Forecasting Performance of Models Using the Box-Cox Transformation

The authors examine the small sample properties and forecasting performance of estimators In models using the Box-Cox transformation via a Monte Carlo experiment They develop a simple,estimator for the expected value of the untransformed dependent variable They show that the sign and magnitude of the transformation parameter Influence the precision of the estimators and the forecasting performance These results support previous research At different values of the transformation parameter, smaller variances of the parameter estimators do not necessarily imply improved goodness of fit for the model