There is a purely geometric proof, but I do not feel like writing it.
Instead, one can also argue like this:
Calculate the cross-ratio $$cr(z_1, z_2, z_3, z_4) = \frac{(z_1 -
z_2) (z_3-z_4)}{(z_2 - z_3)(z_4-z_1)} = \alpha$$ Take a complex
number $a$ such that $a^2 = \alpha$ and pick an arbitrary point
$w_1$. Then form the points $w_2 = w_1 + a, \,\,\,\, w_3 =
w_1 + a + 1, \,\,\,\, w_4 = w_1 + 1$. Together with $w_1$ they form a parallelogram with edges $a$ and $1$. Then if you calculate the
cross ratio $$cr(w_1, w_2, w_3, w_4) = \frac{(w_1 - w_2)
(w_3-w_4)}{(w_2 - w_3)(w_4-w_1)} = \frac{(-a) (a)}{(-1)(1)} = a^2
= \alpha = cr(z_1, z_2, z_3, z_4)$$ Since the cross-ratio is
conformal (Mobius) invariant, there exists a Mobius transformation
(i.e. a complex linear fractional transformation) that maps the
four points $z_1, z_2, z_3$ and $z_4$ to the four points $w_1,
w_2, w_3$ and $w_4$. It can be constructed by solving the equation
$$cr(w, w_2, w_3, w_4) = cr(z, z_2, z_3, z_4)$$ for $w$ obtaining
a linear-fractional map $w = T(z)$ with coefficients formed out of
the points $z_2, z_3, z_4$ and $w_2, w_3, w_4$. The cross-ratio
condition guarantees that $w_j = T(z_j)$ for $j=1,2,3,4$. Now,
take the centroid $G_w$ (the intersection point of the diagonals)
of parallelogram $w_1w_2w_3w_4$ and map it back to $G_z =
T^{-1}(G_w)$ via the inverse Mobius map $T^{-1}$. Any inversion with respect to a circle centered at
$G_z$ and arbitrary radius will map $z_1z_2z_3z_4$ to a
parallelogram.