Difference in Difference Estimation Help- Tax rate vs Revenue

[Gonzalo Etchart]

Greetings,

This is my first post here but hopefully you guys can help me out!

I am currently performing a difference in differences estimation in Stata. I want to find out the impact of a corporate tax increase for agricultural companies in Mozambique on revenue ( tax went from 10% -> 32% in 2016, all non-agricultural industries have always paid 32%).


I have 384 observations in total for company revenue in the years 2015-2016. My treatment group in this case are the 7 agriculture companies in 2015, and the 8 agricultural companies in 2016. The control are the 369 observations detailing revenues figures for other industries (finance, service, construction).

I am assuming a parallel trend for these industries.


My question is: Firstly, is this ratio of treatment/control observations okay? And secondly is it okay if some of the companies in the 2016 dataset aren't in the 2015 one? (In other words, in agriculture for example I have 1 new observation in 2016, can I keep it or must I only look at the same companies in 2 different time periods?)


I really appreciate any feedback/suggestions on this matter.

Thanks in advance.

Kind regards,

Gonzalo

[Phil Bromiley]

Welcome to the Forum.

I don't do d-i-d, but I thought you wanted the firms to be the same in the pre-treatment period and then different after. You have the opposite which might be OK - I just don't know. You should look into it.

You have lots of control firms, the question is whether they are similar enough to the agriculture firms. I would be strongly tempted to consider including whatever other variables have been shown to influence agricultural firm revenues in general. That is, weather and crop prices probably influence ag firm revenues but don't necessarily influence the other firms which means you can't fully count on the d-i-d to handle such factors.

I wouldn't include the firm that appears only in 2016 since you can't have a difference in difference there. [It is interesting that you are looking at a change in revenue based on tax rates, not a change in net income or capital investment or some other variable.]

In future, 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.

[Gonzalo Etchart]

Dear Phil,

Thank you for your answer. I have opted for choosing company revenues as I am interested in the volume of trade rather than private sector profits. (Volume of trade is what affects small stakeholder farmers who sell to these companies the most). There are in fact other factors that would influence agricultural revenue as you said, especially the exchange rate, but I have no idea how I would include this variable in a regression.

Do you recommend I cut my data set to only include observations present in both the pre-post period?

Thanks for your advice as I am still new to the forum.

Kind regards,

Gonzalo