Dear Vysh,

sectoral data for GDP are available on some international datasets like OECD STATS or WIOD/TIVA. To my knowledge, WITS does not provide GDP data. However, in my opinion, even when focussing on specific sectors, the Mass component should be calculated using total GDP as 1) there might be interesectoral demand spillovers 2) demand for final consumption is better captured by total gdp.

I hope this helped,

best,

Piero

Piero Esposito thank you a lot. I thought of that but wasn't sure whether it would be appropriate - thanks for the explanation!

Vysh Nanda, yes the theoretical framework requires sectoral expenditure shares, my previous answer was based on a misunderstanding of your question. Are the 2-3 sectors treated jointly (for each county pair and year you have 1 observation for each sector or you sum the data of the three sectors) or separately (a model for each sector)? How big are these sectors with respect to the total economy?

The separability property allows to estimate an equation for each sector and sectoral expenditure should be used but the caveat is that you fail to capture the susbstitubility of goods across sectors/classes. For this reason, many studies on single sectors use total GDP as mass indicator and, in case, add some proxies for sector specific-demand. The following publications might help you clarify this issue and also address the MRT problem.

https://www.wto.org/english/res_e/booksp_e/advancedwtounctad2016_e.pdf

https://www.unescap.org/sites/default/files/day2%20v2_intraAP%20goods.pdf

https://reader.elsevier.com/reader/sd/pii/S0960148118308437?token=61DD46280585C97B3A0CD9F9159E66777AE1AD63B127793422EAB4B196064752947E94A2B69169C48B8C3F7E7A4846B2

I hope this helped,

best,

Piero

Vysh Nanda,

I stay with my previous answer. With separate sectoral gravity equations, Total GDP is better than sectoral GDP due, among the others, to the fact that the importing sector could not be the same of the exporting one (intersectoral spillovers). In addition, your sectors are pretty large so in my opinion this approximation will have litle negative consequences. As robustenss check you might wanna experiment with manufacturign GDP (if you have it).

Dear Vysh Nanda,

in theory you would need a price index but in practice this variable doesn't exist. There are some ways to overcome the problem based on the use of dummy variables, in the publications I attached in the previous message there is a discussion of the methods to control for MRT.

Best regards,

Piero