Empirical similarity-based approach for selection of unit root test

Abstract

The existence of unit roots in time series processes can impair the choice of techniques for analysis and forecasting time series data. It is of much importance in econometric modelling to determine the integration number of analyzed time series based on unit root tests. Though statistical theory provides broad range of unit root tests in standard softwares, the choice of an appropriate test highly depends on subjective assessment of the analyst. This paper considers similarity-based scoring approach for selecting the most appropriate unit root test for specific type of time series observations based on Chi-square statistic and which is able to reduce subjectivity. Six unit root tests are studied. The utility of the proposed method is illustrated in simulation. The most reliable test, which is found is applied to a real time series of some selected macroeconomic variables