|
ADF | Merit | (i) Test is reliable for high time series length. |
Demerits | (i) Calculates low power for a smaller time series length, often resulting in unit root conclusions even for a stationary time series |
(ii) Inappropriate choice of lag number adversely affects the test results |
Suggestion | (i) Reliable for apt selection of lag number |
|
KPSS | Merits | (i) Test is nonparametric |
(ii) Test outcome indicates stationary if the time series is strongly stationary |
Demerit | (i) The test statistic is vulnerable to type-I errors lowering the test’s reliability |
Suggestion | (i) Reliable for low time series lengths and recommended to be used along with another test |
|
PP | Merits | (i) Test is reliable for high time series lengths |
(ii) Test is nonparametric |
Demerit | (i) Low reliability for small and moderately large time series lengths due to severe size distortions |
Suggestion | (i) Reliable for higher time series lengths and shorter time series lengths having low parameter value |
|
Breitung | Merits | (i) Reliable for any time series length |
(ii) Test result is unbiased by any time series characteristics |
Demerit | (i) Fails to detect the presence of unit root for = 1 in absence of other nonstationary components for lower lengths |
Suggestion | (i) The test is helpful in accurately understanding the impacts of trend, seasonality, and volatility effects through test results |
|
MK | Merit | (i) Presence of the slightest trend component can be effectively detected |
Demerit | (i) Test is not reliable, particularly for higher time series length |
Suggestion | (i) Test cannot be solely used for assessing stationarity |
|
Levene’s | Merit | (i) Completely reliable |
Demerits | (i) Test is parametric |
(ii) Cannot detect the presence of trend component if the variance is constant throughout |
Suggestion | (i) Test is very effective for assessing variance and is recommended for use with some other tests for trend assessment |
|
KW | Merits | (i) Test is nonparametric |
(ii)Test is fairly reliable for higher time series lengths |
Demerits | (i) Cannot detect small differences in mean values |
(ii) Low reliability for lower time series lengths |
Suggestion | (i) Recommended for use with higher time series lengths |
|
SW | Merits | (i) Test is reliable |
(ii) Best performing test in all normality tests |
Demerits | (i) Nonstationary outcome does not mean that the time series is not stationary |
(ii) Low reliability with respect to skewness |
(iii) Significant type-I error for 0 kurtosis |
Suggestion | (i) This test is to be used first. If the test outcome is nonstationary, then other tests are to be used |
|
KS | Merit | (i) Test is nonparametric (two-way) and low type-I errors |
Demerit | (i) Very low reliability with respect to skewness and kurtosis |
Suggestion | (i) Two-way KS test is useful for stationarity assessment, but the use of other tests for confirmation of results is recommended |
|