Global vs Local Approximation

Miranda and Fackler, in their terrific book, directly advocated the use of global approximations ahead of, say, the LQ-method, but their examples mostly included Micro DSGE models where the interest is mostly in the optimal decision rules of agents, not impulse responses.

So is there a particular reason why local rather than global approximation is employed in Dynare and so many Macro DSGE algorithms?

Take note that I am a 1st year graduate student and a relative newcomer to Dynare and DSGE

The issue is speed vs. accuracy: See e.g.