Jarque‑Bera test Fail

[Rose Banda]

Hi everyone,

I am running an time series analysis using ARDL -ECM model consisting of 47 observations

the lag selected from running varsoc for my model is 3

when I carry out diagnostics including running the cumulative sum of recursive residuals (CUSUM) and cumulative sum of recursive residuals of squares (CUSUMSQ. Everything but the JB test passes the test. I get this for JB:

Jarque-Bera normality test: 8.318 Chi(2) .0264
Jarque-Bera test for Ho: normality:


When I change the lag to 2, the JB also passes the test, However, the significance of one of my main variables of interest also changes and becomes insignificant using the lag of 2.

I prefer the model in which the main variable Is significant (at the lag of 3). What can I do to improve the JB result? Are my results still good enough when JB fails as in the case above?


Thanks

[Nick Cox]

What would count as "good enough"? If you want us to guess at what some teacher, examiner, journal reviewer might say, that is really hard. They would usually want to see the dataset, as it is so small, and look at the entire argument of what model you fitted and why. Passing all the tests applied is a requirement in some fields, while others are more relaxed. I would trade any normality test result any day for a sight of qnorm for the same data.

[Rose Banda]

Sounds good! Thanks