On Fri, 1 May 2009, manish shukla @iimk wrote: To avoid predicting negative values, express the dependent variable in log form. Example calculation of calculate RMSE, MAPE: <script> open data9-7 # reserve 6 observations smpl ; 1989:2 arima 1 1 1 ; QNC 0 INCOME fcast 1989:3 1990:4 # focus on the out-of-sample period smpl 1989:3 1990:4 series fc = $fcast series err = QNC - fc print QNC fc err --byobs scalar RMSE = sqrt(sum(err^2)/6) scalar MAPE = sum(abs(err)) / 6 </script> Allin Cottrell