Standardization problem: Mu and sigma not fitting to new values

Hi guys, i got a problem with standardization.

Before I put my data into a linear regression model (fitlm), i standardize. I save the sigma and mu value for late retransformation.

Anyway the model is getting trained. After training I check how big the error between prediction and the actual test data is. Then I calculate the mean squared error (mse), which is aswell standardized since the error is standardized. I use the crossval-function to calculate.

And this is the moment when my problems starts, because when retransforming this standardized mse to the "real world", the value is completly wrong, compared to the mse I would get when retransfrom the data before calculating the mse. The reason for this problem is, as far as I understand, that the errorvalues are not part of the original distribution, so sigma and mu are not accurate.

Is there a way to retransform a standardized mse?