How to determine the shock size

An important question: Is the amount of shock in the model can be arbitrary? if the shock is chosen arbitrarily, is this wrong?For example, We set the exchange rate shock to 10%.

I think you need to be a bit more precise.
Ideally you should calibrate the shocks of a model to observations of such shocks in the data. And then for impulse responses it is standard to report them for a one standard deviation shock.

Please read Setting size of stochastic shock