Q Results – Their overall accuracy is around 53%. When they consider the difference between the top decile and the bottom decile predictions, they get 3.35% per month, or 45.93% annualized return. UQ ???
I don't know your background and neither does anyone else on this forum. That's why when you make baseless anecdotal conjectures based on your alleged experience and personal biases, they count for absolutely nothing. Maybe if you were Mark Zuckerberg or Bill Gates, both college dropouts who managed to figure out how to write software, your opinions would have a little more credibility. But you fail because your opinions will never be able to overcome the facts and logic that I submitted above. Anyone here can test or dispute the validity of my claims. I challenged you to do so, but you could not come up with even one logical error or false premise in my reasoning. Given your difficulty with logic, I think I now understand why you so strongly discourage others from attempting to learn something new on their own. But just because you can't do something doesn't mean that others cannot as well.
@ET180 you're making all sorts of groundless accusations that I have no inclination to waste good Oxygen replying to. Welcome to my blocklist.
Results – Their overall accuracy is around 53%. When they consider the difference between the top decile and the bottom decile predictions, they get 3.35% per month, or 45.93% annualized return. ____________________________________________________________________________ In back testing he gets 3.35% - that means nothing at all since anyone playing with back testing + optimizing can easily come up with something like that. Everyone is a hero with the benefit of hindsight. Any system to brag about needs at least 5 years real world - real money to really even worthy of mentioning.
Thank you very much and I am a little farmiliar with the tools which you show and I will spend more time on them in the future and now I mainly use Matlab to do trading algo; Now I have use 'neural newtworks' to classify two situations: common situation and target situation; I get samples(about 50% belongs to common situation and other 50% belongs to target situation); after training, I get "Training Set Accuracy: 66.327313%"; because this is my first training data by ML and I don't know the result is meaningful or not? or how do you think about it?
To OP, If you are looking for 'big swings,' it is fairly well known that swings or volatility are easier to forecast than direction. The hard part is getting both. You could start to look into GARCH time series modeling to further explore that. People still seem to be stuck on thinking that a fancier implementation, like 'deep learning,' will improve forecasting of trading models. I think they are great at identifying and classifying stable and repeatable objects over time, but with financial ML, it is a different objective altogether.
Thank you for your information; Do you mean that ML would not get big win in finance like in other fields? I am just do this and know a little about ML in finance, so do you have tried it and confirm something?
What I mean is that if your objective is to forecast big price swings, you might be better off studying finance, statistics, and or time series modeling first. It might give you a better foundation to start with.