0
$\begingroup$

I am looking to generate a series of Chow F statistics for an inflation rate series. I am using the strucchange package. Specifically, the Fstats() function. I want to calculate the Chow F statistics using an ARMA(1,1) model on the data. I am not sure how to go about this using the function. Also, I would like to understand how to determine the significance line for the F statistics. When I calculate the F stat for 5% p value using df1 = k , df2 = n-2k, I end up with a different value than the line the graph provides, which is confusing me.

Two methods I tried

1) Fstats(inf ~ arma(inf, order = c(1,1))$fitted.values)

returned parameters:

number of regressor: 2

num of observations: 229

2) Fstats(inf ~ inf(t-1) + error(t-1))

returned parameters:

number of regressors: 3

num of observations: 228

where MA term or error(t-1) is calculated using the entire range of usable data.

Both methods return similar looking graphs with seemingly vertically shifted values. The documentation has an example that sort of follows method 2. I need help interpreting this the right way along with the p-value calculation. (Note: the actual F stats from the graph are F/k.. I saw this in the documentation.)

Method2 graph: alpha = 10%

Method1 graph: alpha = 10%

$\endgroup$

1 Answer 1

0
$\begingroup$

The Fstats() function relies on fitting linear regression models (via least squares), internally. Thus, fitting models with moving average terms is not straightforward. I would recommend to fit a low-order autoregressive model as an approximation to your intended ARMA(1,1) model. The strucchange package contains several worked examples, e.g., a seasonal autoregressive model for the UKDriverDeaths time series, showing how to best accomplish this.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.