This is problematic because frequencies that are not present will have fft magnitudes that are quite close to zero (or perhaps even zero). Inconsequential random noise could result in very large ratios. Thus, you may wish to do some filtering first, perhaps setting very small components to zero. And you are going to have to decide what ratio you expect if one or both components are 0.
If the data length is the same, then you can use the ./ (element-wise division) to construct the ratios. Otherwise, you may wish to interp() to the same frequency and extract the shorter length from each and take the ratios of those.
