当我们在 R 中计算 ARIMA 模型时，如何解决标准误差、z 和 pr 值的 NA 值？ 在 sqrt(diag(x$var.coef)) 中：产生 NaN

Question

I have been doing future forecasting using ARIMA model in R programming language but when I run ARIMA model from the forecast package, my standard errors are not calculated properly and got an NA error. Could you please advise me in order to solve this issue?

arima(5,1,5)

arima515 <- Arima(GSPC$SP500, 
                  order = c(5, 1, 5), 
                  include.constant = TRUE,
                  optim.control = list(maxit = 500),)

coeftest(arima515)

Output:

z test of coefficients:

       Estimate Std. Error z value Pr(>|z|)   
ar1    0.074793         NA      NA       NA   
ar2    0.142322         NA      NA       NA   
ar3    0.754132         NA      NA       NA   
ar4    0.179091         NA      NA       NA   
ar5   -0.370530         NA      NA       NA   
ma1   -0.122067         NA      NA       NA   
ma2   -0.180075         NA      NA       NA   
ma3   -0.751949         NA      NA       NA   
ma4   -0.147119         NA      NA       NA   
ma5    0.381992         NA      NA       NA   
drift  0.387899   0.132093  2.9366 0.003319 **
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Warning message:
In sqrt(diag(se)) : NaNs produced

As you can see above that errors, z values and pr are NA .

Answer 1

I don't know if I understand correctly the question, but if you want only parameter standard errors you could only call arima515 or summary(arima515) , like this

arima515 <- Arima(cop[,1], 
                  order = c(5, 1, 5), 
                  include.constant = TRUE,
                  optim.control = list(maxit = 500))

summary(arima515)

Series: cop[, 1] 
ARIMA(5,1,5) with drift 

Coefficients:
          ar1      ar2      ar3      ar4     ar5      ma1     ma2      ma3
      -0.7413  -1.2671  -0.5753  -0.3700  0.0221  -0.2412  0.5399  -0.7129
s.e.   1.4269   0.6982   1.1093   0.5289  0.0852   1.4266  1.1724   0.8139
          ma4      ma5   drift
      -0.1744  -0.4114  -1e-04
s.e.   1.0093   0.5184   1e-04

sigma^2 estimated as 0.5005:  log likelihood=-801.31
AIC=1626.63   AICc=1627.05   BIC=1682.05

Training set error measures:
                       ME      RMSE      MAE  MPE MAPE      MASE
Training set -0.005256166 0.7017616 0.513783 -Inf  Inf 0.6721109
                      ACF1
Training set -0.0009624799

you will get parameter estimates and se See the documentation for coeftest and see which object you can use for this function. Instead if you wanted to test anything else for your Arima model tell us what you wanted to do.