Roberto Spadim

the idea is good, but the model isn't good, you cannot predict future with volatility and underling model as brownian motion (basic information to create a derivative formula), it simply...

ops wrong paper https://www0.gsb.columbia.edu/mygsb/faculty/research/pubfiles/841/sidbrowne_deadlines.pdf

@Jackal08 did you liked? =)

It’s like a fir filter Implement the kernel and run, nothing more nothing less

Lstm can reproduce a diff frac, the problem is if it will fit at noise or signal

Frac diff can be implemented as a fir filter, lstm can learn the parameters, the point is, instead of feature scaling/manipulation with fracdiff, you are doing a model fitting with...

the point is know what you are doing, lstm/rnn output maybe obfuscate by big math/calcs, a classification problem with good interpretation is better when you have a lot of money...

Number of backtests executed and considerations omitted should be exposed

Not sure but maybe numpy keras Backend? https://github.com/keras-team/keras/blob/master/keras/layers/recurrent.py https://github.com/keras-team/keras/blob/master/keras/backend/numpy_backend.py

* i forgot to sync X axis sharex=true, at subplot, and probably pass X (index) at plot() function