I always wanted to know, if it is possible to create a trading bot with a very small budget and to generate profit. So I started to read some books, watched some tutorials, refreshed my programming skills and created some trading bots. I ran them on a demo account during 2020 and after optimizing them a lot - I think I have finally good versions available ready to trade on the live account. For now, I regard this more in a sense of a proof of concept - so the funding on the real account started with 400 euros and the bots are running since 1st november 2020 on it. I have to consider that each financial instrument I have created a bot for, has its own margin requirements, minimal lot size and so on. So I started small, only with a few small bots and will hopefully bring the larger ones in 2021 when the account has grown. We will see. Enough text for now, please find below the real trading results of november 2020 so far. PS: I try to provide an english screenshot for next time ;-)
I say not possible, loads say they trade automated, nobody has ever backed this up with broker statements. You get that automated games for big money, the 1 that wins once, next time huge loser, conditions suited first time didn't next.
AI trading is not easy. If it's a predictor, you are going to end up high winrate with small profits and low winrate with huge losses. I agree with @Turveyd nobody shares their real statements. There are lots of AI gurus. If they can not write a profitable agent, there's a reason for that. The randomness of the market. Law of large numbers will not let you win.
"I ran them on a demo account during 2020 and after optimizing them a lot... " tl;dr It's easy to overfit, if your optimaztion process was lacking at all, you might find forecast accuracy drops way off as soon at the regime drifts to something new. tl;still gonna read it... Just be super careful about how you go about this. It's a lot easier to overfit a data set then it is to conduct a proper back-test. The latter is incredibly hard to do correctly. Also, my own personal experience is, it's almost impossible to apply ML to a single ticker and get a meaningful result. You have to be relentless with pre-processing the data. Make sure it's stationary, make sure it's balanced, keep features to a bare minimum (just a couple on a single stock can be too many), make double sure the target is meaningful, and even if you do all that, the likelihood that it overfits is still really high. With ML more data is always better and almost mandatory. So, if you want to forecast a single stock, it's better to do it from a large universe, then extract the individual result. All the above still applies, but having thousands of stocks and millions of rows of data that has been properly processed, will go a long way to finding meaningful signal that generalizes better over time.
Yep, they make money, by saying they've got a profitable system, waiting for the PM's then it's the no can't part with it, can't part with it, 10K ohhh go on then, what's it's not working for you, you must be doing something wrong.
Pretty much this. The quant and data science community on the broad stroke is pretty open about what it does. You can find a lot of open source repositories on github and there are some pretty active communities with a lot of robust discussion (and disagreements) about what works and what doesn't. What you won't find are any robust feature sets laying around. If someone can actually generate alpha from the markets they're not selling it for 49.99 a month.