The Effects of Sampling Frequency on Detrending Methods for Unit Root Tests

This paper analyses the effects of sampling frequency on detrending methods based on an underlying continuous time representation of the process of interest. Such an approach has the advantage of allowing for the explicit - and different - treatment of the ways in which stock and flow variables are actually observed. Some general results are provided before the focus turns to three particular detrending methods that have found widespread use in the conduct of tests for a unit root, these being GLS detrending, OLS detrending, and first differencing, and which correspond to particular values of the generic detrending parameter. In addition, three different scenarios concerning sampling frequency and data span, in each of which the number of observations increases, are considered for each detrending method. The limit properties of the detrending coeffcient estimates, as well as an invariance principle for the detrended variable, are derived. An example of the application of the techniques to testing for a unit root, using GLS detrending on an intercept, is provided and the results of a simulation exercise to analyse the size and power properties of the test in the three different sampling scenarios are reported.