I can give a counterexample where $A:H\to H$ is a linear, bijective, unbounded self map of a Hilbert space with $A^2=\text{id}$.
Something similar can probably be done for many other Banach spaces as well, but I hope this gives a flavor of what might happen.
It might be that no counterexamples exits in somehow reasonable cases, but I don't know what reasonable should mean here.
What I do know is that this construction is not very reasonable, at least from the metric point of view.
Let $H$ be an infinite dimensional separable Hilbert space and $B$ a Hilbert basis (orthonormal, of course).
Write $B$ as a disjoint union of two countable sets: $B=\{a_i;i\in\mathbb N\}\cup\{b_i;i\in\mathbb N\}$.
($0\notin\mathbb N$.)
Now extend $B$ to a Hamel basis $C$ of $H$.
Define $A(a_i)=ib_i$, $A(b_i)=a_i/i$ and $A(x)=x$ for $x\in C\setminus B$ and extend linearly.
It is easy to see that $A:H\to H$ is unbounded and bijective and that $A^2=\text{id}$.
Note: If you restrict $A$ to the span of $B$ (Hamel span, without the closure), you get a more naturally defined operator with dense domain and range.