Dear All,

Good morning,

I have an important issue troubling me. I had some prior trouble with the AR(2) test in my initial dataset. To this end, I

have increased the N (53) and reduced the T (4 and 3) of my dataset to mitigate potential issues of instrument proliferation. Hence, I'm now working with 53 cross-sections of 4-year and 3-year non-overlapping intervals. I am now more confident with the identification restrictions (more or less four degrees of freedom in overidentification).

But my concern persists: after performing the two-step System GMM, the results for the AR(2) errors test reads 'not applicable'. Can I use the AR(1) test as the information criterion for the Arellano & Bond autocorrelation test?

Thank you for the opportunity of guidance.

Best,

Simplice