconceived limitations. you can really pretty much get it to do anything you want. you can tell it what you expect the results to look like and instead of you looking for it thru optimization processes it will instead find combinations that meet your specifications or not. increasing the detailed data you feed it will better your chances of something working out. good data in great results out. Impact: 55,986 sequences and 184.27 billion combinations enable profitable setups in any market. so what does this mean? i have defined the market i trade so well that ai was able to process that data and return to me more profitable combinations that i ever could with a old cray t-90 back in the day and do it far quicker than any trading software that has ever been written. ai will number crunch on a whole different level, you either owned a cray or you rented or bootlegged time from an astrophysics department somewhere. i have found grok3 to be far superior "based in texas" to anything to date. also ai has allowed me to find a way to actually auto trade using pseudo code to hook into my existing trading platform and achieve what was impossible before as far as order placement. would love to answer questions as it makes me sharper and may benefit others thinking about ai and how to use it for trading systems.
Backtesting is curve fitting. So do you wanna say more curve fitting you get better results in the past or better realtime tracking results. Because I would like to see realtime results after all those optimizations are done. Usually this is the hardest part and not doing more optimizations in larger input combo space. I mean I stopped looking for better curve fitting or doing more optimzations and even other strategies after I have achieved Sharpe 14+ in past data. Because it does not make any sense to look further. But in realtime I never got beyond Sharpe 2. Only on some lucky periods I got a realtime Sharpe 3 for 1 year. But that is all. So the key is to have high Sharpe on realtime tracking or when you trade actually after all that kind of looking on calculating on past data is done. There is a huge discrepancy on past data and future data. What kind of results with Grok3 did you get in realtime tracking so far ? What can you show here ?
So what's the bottom line...did you invent a Holy ATM money machine that will make you ten million a year All that scientific technical talk is great, but did you develop a nuclear bomb of trading cash money returns
i have backtested since the 80's but this is different it's like data mining. optimization will most the time give false positive results because it's looking for best case fit. it will allow trades to be valid that in real life trade cannot be obtained. probably we have all been duped by test results from optimization. why this is so different is because it is just taking the data you give it and sorting it out in far more combinations that you could in a lifetime. it just gives you back what you give it but organized in a way that you can make sense of it. added: so i am very happy with the results but what i am blown away by is the ability to general fit the results to any market. i mainly specialized in bonds and futures but now i can draw out the data from any market and universally fit patterns and find satisfactory results. my possible pool of markets has opened up. it's important to stress ai will not optimize unless you ask it rather best just to sort through the data and find what is there that meets your criteria you specify. btw sharpe 2 is a holy grail to most i made a living below 1
i always thought i had the holy grail already which to me is replicable way to make money. ai is just helping improve what i already have in ways i could have never imagined. i am not a salesman for ai. i am just posting this because someone may not be taking advantage of a tool that they may not understand how to squeeze for results that would benefit them. first step to using ai is to provide it with the raw data it needs. most will take .csv or .txt files
I've done roughly similar things with genetic programming and genetic optimization of parameters in manually-written rules. It's much easier to get good results from training data than on out-of-sample data.
glad you mentioned that - i am not a fan of out of sample at all - that's why i went to range bars to eliminate the need for out of sample. range bars neatly compact the price into a bar of range no matter the time. so every bar looks the same and is mathematically uniform - whereas time-based data is all over the place. this coupled with my theory that price is all that matters nothing else, time to me is a pollutant to consistent math analysis of price action. things start making sense when you eliminate time. i think time-based bars are a cruel joke perpetrated by the industry "exchanges". so the only way to battle that is to throw time out.
Price data is "all over the place" too (at least for price swing ratios on daily data) as in https://www.elitetrader.com/et/threads/fibs-dont-work.367143/page-6#post-5600830 and https://www.elitetrader.com/et/threads/fibs-dont-work.367143/page-7#post-5600932
interesting - i use to dismiss volume attached to price and so recently since i started including it with the data for feeding to ai. i attach volume per second and time it takes for the bar to form "bar duration". it's like putting time back in to track acceleration and how it relates to price movement and direction. i never had unlimited power to run analysis like this, let alone it will write the code to process. i loaded up every python library i could on my puter and now ai will write and execute the code and format the results into any format i wish. all i have to do is use my headset and prompt it. this is an amazing time for anyone because i can see the day when software products as we know maybe extinct. ai will be the software and it's better than dedicated purpose software that i have paid thousands to license...