I am planning to investigate the effect of the war in Ukraine on the stock market prices by including in a GARCH model a dummy identifying the burst of the war.

I am wondering if I should strictly use the data post pandemic in order to rule out the effects of it in my analysis. Are 600 observations of prices (01 January 2021- 12 May 2023) enough to perform a good analysis?


1 Answer 1


I do not think you censoring your dataset in such way is good idea.

A) The pandemic actually overlaps with escalation of war in Ukraine (since the start of the war is actually being dated to 2014), even though media kinda moved on. However, the pandemic should still have an effect on agricultural output because many countries had some lockdown restrictions pass the Russian 2022 invasion. In fact some countries still have at least some restrictions in place (see data), it would require deeper analysis if those remaining restrictions still affect food production.

B) To get good estimates of what effect the escalation of war had on food prices you need to have data not just after the escalation but also before. Your dataset will in any case still include the effect of pandemic restrictions if you want to look at the effect of war escalation.

Hence what you should do rather than drop part of your dataset you should use even pandemic data but you should try to explicitly control for the effect of pandemic. You could for example use the stringency index or perhaps some other variable that you believe better captures the effect on food prices (e.g. you can dig deeper into raw data on the lockdown stringency and separate measures that impact food production from others that don't and create new index).

  • $\begingroup$ but the more observations you have the better This is generally true for cross-sectional but not for time series data. Many data generating processes (DGPs) in economics change over time, so that data points from a long time ago are no longer representative of the current DGP. Therefore, it may sometimes be beneficial to use shorter samples or downweigh the older data points in estimation. $\endgroup$ Commented May 15, 2023 at 7:16
  • $\begingroup$ @RichardHardy you are right thanks for pointing that out I edited my answer $\endgroup$
    – 1muflon1
    Commented May 15, 2023 at 8:44

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.