Solved – Durbin-Watson test for residuals of SARIMA(1,1,1) in R

rtime series

I'm modelling a time series data using ARIMA. Now, I'm trying to test for the serial correlation of my model SARIMA(1,1,1) using the durbin watson test.

My problem is that I don't know what linear model I would put on the formula of the dwtest function in R. Here's the usage of the function,

dwtest(formula, order.by = NULL, alternative = c("greater", "two.sided", "less"),
       iterations = 15, exact = NULL, tol = 1e-10, data = list())

Here's my code below,

Data: http://iitstat.weebly.com/uploads/7/3/4/0/7340846/chickenprod.rdata

To download the data just right click the link and click "Save Link As…"

library(forecast)
library(lmtest)
ChickenProd <- ts(ChickenProd, start = 1980, frequency = 4)
SARIMA111 <- Arima(ChickenProd, seasonal = list(order = c(1,1,1), period = 4))

The residuals of my model SARIMA111 is obtain by

SARIMA111[["residuals"]]

Now, I want to test the serial correlation of it using the Durbin-Watson test, but I don't know what linear model I would use in the formula argument of dwtest function in R. Is it the SARIMA(1,1,1) model? If so, how will I extract the coefficients of the SARIMA(1,1,1) model, and make a linear model formula in R?

Thank you in Advance!

Best Answer

I'm not sure how you would use the function with an ARIMA model - it requires a linear model object. Fortunately, the test is really simple to set up without the function.

Your test statistic (d) would be

   d = sum((SARIMA111$residuals - lag(SARIMA111$residuals))^2, na.rm = TRUE) /
       sum(SARIMA111$residuals^2, na.rm = TRUE)

That worked for an AR model that I had (I got d = 1.535959 with my data), so I hope it works for you. You'll have to look up this value in a DW table.

The na.rm = TRUE option is necessary because the first value of the residuals is NA.

Related Question