Scrutinizing the capacity of unit root tests in detecting non-stationary time series

• Main Author: Rivera, John Paolo R.
• Format: text
• Language: English
• Published: Animo Repository 2009
• Online Access: https://animorepository.dlsu.edu.ph/etd_masteral/4424
• Institution: De La Salle University

Time series data such as asset prices, gross domestic product, and exchange rates almost always exhibit non-stationary in the mean. Hence, a mandatory econometric task before performing time series analysis is to determine whether the series is stationary to avoid spurious regression results. Non-stationary time series can be detected through various unit root testing procedures such as the Augmented Dickey-Fuller (ADF) and Phillips-Perron. As such, this study exposits unit root testing in detecting non-stationary and analyzes its respective capacity given the various statistical properties present. Results show that the ADF and PP unit root tests perform relatively the same with each other. However, the ADF test is more efficient in detecting that a random walk, random walk with drift, and random walk with drift along a deterministic trend is non-stationary while the PP test is more efficient in detecting stationary among the family of AR(1) data generating processes (DGPs). However, there are instances where the ADF and PP tests may not agree with each other. Thus, the properties of the DGP and the descriptive statistics of a series must be considered in deciding which unit root test to use and which unit root test must prevail. Unit root testing must be taken seriously in empirical research. The accurate detection of stationarity in time series is vital on the selection of the best DGP for accurate time series analyses.