Researchers may want to estimate the percentage change of a variable, such as household wealth or corporate profits, that takes on economically significant nonpositive values. Using the logarithmic transformation, however, requires discarding observations with nonpositive values. This paper describes a possible solution to this problem-the inverse hyperbolic sine transformation-and shows how to implement this transformation optimally in the case of median regression. As an illustration of the usefulness of this transformation, I revisit a specification sometimes used to estimate the effect of tax incentives on household saving.
©2011 Walter de Gruyter GmbH & Co. KG, Berlin/Boston