Abstract
This article evaluates the Smallest Canonical Correlation Method (SCAN) and the Extended Sample Autocorrelation Function (ESACF), automated methods for the Autoregressive Integrated Moving-Average (ARIMA) model selection commonly available in current versions of SAS for Windows, as identification tools for integrated processes. SCAN and ESACF can be applied to either nontransformed or differenced series, so the advantages and drawbacks of both procedures were compared. The best results were 79% of correct identifications for SCAN and 80% for ESACF. For some models and parameterizations, the accuracy of SCAN and ESACF was disappointing. The key finding of the study is that both human experts and automated methods provide inconsistent model identifications. Hence an elaborated strategy for model selection combining different techniques was developed and demonstrated on 2 empirical examples.
Notes
1In the following square brackets are used for [P, q] models where P = p + d.
2It is noteworthy that for the second-order parameters 0.8 and −0.2, ARIMA (1, 1, 0) = ARMA [2, 0] is a parsimonious nearly equivalent mathematical representation for ARIMA (2, 1, 0) = ARMA [3, 0]; and ARIMA (0, 1, 1) = ARMA [1, 1] is hardly distinguishable from ARIMA (0, 1, 2) = ARMA [1, 2].
3 A situation of nonstationarity is called the unit root problem, if in the first order autoregressive model Y t = φ Y t−1+ u t φ is 1. The name unit root is due to the fact that φ = 1: The autoregressive model can also be written as (1 − L)Y t = u t . The term unit root refers to the root of the polynomial in the lag operator. Lag operator L: LY t = Y t−1, L 2 Y t = Y t−2 and so on. If (1 − L) = 0, we obtain L = 1, hence the name unit root.
* p < .05.
** p < .01.