Impact of Unit Root Test Selection on Automated ARIMA Model Order Identification: A Comparative Study of ADF, KPSS, and PP Tests

The auto.arima() function in R is widely used for automated ARIMA model selection in time series analysis. However, its performance may depends on the choice of stationarity test for determining the order (p,d,q). This study investigates how the auto.arima() function behaves under three unit root tests (three stationarity tests); Augmented Dickey-Fuller (ADF), Kwiatkowski-Phillips-Schmidt-Shin (KPSS), and Phillips-Perron (PP), when determining the correct ARIMA model order (p,d,q). Using ۱۰ simulated ARIMA(p,d,q) models with varying parameters, we compute the percentage of correct identification of (p,d,q) simultaneously for each test. Among the three unit root tests, auto.arima() function on KPSS achieves the highest mean correct identification rate (۲۹-۳۵-۴۰%), followed by ADF (۲۸-۳۴-۳۸%) and PP (≈ ۲۸-۳۳-۳۸%).

کلیدواژه ها:

ARIMA model ، auto.arima() ، ADF test ، KPSS test ، PP test ، time series ، stationarity