I have a question about the estimated smoothed shocks.
As far as I know, the smoothed shocks are the best guess based on all observations.

The problem is that I’ve gotten a large scale (around 1e+04) of smoothed shocks over the initial 20 quarters.
Although it converges to zero after the initial 20 quarters, I was wondering what causes the initial spikes of the smoothed shocks.

The initial values for the Kalman filter are the steady state and the unconditional variance. If your observation equation is wrong, that starting value will of course be off.