Endogeneity concerns: Suggestions for alternative to IV estimation

[Niels Meijer]

Statalist users,

I am conducting research on the effects of board [of directors] composition on risk-taking in banks. My dependent variable is a proxy for risk-taking.
My explanatory variables are:
Board size
Board independence
CEO Duality (CEO also serves as chairman of the board, dummy variable)

I also employ various bank level control variables and country level control variables.

Now I have run my regressions using xtregar command (Random effects GLS). Results are largely in line with expectations. Now I am looking for a way to address endogeneity concerns. A common way to do this is the Instrumental Variables method, i.e. re-estimate my model with IV. However, I am having difficulty finding instruments. Does anyone have a suggestion for an alternative way to address endogeneity concerns? Any input is very much appreciated. Any suggestions for instruments are ofcourse also appreciated.

[Niels Meijer]

Would a 2SLS estimation be suitable to alleviate endogeneity concerns?

[Phil Bromiley]

You didn't get a quick answer. You'll increase your chances of a useful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex.

2SLS is the most common way to address endogeneity, although Stata offers many other alternatives. [By the way, I would want to see if fixed effects give you different results on the variables of interest. This might also change the level of serial correlation in the errors.] The issue is you still need instruments. The problem is simply not identified without at least one instrument per endogenous variable.

Which variable do you think might be endogenous? With bank data, financial data several years before, geographic location, etc., might act as instruments.

[Attila Balogh]

Further to Phil's thoughts, you may need to first think about the suspected source of endogeneity (omitted variable bias or the related functional form misspecification, reverse causality, measurement error). That will perhaps provide the insight as to whether the instrumental variable approach is most suitable. Some endogeneity concerns may be addressed with difference-in-differences or propensity score matching methods, among others.

In this regard, your specific estimation method only comes after you have developed a credible identification strategy. Because you cannot explicitly test for instrumental variable exogeneity, the success of your instrument will largely depending on how convincing your writing is.

[Niels Meijer]

Thank you both for your helpful responses. I've managed to identify potential instruments for my explanatory variables. I will read in to how run an IV regression now, as I have not done it before.