考虑一个ARIMA(2,0,2)。如果我的标准化残差的ACF表明MA(5),那么我推定的模型将如下所示:
P <- arima(x, order=c(2,0,2), include.mean=TRUE)
我的一般模型看起来像这样:
G <- arima(x,c(2,0,5), fixed=c(NA,NA,NA,NA,0,0,0,NA))
以下是我的看法:
我有2个AR,其系数不同于0,然后是2个MA,其系数与0不同,所以我在开始时有4个NA。一般模型有2 + 5 + 1 = 8个参数来估算(2对于AR,5表示MA,1表示常数)。无论AR或MA的任何位置没有不等于零的系数,我都用零填充它。这就是为什么第5,第6和第7位置充满了zereos。
这是真的吗?如果没有,我该怎么做?
答案 0 :(得分:0)
您的一般模型的代码是为MA(5)添加术语,但是您使用fixed = c(NA,NA,NA,NA,0,0,0,NA)
强制所有这些术语为零,因此这两个模型基本相同。
> set.seed(1)
> x = rnorm(100)
> arima(x, order = c(2,0,2))
Call:
arima(x = x, order = c(2, 0, 2))
Coefficients:
ar1 ar2 ma1 ma2 intercept
1.2546 -0.3933 -1.3614 0.3614 0.1172
s.e. 0.5075 0.4106 0.5237 0.5231 0.0123
sigma^2 estimated as 0.7115: log likelihood = -126.24, aic = 264.47
> arima(x, order = c(2,0,5), fixed = c(NA,NA,NA,NA,0,0,0,NA))
Call:
arima(x = x, order = c(2, 0, 5), fixed = c(NA, NA, NA, NA, 0, 0, 0, NA))
Coefficients:
ar1 ar2 ma1 ma2 ma3 ma4 ma5 intercept
1.2546 -0.3933 -1.3614 0.3614 0 0 0 0.1172
s.e. 0.5075 0.4106 0.5237 0.5231 0 0 0 0.0123
sigma^2 estimated as 0.7115: log likelihood = -126.24, aic = 264.47
如果你想将ARIMA(2,0,2)与ARIMA(2,0,5)进行实际比较,那么你的一般模型应该是:
> arima(x, c(2,0,5))
Call:
arima(x = x, order = c(2, 0, 5))
Coefficients:
ar1 ar2 ma1 ma2 ma3 ma4 ma5 intercept
-0.1300 0.6333 0.0183 -0.7853 -0.1296 -0.1167 0.0132 0.1174
s.e. 0.2908 0.1944 0.3009 0.2166 0.1413 0.1511 0.1675 0.0124
sigma^2 estimated as 0.7067: log likelihood = -125.9, aic = 269.79