Financial Markets project Doubts

[klaudiaplk]

Hello Professor!

I have a question about this subsection: compare the performance against the naive LSTM approach. Is there any specific architecture that I need to compare my target solution with? And for: implement your own sequence forecasting system (with significant differences from existing methods) - Can I for example use here GAN network with this LSTM as generator?

Thank you!
Klaudia Palak

[lucmos]

Hi!

I would say you need to compare your approach with at least the single LSTM model. You can find a reference in the section "RNN with LSTM Model (LSTM);" in this report in the project references.

Yes a GAN approach with the LSTM as the generator is enough regarding the "significant differences"!

[LeonardoEmili]

Hi!
We are also working on this topic and are tackling the problem of stock price forecasting using recurrent neural networks: ranging from simple RNNs up to recurrent networks that employ attention mechanisms and convolutional layers on top of the recurrent architecture. Moreover, lots of effort was put in the direction of feature engineering in order to improve the model's capabilities. Do you think this system shows significant differences from existing methods? @lucmos

[lucmos]

Hi @LeonardoEmili
Yes, I think that works too.

As a rule of thumb: it is not ok if your project is just a collage of tutorials/snippets that can be found online (e.g. notebook in kaggle competitions, blogpost); it must contain some novelty