Innovative Outlier Detection
This function serves to detect whether there are any innovative outliers (IO). It implements the test statistic lambda_{2,t} proposed by Chang, Chen and Tiao (1988).
detectIO(object, alpha = 0.05, robust = TRUE)
object |
a fitted ARIMA model |
alpha |
family significance level (5% is the default) Bonferroni rule is used to control the family error rate. |
robust |
if true, the noise standard deviation is estimated by mean absolute residuals times sqrt(pi/2). Otherwise, it is the estimated by sqrt(sigma2) from the arima fit. |
A list containing the following components:
ind |
the time indices of potential AO |
lambda1 |
the corresponding test statistics |
Kung-Sik Chan
Chang, I.H., Tiao, G.C. and C. Chen (1988). Estimation of Time Series Parameters in the Presence of Outliers. Technometrics, 30, 193-204.
set.seed(12345) y=arima.sim(model=list(ar=.8,ma=.5),n.start=158,n=100) y[10] y[10]=10 y=ts(y,freq=1,start=1) plot(y,type='o') acf(y) pacf(y) eacf(y) m1=arima(y,order=c(1,0,0)) m1 detectAO(m1) detectAO(m1, robust=FALSE) detectIO(m1)
Please choose more modern alternatives, such as Google Chrome or Mozilla Firefox.