Estimating the number of unit roots: A multiple decision approach

The problem of detecting unit roots in univariate and multivariate time series data is treated as a problem of multiple decisions instead of a testing problem, as is otherwise common in the econometric and statistical literature. Four examples for such multiple decision designs are considered: first- and second-order integrated univariate processes; conintegration in a bivariate model; seasonal integration for semester data; seasonal integration for quarterly data. In all cases, restrictedly optimum decision rules are found for finite samples based on Monte Carlo simulation.