One step ahead prediction using NARX networks

Thanks in advance for helping me. I am totally new to Matlab and to Neural Networks. I searched in google and in different forums but I still have doubts about a couple of things.

I am trying to program a NN that returns a prediction for prices of a commodity for the next day when introducing 7 different technical indicators and the closing price of the previous day. It is a project for university so it does not really matter much how accurate it predicts.

My doubts are the following ones:

1)I programmed and trained a NARX network:

if true
  % code
end
%%TRAINS AND CREATES A ONE STEP AHEAD NEURAL NETWORK
%%Takes the data
X = tonndata(transpose(INPUT),true,false);
T = tonndata(transpose(OUTPUT3),true,false);
%%narxnet(inputDelays,feedbackDelays,hiddenSizes,trainFcn) takes these arguments
net = narxnet(1:1,1:1,8,'none','trainbr');
[Xs,Xi,Ai,Ts] = preparets(net,X,{},T);
net = train(net,Xs,Ts,Xi,Ai);
Y = net(Xs,Xi,Ai);
perf = perform(net,Ts,Y)
%%Creates the One-Step-Ahead network
netp = removedelay(net);
netp.name = [net.name ' - Predict One Step Ahead'];
[Xs,Xi,Ai,Ts] = preparets(netp,X,{},T);
Y = netp(Xs,Xi,Ai);
stepAheadPerformance = perform(netp,Ts,Y)
view(netp)

¿Should I train the removedelay network also? ¿Or with the first NARX training is enough?

2)If I need to predict one step ahead. ¿How can I call the NN without using any output data? Because I am not suppose to have them yet.

if true
  % code
end
%%Takes the data
X = tonndata(transpose(INPUTtest),true,false);
T = tonndata(transpose(OUTPUTtest),true,false);
%%Returns the results
[Xs,Xi,Ai,Ts] = preparets(netp,X,{},T);
RESULTS = netp(Xs,Xi,Ai);

Thank you very much in advance. I know that they are probably basic questions, but I felt lost.

Answer 1

[Greg Heath]

I HAVE COVERED MOST ASPECTS OF NARXNET DESIGN

   SEARCHWORDS             HITS        

                   NEWSGROUP   ANSWERS   
   GREG NARX          53         231
   GREG NARXNET       75         276
   NARX               86         379
   NARXNET            94         359

Since there will be many duplicates, the best bet is to search by category size ( i.e., 53, 75, ...,359, 379) IN REVERSE CHRONOLOGICAL ORDER !

Hope this helps.

Thank you for accepting my answer

Greg

P.S. Do not overlook the importance of using the auto and cross correlation functions to determine the statistically significant lags. It is not recommended to just assume how many and which lags are significant.