random_coefmats generates random VAR model coefficient matrices.
random_coefmats(d, how_many, scale)Returns \(((how_many*d^2)x1)\) vector containing vectorized coefficient
matrices \((vec(A_{1}),...,vec(A_{how_many}))\). Note that if how_many==p,
then the returned vector equals \(\phi_{m}\).
how many \((dxd)\) coefficient matrices \(A\) should be drawn?
non-diagonal elements will be drawn from mean zero normal distribution
with sd=0.3/scale and diagonal elements from one with sd=0.6/scale.
Larger scale will hence more likely result stationary coefficient matrices, but
will explore smaller area of the parameter space. Can be for example
1 + log(2*mean(c((p-0.2)^(1.25), d))).