No, there are no problems. The stack of principally polarized abelian varieties $\mathcal A_g$ has a universal family $\mathcal X_g \to \mathcal A_g$, which is a relative group scheme. As such, it has an endomorphism $\mathcal X_g \to \mathcal X_g$ representing multiplication by $N$, which is finite and flat. Its kernel is the stack you want; the map to $\mathcal A_g$ is finite and flat. In characteristic prime to $N$, it is also étale.
[Edit] I was too hasty. As Kevin points out, one should take points of order $N$ in the kernel. This is an open and closed substack when $N$ is invertible. Over $\mathbb Z$, the kernel will have several irreducible components that will meet over primes dividing $N$, and it is not so clear to me how to distinguish them. For example, suppose that $N = p$, where $p$ is a prime; then all the components of the kernel will intersect over the locus of abelian varieties with $p$-rank $0$. It is not clear to me how to give a modular interpretation of the closure of the locus of points of order $N$ over $\mathbb Z[1/N]$.