代码如下文所示,第一个命令是直接进行ARIMA(2,2,1)模型的拟合,第二个是对数据先进行2阶差分,然后对模型ARIMA(2,0,1)J进行拟合,都使用的默认方法(最小二乘法)。照理说这两个命令表达的完全是一回事,为什么参数估计值会差这么多?
将方法改成极大似然估计后(加参数method='ML'),直接拟合的结果又不一样了?!?( ar1 ar2 ma1 mean=
0.5045 0.0185 1.0000 0.2113)ㄟ(▔=▔)ㄏ。但是先作差分再拟合的话,结果和最小二乘法是一样的。
求大神帮助怎么解释?谢谢!
> arima(logts76,order = c(2,2,1))#直接拟合
Call:
arima(x = logts76, order = c(2, 2, 1))
Coefficients:
ar1 ar2 ma1
-0.4275 -0.4258 0.0811
s.e. 0.9527 0.2463 1.1072
sigma^2 estimated as 0.004866: log likelihood = 50.78, aic = -93.56
> arima(diff(logts76,1,2),order = c(2,0,1))#先差分后拟合
Call:
arima(x = diff(logts76, 1, 2), order = c(2, 0, 1))
Coefficients:
ar1 ar2 ma1 intercept
0.2168 -0.2399 -0.7294 0.0021
s.e. 0.2252 0.1883 0.1957 0.0031
sigma^2 estimated as 0.004594: log likelihood = 51.75, aic = -93.5