Choosing between Random Effect and Fixed Effect Model

[Guest]

Dear All,

I have a problem when choosing the right model. I investigate how additional capital influences the company bankruptcy risk. I expect the coefficient on capital to be negative and significant.
I have unbalanced panel data. The firms from the sample are not randomly chosen. The coefficient for the bankruptcy risk is computed only for them.
According to random effect model the coefficient is significant and negative. When applying fixed effect model the coefficient is negative and not significant. I conducted a Hausman test. The results were significant so I should reject the null hypothesis and choose the fixed effect model. To be on the safe side I applied LM multiplier. I saw the value Prob>chibar=1.000 So should I trust rather Random Effects? However a value 1 seems bizzare to me. I'm really confused.

When controlling the Random Effects Model for years I still have significant and negative coefficient.

I would be grateful for any hints or literature

[Maxence Morlet]

By default, I would always use a two-way fixed effects (TWFE) model and consider this as baseline. You seem to be a fellow economist, and economists are obsessed (rightfully so) with causality. The identification assumption made by RE is generally really not plausible, therefore results will most likely be biased; you'll have a hard time convincing any reviewers that RE results are causal.

However, with this being said, TWFE is not a magic wand that solves all of our problems, despite it being a powerful tool. Wooldridge (2021) dicusses it at length, I really recommend you give his paper a read: https://papers.ssrn.com/sol3/papers....act_id=3906345

As an aside, the Hausman test makes a lot of unrealistic assumptions and may yield false results if these assumptions are not met...