Noisy Data and Uncertain Coefficients: A Reply

In his comment on our paper, John H. Herbert brings out an important issue: [N]o one (except KSS, as far as I know) has recommended using the bonds approach for forecasting. KSS however, does not reply on a well-founded statistical basis for their forecasting procedure: it is decidedly ad hoc. We are pleased that Herbert has recognized the potential of "the bounds approach for forecasting" as a viable alternative for forecasting. His major concern is with the "ad hoc" nature of our forecasting procedure. This apparent occurs because forecasts generated by the NVM do not come with confidence intervals similar to those placed on forecasts generated by classical regression techniques. We wish to reply.