Research Design problems

[Joris Verpaux]

Dear all,

I have a question regarding my research into the effect of new covid cases reported (per million inhabitants) in 10 EMU countries on the yields on 10-year government bonds.
My research period is from 1 year before the first covid case (feb 2019) until one year after the first covid case reported (feb 2021) on a weekly basis.
My regression is as following:
Yi,t = B1*CovidCases i,t+ B2*CovidCases i, t *PostperiodDummy + B3*PostDummy + Control variables + FE. (Where Post is a dummy var. which equals 1 after first covid case reported)
But this regression leads to nonsense output due to the fact that the B2 isn't allowed due to multicollinearity problems and B1 is not useful because it also accounts for the first period where the covid cases are zero..

Does anyone have any advice on how to design this regression to get useful results on the effect of covid cases on the yields?

Thanks in advance,
Joris

[Fei Wang]

Joris, "CovidCases" itself measures the "shock" which is similar to the interaction term in a conventional DiD set-up. Therefore, "CovidCases * Post" is unnecessary. Just control for Covidcases, country FE, Post dummy (or quarterly or monthly dummies), and other covariates. As you have a small N-large T panel structure, some guidance can be found from https://www.statalist.org/forums/for...panel-analysis.

[Joris Verpaux]

Hi Fei,
Thanks for your quick response. I used xtscc on the regression equation minus the interaction term between Post and Covid Cases. However, I'm not sure if my Beta 1 on Covid Cases is relevant, due to the fact that it also takes into account the first year where the reported covid cases are 0. I've done the regression as well on the Post-period only and there the coefficient of Covid Cases is way higher and significanter.

regards,
Joris

[Fei Wang]

Joris, the CovidCases is a valid variable measuring the shock density of the pandemic. Zero means no shock, small numbers mean small shocks, and large numbers means large shocks -- All are components of the shock density. When focusing on the post-period, you are looking into the effect of pandemic shock density conditional on that pandemic has begun -- It's a valid research question, but is a different one.