Command xtabond2 gives error message: factor variables may not contain noninteger values

[Hjalmar Jernstrom]

Hello. I am trying to run the xtabond2 command in Stata 14 and I have encountered a problem that I could not solve using the solutions in previous posts on the matter. I am trying to apply the system GMM estimators in a panel VAR model. I am trying to assess the impact of investor sentiment obtained from social media (values ranging from [-1] to [1] on stock returns (expressed in decimals) and trading volume (absolute numbers). Below is my proposed command for the equation for the return variable.

xtabond2 return 1.return sentiment, gmm (sentiment, lag(2 2) collapse) small nonconst

Where:

• return is the daily stock returns for the sample companies
• sentiment is the value for tweet sentiment regarding the company (values -1 to 1)
• gmm estimates the variable that I believe might be exogenous

I will later run a second regression replacing return with volume. I use collapse as my instruments widely exceeds my groups (9 groups, 174 instruments). I use small because the sample is small and i want to interpret t-statistics.

When I try to run the command, i get the following error message: factor variables may not contain noninteger values. I believe this means I have to convert the variable to a continuous variable, but I am unsure wether that is a good solution, and I also do not know how to execute such a commando. Does anyone have an idea on how to resolve this error message? Any help is greatly appreciated! Hjalmar

[Dario Maimone Ansaldo Patti]

I think you wrote 1.return while you should have L.return. You want the lagged values of return don't you? In addition I would also use some lagged values of returns as instruments.

Also it might be preferable to use a two-step estimators (option twos).

Finally I do not think you are estimating a panel VAR model. This is just a dynamic panel data model.

[Hjalmar Jernstrom]

I think you wrote 1.return while you should have L.return. You want the lagged values of return don't you? In addition I would also use some lagged values of returns as instruments.

Also it might be preferable to use a two-step estimators (option twos).

Finally I do not think you are estimating a panel VAR model. This is just a dynamic panel data model.

You are correct, thank for your help dmaimone!