Source: Deep Learning on Medium
Hi there Akil, thanks for the response!
1. Ah I see, so perhaps it would have been more useful to use the SMA and then some indicator for momentum or volatility, like RSI. I read an article from almost a year ago now where someone attempted to do basically exactly the same thing as me and they did something interesting. They used a stacked autoencoder to perform feature extraction on the stock history and then feed the resulting latent vector into the neural network, as opposed to hand picking technical indicators themselves. This sounds like a promising approach, and I’d like to try and implement it in the future.
2. So it’s like individual stocks have their own behaviours? Interesting, the annoying thing is the dataset does feel quite limited with history only going back to 1999, if there is any larger datasets online I think it could be helpful with training a more accurate model, especially if the model is made more complex.