Instrument exogeneity tests.

[Rishi Varghese]

Hello Statalists,

I had a question relating both to STATA's instrumental variable post-estimation commands as well as general economic theory - figured this place would be the best to ask.

I have used a lag of an arguably exogenous variable as my instrument but the exogeneity tests both come back significant suggesting this IV is endogenous.

Given that this is panel data and I'm using the xtivreg2 command, is the test significant because of the first-order correlation that comes with a lagged variable or is there just endogeneity that I haven't come to realise? Does first-order correlation still imply endogeneity?

Thank you for your input.

Rishi

[Maxence Morlet]

What test have you run? Certain tests (e.g. Sargan if memory serves) make unrealistic assumptions, e.g. homoscedasticity. You may want to read Hayashi (2000).

[Rishi Varghese]

Hi Maxence,

Well when I run the xtivreg2, first I get the Anderson-Rubin Wald result and LM statistic. But point to the instrument being endogenous, even though the instrument, which is crop price, is mainly determined by rainfall which is fully exogenous.