ok so kind of a plot twist. i might be getting staked 50k by a chicago prop shop and running this in a semi automated context for them. salient details... 80/20 split on profits (my favor), keeping the current application layer intact, no quota, can scale up to 150k backing. i retain ownership of all code.

will be finalizing details in the next 2-4 weeks but some papers have been signed and they come to the table with a solid track record & reputation.

ive been essentially papertrading using the alg and so far im not at levels where i feel comfortable going live. two issues are rat fucking me atm:

- i havent moved all the resources to the new server and i feel like im actually getting legit hamstrung by latency issues.. like for example im relying on httpd to retrieve price information on ticks. thats ... not great. but once everything is just a qemu vm on a flat vlan, i can start getting slick with samba/afs/nfs/smb/whatever which should cut down on latency by 90%, at least, and also make it easier to implement proper database logging.

- in the morning, with high volume, my alg is pretty efficient with a pretty high win rate. after about 1pm things start to get complicated because things get a lot more choppy as the volume pressure gets less intense. and at night every single fucking position turns into a random walk. so theres that.

bonus issue as my code gets more involved, the bugs get more insanely shitty. lately ive been seeing more situations where it enters a position and then immediately dies quietly without exiting the position. so ill get a cup of coffee and come back and suddenly its basically down a grand and dropping but the alg doesnt even know its in a position. so its absolutely fucking imperative that i write some sort of local logic into ninjatrader to automatically exit a position thats down more than X points. and im fucking dreading it.

honestly i fucking hate hacking my way through these issues because theres no real direct access to ninjatrader via api, so i have to rely on all these wack ass abstractions and just charge it to the game when i end up getting stomped on by slippage in one form or another.

also to be clear ninjatrader's idea of an API is a fucking DLL so dont @ me

at this point im considering offloading a fucking ton of hamfisted python to the ninjatrader ATM system which on the one hand means it will actually do what i want but on the other hand it introduces an ungodly race condition into the equation where the logic center cant view what is actually happening within ninjatrader, so i end up with the very real possibility of multiple orders stacking up at once, potentially cancelling each other out or just generally ratfucking my risk profile..

also looking forward to mondays is a lifestyle thats taking some getting used to.

so an issue im having is that im frantically tuning behavior to adjust to a market personality and right around the time i get it dialed in, everything changes up because of course it does.

ergo turning this loose with a prop account is manifestly suicidal.

the biggest issue i have by far is the lack of data access.

1) i dont actually have a level 2 feed, im stuck with OHLC.

2) when i dial into the super granular OHLC data it only updates when the price updates. and because my indicator code assumes every slice of data represents the same chunk of time, it registers momentum changes inaccurately.

so i have an idea to essentially pre-process all the OHLC data to backfill gaps in so many words but HFC + preprocessing = anathema due to the latency concerns.

on the plus side its doable and its possible that the latency concerns are non-biased, meaning they work in my favor as often as they dont.

but if im going to lean into that school of hyper granular HFT trading, doing so without level 2 data is like boxing with one arm and no legs; i have to offset the horrific transaction fees for every bet.

there is another option that ive been avoiding because it felt a bit ambitious and i didnt have the base functionality in place to support it; offloading momentum change detection to a ML layer and removing the static indicator detection code entirely.......... ... .

Sonatine, have you ever looked at applying any of this witchcraft to sports betting?

honestly i constantly look towards leveraging AI/ML towards sports and i'm pretty much locked and loaded on getting a rtx 3090 so i can work with larger table spaces / faster.

but right now, virtually none of this project is AI/ML for lack of good, consistent, sanitized, low latency data.

something i did discover last year when i was making tentative steps into sports betting using AI/ML, is that more often than not i ended up with nothing to work with but the existing lines.

like id crunch all the numbers overnight, wake up the next day all hot and bothered and see the saints won 66% of simulated games and then i go to 5dimes or whatever and thats literally the odds they are laying; there wasnt any real edge because they are essentially using the same algs / data to set the lines.

im much more interested in fantasy sports betting via ML but again i question where my edge is. we all know whats going on there; the same 2% of betters dominate 20% of the top payouts by using exel algs optimized over years and years and spamming hundreds of cards.. they might take down mid five figures in winnings in any given contest but after 20% rake and entry feels their PnL is break even or worse.

https://www.theguardian.com/science/...a-breakthrough

found a way to solve very complex stochastic systems which almost always end in a chaotic system

one of his colleagues said that "aliens must have given him the solution telepathically" because its very difficult to find solutions for this

maybe it will revolutionize algorithmic trading

its 180 pages long

he won the 3 million $ breakthrough price https://breakthroughprize.org/ because of his work

he didnt even expect to win it

god i wish i could digest stuff like this on its own terms.

so quick update on this..

ive been tweaking this for the last 4 months and have made a few significant changes.

originally i was relying on pretty one dimensional scalp-oriented signal generation for entry/exit and a complicated automated order strategy involving trailing stops etc. this worked great in very specific market conditions but generally resulted in the signal generation being stale in relation to actual market momentum, ergo ipso me getting double-fucked by stop-loss and contract fees.

tonight its going live again on a paper account with a more sophisticated indicator routine. this routine will largely determine entry/exit signals; ive removed most of the dynamic logic from the orders on the client and will be relying on a simple stop loss as well as a 'take profit' instruction which will insulate me against catastrophic reversals eg bad unemployment numbers dropping and slamming me out of a profitable swing before indicators can register the reversal.

first 24 hours are in the bag: - $1163

was trading 3 ES_F emini contracts.. started trading at 6:00pm, was down $1500 by 8pm or so after a much too conservative stop loss kicked in and then another swing just broke the other direction. set a far more lax stop-loss (questionable) and by 2am was up about $1200. woke up at +$1000. got absolutely ransacked on a couple of reversals to the downside and closed out the day at -$1163.

to put that in perspective, a $1 shift in SPY is about a 10 point / 40 tick shift in ES. at 3 contracts, 10 points = $1650. so the alg entered a total of 9 positions and lost the equivalent of a 70 cent shift in SPY. im not mad at that.

did some backtesting on NQ with a indicator alg that shifted signal authority weight to shorter horizons and i liked what i saw. burnt it into the platform and pointed it at NQ, next 24 hours commences at 6:01pm.

so i hit that 5am NQ drawdown last night for +$17,516. not a typo.

got chewed up by 6am chop and ended up holding steady at around +13k but decided to switch back to ES. the reversals on NQ are just too violent and too fast to implement any sort of automated response that doesnt either expose me to unacceptable risk or bottom out too many times before we stabilize and move into profit.

as discussed, im looking to work with backers and they have a 3k per day drawdown limit. at its current evolutionary stage i cant fade that consistently on this alg, not in this market.

so i switched back to ES, kept the more sensitive trigger / weight system, and modified it so it builds positions over time instead of just leaping dick first into every trend with a max-buy.

so right now im heavily insulated against black swan events but also positioned to catch those massive swings.

day 2: +$12,850.00

total: +$11,461.9

got my teeth kicked in slowly to the tune of $1800'ish in about 5 hours and its mostly because of changes i made to avoid getting my teeth kicked in quickly.

will contemplate this on the tree of woe and fire it up again tomorrow evening at market open.