Compare IRF results from different studies

[barangaroo]

I try to compare the impulse responses from different studies. However, I am not sure if my conclusions are correct. This is my problem:

Two studies that use monthly data report the responses of stock returns to a shock in oil prices. Both studies use vector autoregressive models and then construct impulse responses with a 1 standard deviation shock in oil prices. Now I want to compare both studies and thus I look (for example) at the response after 3 months. However, I am not sure if I can directly compare the absolute values of the responses? Or do I have to know the standard deviation for the oil price time series from both studies and then I can compare 0.0017/SD_1 (first study) with -0.01/SD_2 (second study)?

Why is the size of the shocks usually not given in absolute values (e.g., a 10% shock in oil prices)? If I would know that the shock in the first IRF is 10% and in the second its 20%, then I could divide the second response by 2 and then both values for the IRFs after 3 months should be directly comparable?

This figure illustrates my problem:




Thanks a lot for your help!

[TomDoan]

Those graphs in isolation wouldn't tell you much since you can't figure out the size of the oil shock just from the responses. You're correct that you can rescale the responses if you know what factor you need to create a standardized shock size. However, there can also be differences due to how the shocks are constructed. The first graph, for instance, is a one-standard deviation "generalized" shock (the adjectives on the graph are backwards from what they should be) which means it assigns the contemporaneous correlation between the two variables to oil, thus it has a non-zero impact on stocks. The second graph may be doing something different.

Standard deviation shocks are used because they have a size which is historically "typical". However, some papers do standardize to a particular size when shocks to only one variable are of interest.

[barangaroo]

Thank you for the answer!

So, just to be sure that I got it correctly: if both studies report that their shock is a one standard deviation shock and if both studies report the standard deviation of the shock variable time series (here, SD_1 and SD_2), then I can directly compare the absolute values for both responses (i.e., in the two plots this would be 0.0017/SD_1 (first study) compared to -0.01/SD_2 (second study)? And of course assuming that the shocks are constructed in the same way.

[TomDoan]

If you're given the scale information, then yes; you can use that to standardize the responses.