Possible discrepancies in System Identification Toolbox?

I'm using MATLAB R2014a, and trying to use the System Identification toolbox to find the transfer function of a given data set consisting of input and output time domain signals.

When I use the System Identification app, as shown by the tutorial, I remove means from my data and split my input into two halves (for estimation and validation). Then, I use the Transfer Function Models estimator (which initializes using "iv" method and uses TFEST to estimate) and get a best fit of ~45%.

When I use the same input and output data and use iddata() and tfest() in the command line, I get a FitPercent of ~61%.

Then, if I import the estimated TF obtained from the command line into the System Identification app, and validate against the validation data I'd defined already, I get a best fit of ~68%.

Can someone explain why I'm getting three different Fit estimates for the same input and output data, considering I'm using the same data set, but two different methods of estimation (the app and command line) which apparently use the same estimation algorithm (tfest)?

For reference, it's a 4th-order system.