Testing for serial correlation with lagged variable - error

[Dan Mikael Larsen]

Dear all,

I am currently trying to estimate the effect of language training for immigrants on the employment of immigrants on a municipal level. The data contains 370 municipalities and three time periods (2014, 2015 and 2016).

Dependent variable: ln_employed_immigrants
Variable of interest: ln_lag_participants
Control variables: ln_total_immigrant_population, ln_eligible_immigrants, prodindex, ln(revenues), ln_expenses
Dummies: urban, left, right, list

I am estimating the model with OLS and would like to test for serial correlation. However, as the variable of interest is lagged using the lag-operator (l.) I cannot perform the command -xtserial-.

It should be noted that data is not available before 2014, hence all observations for 2014 appear as missing in the lagged variable of interest.

My question is thus if there are other ways to test for serial correlation in my model? And if so, what would be the command?

Lagging variable of interest

gen ln_lag_participants = l.ln_participants

Regression

reg ln_employed_immigrants ln_lag_participants ln_total_immigrant_population ln_eligible_immigrants ln_revenue prodindex ln_expenses urban left right list i.year

Test for serial correlation

xtserial ln_employed_immigrants ln_lag_participants ln_total_immigrant_population ln_eligible_immigrants ln_revenue prodindex ln_expenses
no observations
r(2000) ;

Thanks,
Dan

[Phil Bromiley]

I suspect it doesn't make a lot of sense to test for a serial correlation coefficient when you only have two usable annual observations, but I can't readily explain why.

[Dan Mikael Larsen]

I managed to perform an Arellano-Bond test for serial correlation with the command -abar, lags()- that rejected the null at 5 per cent significance level.