Doubt NarX network and its implementation to meet future value

I would ask a question and I comentarais if possible. I have long been trying to establish a network of university Narxen work. My purpose is to predict the behavior of the closing price from a few inputs (volumen. ...) and some targets (closing price). I wish you would tell me if it is possible to do this with the network NarX. I developed a code in which I try to do 30 iterations in closeloop but I realized that future values introduced in this matrix, which is impossible because I can not introduce variables that will happen tomorrow for example. If anyone knows the answer it would be greatly appreciated. Thank you very much.

Answer 1

[Greg Heath]

Future values are NOT intoduced.

PREDICTIONS of future values are introduced.

Understand the difference now?

Hope this helps.

Thank you for formally accepting my answer.

Greg

[FRANCISCO]

One last question on this. If I have as inputs EMA and RSI and I predict LOCKS, get into the closed loop of the EMA and RSI today and tomorrow get closure. If you want to keep iterating to obtain several steps ahead, should get the EMA and RSI tomorrow and so on for several steps ahead. And my question is: How do I get the EMA and RSI morning and later days to continue iterating?. Or you can do this otherwise?. I know I should use the predicted values but I can not understand this step. Thank you very much.

[Greg Heath]

Have you run and closely examined the example in the documentation?

Have you compared open loop and closed loop performance on original data?

Have you compared open loop and closed loop performance on new data?

Will take a closer look at your problem this evening.

[FRANCISCO]

As it was in the stage of evaluating the accuracy was openloop and R2 have obtained the previously posted. Basically get R2 around 97-98% in train, validation and test, question that puzzled me a bit with respect to overfitting. In closeloop the mean square error is 2.766 e-4 and graphic representation is very similar to the actual data. In trying to evaluate the accuracy out of that sample is where I got the question of how to perform several iterations with the predicted values if the inputs I have not, that is the question that I mentioned earlier about the RSI and EMA. I must say that the inputs are highly correlated with the target and that can lead to greater accuracy but also if I use a lot of data, in this case 2200, overfitting may occur. I'm a little confused because I'm not sure how to evaluate the network closeloop not really know if I can produce overfitting, and finally do not know how to get several iterations with the predicted values for several steps ahead (espuesto example in this post ). Greg thank you very much for your time.