No, you can keep the exp(). The simply means you are approximating in percentages. Note that I tried the SGU 2003 model with third-order perturbation. It turns out that the approximation order hardly matters, because the original model is quite close to linear. Thus, you will get almost the same results as the original paper.