Neural networks for algorithmic trading. Simple time series forecasting
Alex Honchar
28314

i think you did a good job explaining the concepts and whenever things got too deep you linked to other resources. i liked that.

As others have pointed out you made the mistake of using future values for scaling and predicting absolute values instead of relative changes. My biggest complaint however is something else. Your graphs may imply to some people that you’re very accurately predicting future prices for 150 days straight. This is not even remotely the case.

You are predicting n+1 based on all previous values for 150 times. You could probably get a similar accuracy by simply guessing the the value of n for n+1. A model like this is useless.

Show your support

Clapping shows how much you appreciated Paul-Louis Pröve’s story.