Because parametric tests require independence which permutation tests do not.

To avoid reliance on distributional assumptions like normality.

Because permutation tests always yield a smaller p-value.

To increase the Type I error rate.