On data transformations and evidence of nonlinearity

In this paper we examine the interaction between data transformation and the empirical evidence obtained when testing for (non-)linearity. For this purpose we examine nonlinear features in 64 monthly and 53 quarterly US macroeconomic variables for a range of Box-Cox data transformations. Our general finding is that evidence of nonlinearity is not independent of the data transformation. Results of simulation experiments substantiate this finding.