[Sebastian Kripfganz]

Using deeper lags generally has the side effects that those instruments become weaker. With weaker instruments, standard errors tend to become larger. This could explain your observation that the coefficient of the lagged dependent variables became statistically insignificant.