I need you to be completely honest with me. Nobody is trolling you. Look at the first three questions there again. You sincerely can't find one simple trick in all of them such that they can all be solved together in your head in less than a minute?
You're pressed for time solving (5 * 10 * 2) + 15 and putting some zeroes back? (13 + 11.3 + .7) * 10? 114 / 57? There's zero time investment for an analytical mind to put things into these terms and just blast through them.
Soooo much time leftover to work the main problem
Lmao
Ended up liquidating and moving stuff out of my accounts, now around 50% in cash now. Fucking hate doing this, but it gives me too much stress to let it all sit in equities while shit is hitting the fan, then I see shit like this popping up on the news feed XD
"Blessed are the idiots, for they are the happiest people on earth"
That sounds like the right thing to do. In fact, do that forever.
Good time to buy lol
This might be the conclusion of my trading bot. I discovered that my dataset had leakage so that's the main reason the bot was doing so well, once I rebuilt my dataset without leakage the edge dropped close to 0. My best 5 minute model now trades and makes 6k over the course of 3 years backtesting, and another model I have for hourly does better at 15k, so I was going to bundle them up together and run them in parallel to try to extract something on the order of 20k a year if I am lucky and the alpha holds.
This is a far cry from the millions of dollars I saw myself making... I am going to spin up these models and let them run on the paper trader but I don't know where else to go from here, there's no holes to plug up, no issues to fix, the models are running fine and outside of finding more historical data to train on...
I thought historical data source was the issue, but I was wrong, actually my historical was fine, but the features I built on that historical were able to look forward 1 hour for some indicators which explained why backtest did so much better than live trading... So no real issues to fix I think, I can try to start looking for historical/training data for other pairs like forex to try to hook up with IBKR, but I really don't know if it's worth it. I can try to build a separate model for natural gas as well.
Anyway, I feel like I've hit a brick wall, but I did manage to build a profitable model, I think I can say that with confidence, but it is squeezing out value. Barely making money. Lol. Slave wages. Expected earnings of 7k per year :) so, first struggle is to make a profitable model, now it's to make a profitable model that makes you a lot of money. Outside of leveraging up, I don't know where to go from here. I might try out the ensemble strategy because I have spent so much time building all of the infrastructure already and back end testing, I even hooked up my code to google cloud to run my training and am now looking to host my bot online so I can run parallel models and have solid uptime, but that's for much later if I am actually able to build different models that have positive pnl that are orthogonal to each other.
Get an astrophysics degree and learn the art of extracting the most information from the least amount of data. You need far fewer predictors and you need to do actual modeling and regression tests.
The astrostatistics symposium I attended was quite interesting, but not interesting enough. No thanks
fuck you budy
nah im just kiddin
Fuck YOU
Congrats man that's pretty kewl
Lol
OK I've found another potential breakthrough. Will need to find out and see but I tuned up my parameter search and got the model to make about 10k per year.
I'm running some more experiments later to test changing my rolling window training methods for walk forward to see if dropping older training data between 2009-2018 might improve the model and see what that perfect window of training data might be, maybe it's everything like I'm doing now for the walk forward training or maybe it's only the last 4 years, also going to implement live retraining on the bot to see if that improves testing too.
So it turns out there's still a lot of things to try to test out and see if it improves alpha I think it might. Looking forward to it. If I am able to potentially squeeze out another 10k so we are at 20k per year for the bot I think that's good enough that I'll invest in infrastructure to start trying to run multiple bots in parallel for different symbols because I think the methodology can splash over easily to different tickers.
Walk me thru it sometime
ive found another breakthrough. I have no thesis. I'm just going to give my bot more data and hope my model somehow gets better for some reason.
Can't you just stack these bots and make infinite amount of money or the buffs don't stack?
New breakthrough. The bot has begun to develop my tendencies. Mirror my actions. He started withholding information and asking more about my wife and kids. I think he wants to replace me. I’ve shut down the bot.
Yo that format kinda goes hard tho wth
