LS-SVM time series forecasting

I’m trying to forecast a time series of air passengers using LSSVM with the help of the LS-SVMLab toolbox v1.8 from http://www.esat.kuleuven.be/sista/lssvmlab/, specifically the NARX model function.

I’ve read that LS-SVM as a forecasting method is a “no-model” methodology in a strict sense. Does that mean that every run I make for a one-step prediction on a particular observation would yield different results and that the hyperparameters would be different each run?

If so, how do I choose which predicted values to accept, since the hyperparameters per run varies tremendously.

Some papers have stated their “best” hyperparameters which led me to this confusion on whether specific hyperparameter values are obtainable to ‘hardcode’ it for subsequent forecasts.

Thank you in advanced.