unobserved heterogeneity and time fixed effects

Question

I understand that if there are correlated effects i.e. (one of) our explanatory variables are correlated with unobserved heterogeneity $\alpha_{i}$ then there will be omitted variable bias. Because this $\alpha_{i}$ affects the dependent variable and is correlated to our explanatory variable. Therefore, it's important to account for time-invariant unobserved heterogeneity in a fixed effects regression.

What I am struggling with is why should we also include time fixed effects (i.e. year fixed effects for example). There might be a national policy in a specific year that affects all 50 US states in a given year. This could cause a shock on the dependent variable. But must these national policies not be somehow correlated to at least one of the explanatory variables to cause inconsistent estimates? Can somebody maybe name an example for more intuition where this is the case?

Thank you in advance.

Answer 1

You should include fixed effects whenever there are some time variant omitted variables that would affect all states/firms/individuals in your panel at the same time. For example, if you are looking at firm performance but you don't control for boom and bust using GDP you would want to include time fixed effects. Other examples of things that the fixed effects might absorb are things like impact of war, seasonal effects (if you have quarterly or monthly data), and so on. Generally as long as you believe there is anything that has time varying effect on all panel members (fixed effects do not help with time and cross section varying effects) you should have time fixed effects there to avoid omitted variable bias.