Panel Data - Time Trend

[Mohsin Javed]

I came across the following paper “Equity and bond flows to Latin America and Asia: the role of global and country factors”.

The paper uses panel data analysis, in particular, random effects model for the two regions. While I understand the methodology completely, one thing which I found new is that for all the regressions a “time trend” was included as an independent variable.

I have two questions:

1) In order to add a time trend, we use -c.timevar-
Is it the correct command? I have my timevar in quarterly frequency.

2) What is the rationale behind adding the time trend? I understand the rationale behind time fixed effects and individual fixed effects but this appears to be something new. The paper states that it is used to deal with the problem of possible stationary around a time trend.

Many thanks.

[Clyde Schechter]

1) Yes, adding c.timevar to the variable list introduces a time trend.

2) Well, in my field, epidemiology, we use time trends quite often. A time trend is used when you believe that the outcome variable in the model has a tendency to grow linearly over time. Occasionally a quadratic time trend is used, represented in regression commands in Stata as c.timevar##c.timevar. Either way, this is different from the usual situation in, say, finance, where there are usually just haphazard shocks to the outcome during each time period with no longer term linear or quadratic tendencies.

As for why the authors believe that in their situation a time trend is appropriate, if they do not explain their reasoning in the article, then you should contact them and ask them.

[Mohsin Javed]

Thanks.

As far as the interpretation of the time trend coefficient is concerned I am assuming it can be interpreted in the usual way, that is, suppose it is positive then we can say that the outcome variable increases with time. Is that correct?

[Clyde Schechter]

Yes, that's correct.

[Roman Li]

Dear professor Clyde Schechter ,
I would like just to ask you a small question: is it correct to code: c.timevar##i.Industry to capture Industry-specific trend?

[Clyde Schechter]

If you mean to ask whether including c.timevar##i.industry in a model will enable you to estimate a separate time-trend for each industry, the answer is yes.

[Adriana Dimas]

Hello,
I dont understand that time trend variable, is it another column of data i need to have or just run the regression and add that command?

[Carlo Lazzaro]

Adriana:
welcome to this forum.
Just add in the right-hand side of your regression equation:

c.timevar##c.timevar

(please replace the general term -timevar- with the corresponding variable that is present in your dataset).

[Adriana Dimas]

Thank you very much for your answer. I am stuck trying to figure out how to run a DID model for a term paper.

My data set looks like this:

URH is the Unemployment Rate (an average from before and after policy, separated by the bold line), TREAT indicates the presence/absence of the treatment, TIME indicates pre/post policy, and TREATxTIME is the interaction, and ATW and EAP are control variables.

My professor told me to add year and region fixed effect, I saw an example where the values in the data set are like these, (1 and 2 for year fixed effect in both periods of time, and 1,2,3... for region fixed effect). Is this correct? I tried running a regression in STATA but it will appear year fixed effect is omitted because of collinearity. How can I fix this mistake?
My professor asked me to add time trend, so I add to the right-hand side that STATA code?

I am attaching the specifications below:
(i) Baseline without any controls
(ii) with state-fixed effect
(iii) with year-fixed effect
(iii) with state and year fixed effect
(iv) with state-fixed effect and time trend
(v) (iii) with other control variables
(vi) (iv) with other control variables

I'd greatly appreciate any help.