A Note on the Use of a Logarithmic Time Trend

It is shown that parameter estimates in a regression with a logarithmic time trend are not invariant to the choice of the starting point of that time trend. Possible solutions to this problem are discussed and applied to data from a recent article in this Review where a logarithmic time trend was included. It is suggested that the appropriate course is either to estimate the starting date directly or to adopt a functional form invariant to the choice of starting date.