Hi guys, newbie here. I have been into algo trading for a while now and I'm at the point where I start wondering if it's possible to build something that is profitable over long time. I have built few models in Python for stocks on several exchanges (US, Europe) and all of them predict the next day's closing price based on the data from the past. The accuracy is fine on the test set etc... but in real life it doesn't work. Another prove of efficient market theory. And if I just think logical it does make sense. Data from the past just doesn't say anything about today because today is already incorporating future. This made me think about the model features. If there would be no limitations, what features would I choose for the model. The price at the moment is a reflection of ask and bid. Is there any place where we can get this? Obviously there is the order book, but that is already status from the past, so according to EMT it's already affecting the price. People submitting the orders is the point where future price is being created. This means that brokers have possible advantage. The bigger the broker (market share) the better their data is. I think crypto manipulation by exchanges shows this as well. What do you think? Is this the only possible way to really predict future price correct over a long period?
you have just began a 10,000 mile journey on foot. ps on bare foot walking over nails and broken glass, with razor blades oh and super glue.
Algo trading is much more difficult than people imagine. Intraday is ridiculously competitive due to the world's smartest people wanting a piece of the cake (and indeed securing most of it). The market is not efficient, it's however far more efficient than you might believe when reading your stereotypical trading forum though, which generally hints there are mysterious price patterns in random walks everywhere. Claiming "EMT" becomes something of an excuse for failure, to protect the ego. Instead, it's probably much more productive to spin it like "there are far more effective uses of my time" (which is true for most) if you want to quit. Algo trading is an incredibly saturated industry.
saturated with those who refuse the honest truth, more so overrun with those who lust for money. the only true algo conquerors are in it for the passion of figuring it out, without passion all you have is failure.
That's almost religious. I'm more into facts. Is there anyone who has successfully built a strategy which works over a long period, except hft which obviously falls outside the EMT? Btw, I'm not new in this. I have been around with data for 20 years. Wouldn't last that long without passion.
Personally, delayed reactions to Fed interest rate decisions for instance [back when the market cared about those, post COVID this feature is kind of dead for the time being]. You could enter early and get a tail for hours after the announcement with huge reward per risk. Then there are obviously people I trust that algo trade (even on ET), but I realize second hand experiences is not what you're after.
Sure the reaction is stochastic, as is the win/loss binary outcome. On expectation you can formulate a strategy with positive expectancy though. Most notably with a trend following strategy that adds to winners (if you want to keep max loss manageable by sacrificing max and expected winnings). I should add this is as a discretionary trader able to somewhat dissect the information in the announcement as to filter out a number of "do nothing" events early, but pretty sure an algo could do it too. I just wanted to submit a concrete counter point to EMH.
20 years of data mining and yet, you still couldn't figure out at least one setup to make money??? Maybe it's time to change your passion lol...