Future Prediction Narx-Net vs Nar-net

Recently i am working on a project to predict cryptocurencies prices and find future treds, through Neural Networks. I have as exogenus input X of 7 variables and output Y as 1 variable . In variable X i have some technical indicator such as RSI, MACD and some other factors such as Volume,MarketCap etc, and Y is the closing price each day.

My doubt is that i can use a NAR-NET with only Y(t) and try to forecast the future prices and find a future trend with some accuracy.

On the other hand i assume its more accurate to forecast with the use of 7 Narnets the future inputs of X and then try to predict Y, with the use of a Narx.

I know that the second way needs more time and memory but actually the market is affected with more than one variables and in my eyes its impossible to make future forecasts with just a simple NARX-Net due to the lack of the future Inputs.

Finally i need some advices on what way to choose to make a complete and effective project about cryptos.

Answer 1

[Greg Heath]

> My doubt is that i can use a NAR-NET with only Y(t) and try to forecast the future prices and find a future trend with some accuracy.

Nevertheless, since NARXNET uses BOTH output feedback and external inputs, the best way to start is

1. Calculate the autocorrelation function of the target T and design the corresponding NARNET

2. Calculate the 7 cross-correlation functions and design the corresponding TIMEDELAYNET.

Believe me, this will help immensely!

Thank you for formerly accepting my answer

Greg