Panel data with non-stationary DV

[Matt Roberts]

Hi, I'm carrying out some research using panel data. The dep var is non-stationary and I am unsure how to rectify the problem. I would normally just use a first difference model but in this instance I need to use examine the level of the dep var for theoretical reasons.

My initial thought was to use xtpcse with the AR1 option to avoid using a lagged dep var. Would this address the non-ststionarity issue?

[Phil Bromiley]

You'll increase your chances of a helpful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex.

Different disciplines treat stationarity differently. Some would insist on a lagged dv and would not accept just controlling for serial correlation.

[Marcos Almeida]

I wonder whether the command - xtgee - following a thorough ‘check up’ of the variance-covariance, fiddling with autoregressive options, then looking at the information criteria as well, wouldn’t fit in your needs.