I am currently working on a research project regarding the introduction of mandatory earnings guidance regulation in multiple countries within a time span of 10 years (staggered adoption). As I can see that my results depend on the FE I use, I tried to figure out how to determine which FE I have to use in this setting. Literature doesn't give a clear answer as different authors use different FE in similar models (firm, year, country, industry or a combination of those four). That's why I wanted to ask the community: Is there any rule-of-thumb or a precise answer on which FE I do have to include in my multi-year multi-country model (using linear model with an unbalanced dataset) in the form of:

Market liquidity measure = b0 + Analyst_following * x1 + US_Listing * x2 + IFRS_User * x3 + MarketCapitalization * x4 + ShareVolatility * x6 + e

I have detailed firm level data for all variables. I would suggest industry-year FE, but don't have any arguments why I should use them (just intuition). Is there any clear guidance on which FE to use or any recommendation for a book adressing this topic in an inuitive way? Most books deal with social research questions (using e.g. race as FE as it does not change) but it is hard for me to transfer this logic to a business regulation setting... What I want to measure is, whether the staggered adoption of the regulation has a significant impact on market/stock liquidity. Dependant/independant variables are derived from well-established papers in my field.

Any help appreciated!

  • $\begingroup$ I would need to see the exact structure of your panel data first, but it is most common to use the finest level FE + year FE. For example, use (firm FE + year FE) for firm-level panel data. But nobody says you must use them. It's all your choice, and it depends on what you want to capture. Different models are interpreted differently. The reason you don't see a clear answer is that there is no clear answer. $\endgroup$
    – chan1142
    May 23, 2019 at 13:35
  • $\begingroup$ That said, it would be easier for us to help you if you state the structure of the data, the real model (not in terms of y, x1, x2, etc.) and what you want to see (not in mathematical language but in economic language). Modeling is more about economics (as opposed to econometrics or statistics or mathematics). $\endgroup$
    – chan1142
    May 23, 2019 at 13:40
  • $\begingroup$ @chan1142: Thanks for your response. I updated my question. If you still need more information let me know! For the moment I use country and industry-year fixed effects. Substituting country for firm FE does not change my results but it would be interesting to know how to determine the right set of Fixed Effects for this research question! $\endgroup$ May 23, 2019 at 16:18

1 Answer 1


It seems that you have firm-level panel data over multiple industries in multiple countries.

Year fixed effects control for trends. Do not omit them. For others, a general rule is: ___ FE's control for time-invariant heterogeneity across different ____, but do not control for heterogeneity within the same ____. You replace ____ with "firm", "industry", "country", "industry-country", etc.

Firm FE's successfully control for all firm-level time-invariant effects that are possibly correlated with the RHS variables. So you compare different years within the same firm. In contrast, industry FE's control for time-invariant effects (that are possibly correlated with the RHS variables) across different industries, but do not control for firm-level heterogeneity (fixed effects) within the same industry. And so on.

I personally think a model with firm FE (and year effects) is easiest to defend using. If you can explain why firms within the same industry are homogenous, you can use industry FE, etc.

  • $\begingroup$ Alright, this means, when having firm-level panel data over multiple industries (which is most common in my research field) then Firm FE are kind of mandatory. Thanks for your reply! $\endgroup$ May 27, 2019 at 8:20

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.