Super great stuff. I have actually worked on this kind project last year with my co-worker. And we have proposed once the idea of using RNN and CNN for our target. But we cannot go that far.
This article gives me a very clear idea about how RNN work and it is very likely that RNN can be used in the stock market prediction.
I have some concerns here:
1, will the sampling data be enough to train the massive among of parameters in the model? Based on my understanding, in order to have a good train of the NN parameters, we need lots of training data. But for the stock market, even we take the time-step 10 mins, we only get around 150k training data. Is that number enough to train the RNN parameters?
2, Other than the word prediction, the model of the stock market changes more often. For example, the interest rate has been reduced year-by-year over the past 30 years and it is almost zero right now. And we face more-and-more shortage of the natural resources year-by-year. Anyway, the stock market model could slightly change by time. I believe RNN can be adaptive to the model change. But my question is: is that fast enough?