Performance of MIMO NARX neural network

I have a multi-input multi-output NARX NN. I am testing its performance with perform(). I have 2 main questions:

1. Do I have to switch Z and held_back_targets on perform()? Because Z is my net output data, right? and held_back_targets is the fed in target data, right?

2. I am using mse. Is the single value from perform() based of both outputs? How can I calculate the mse of each of the outputs separately?

Here is my code, the "..." means that I have code in between and the whole code is running correclty.

...
net = narxnet(inputDelays,feedbackDelays,hiddenLayerSize);
net = openloop(net);
...
[net,tr] = train(net,inputs,targets,inputStates,layerStates);

%Switch the network to predictive mode
newnet = removedelay(net,delay);
...
% Performance Testing
[held_back_inputs,held_back_inputStates,held_back_layerStates,held_back_targets,held_back_EWs,held_back_SHIFT] = preparets(net,held_back_inputSeries_altered,{},held_back_targetSeries_altered);
...
Z= net(held_back_inputs,held_back_inputStates,held_back_layerStates)
...
perform(newnet,Z,held_back_targets)

Thanks