I am wondering about omitted variables in the context of intervention analysis. In my research, I have a time series of price differences between two regional commodity markets as the dependent variable. Those price differences were possibly influenced by a political intervention. Is it sufficient to include a dummy variable, that would be 1 after the intervention, and 0 before the intervention, in an ARIMA model?

Obviously, the transportation costs between the regional markets in my example could also influence the price differences. If I would have time series about the transportation costs, how could I include them in the intervention analysis? Since feedback between transportation costs and price differences is likely, I think that I would need a VAR approach?

  • $\begingroup$ @gung, I am curious in which cases it is advisable to add the econometrics tag. Regardless of what the tag Wiki says, I think of the distinction of econometrics from the rest of statistics more in terms of models and techniques (such as structural models based on economic theory and instrumental variables estimation) rather than just the applications. Because on one hand one can do a "purely statistical" modelling of an applied economics problem, but on the other hand one can do something "typically econometric" (distinct from a typical statistical application). $\endgroup$ Commented Nov 14, 2016 at 18:31
  • $\begingroup$ @RichardHardy, that's a reasonable point. I added the tag here b/c of the subject matter to which the model will be applied (as you surmise). It seems to me that knowing something about a topic is relevant to modeling generally, & that the issues that will need to be taken into consideration in this specific case are issues economists would readily notice & be used to considering. You could raise the issue on meta.CV for discussion, if you'd like. $\endgroup$ Commented Nov 14, 2016 at 19:04
  • $\begingroup$ @gung, OK. Because I treat the tags macroeconomics and finance are more like you treated econometrics here. $\endgroup$ Commented Nov 14, 2016 at 20:03
  • $\begingroup$ @RichardHardy, you could swap it out for one of those, if you think one is more appropriate. I don't doubt you are more savvy with these terms than I am. $\endgroup$ Commented Nov 14, 2016 at 20:08
  • 1
    $\begingroup$ @gung, as per your encouragement, I replaced econometrics with finance, although in this case it is not entirely clear cut. But I will try to ask the question on Meta tomorrow, I am curious about others' opinions. $\endgroup$ Commented Nov 14, 2016 at 20:13

1 Answer 1


Intervention Variables can be pulses (one-time effects) or level shifts/time/trends/seasonal pulses reflecting multi-period effects. Often the actual point of intervention can be different from the "known point" as there may have been either anticipation and/or lagged effects. I would not model the differences but rather model one price being predicted/affected by the other.

  • $\begingroup$ If I model one price being affected by the other, I unterstand that I would have feedback between the variables. Then a VAR model with the two prices and the intervention variable could be a suitable approach? $\endgroup$
    – Fabian
    Commented Nov 14, 2016 at 23:09
  • $\begingroup$ if you wished to predict there two series simultaneously ... yes $\endgroup$
    – IrishStat
    Commented Nov 14, 2016 at 23:16
  • $\begingroup$ Would I also need to include possible fundamental influences on the commodity prices in such a VAR approach, or could there be reasons to not include them? I expect that the political intervention only effected the price differences between the markets, not the absolute level of the prices. Therefore, I initially thought that only variables affecting the price differences (e.g. transportation costs) would be of relevance for me. However, I now think that I require a larger model that also accounts for possible fundamental influences on the absolute level of prices. $\endgroup$
    – Fabian
    Commented Nov 16, 2016 at 10:28
  • $\begingroup$ definitely include fundamental influences in the model $\endgroup$
    – IrishStat
    Commented Nov 16, 2016 at 13:09

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.