Interpreting IRF graphs

I use dynare to estimate a loglinearized model. after estimation, I run the stoch_simul command. I find the impulse response functions are not start from 0 in the graph. But for loglinearized model, steady state is all 0s. and also some impulse response functions are in wave shape instead of hump shape. Why does that happen?

IRFs do not have to start at zero, because the first period is the one where the shock hits. If the respective variable is a jump variable, it will react immediately. Regarding the wave shape, I guess you refer to oscillating behavior. In that case you usually have complex eigenvalue, which often means there is still something wrong (I would guess a timing error). Another possibility is that your estimated parameter values do not really make sense and there still is some issue with the estimation.