areg in cross-sectional data and multicollinearity

[Zygimantas Svitojus]

Hello, I have a cross sectional data consisting of 632 banks in 67 countries. In my dataset I have many variables with banks ratios, such as Tier 1 capital, Deposits, Loans ratios and etc (at one point in time). Following Beltratti and Stulz (2012) paper, I want to include country fixed effects and to cluster at the country level. I decide to use the following code in Stata: areg Y All_Xs, absorb(CountryID) vce(cluster CountryID) Is this is a correct code to use with my data? I'm a beginner in STATA and I read that fixed effects are normally applied in panel data, so I'm a bit confused if what I'm doing make sense. Also, I want to test for the multicollinearity. I use simple corr ALL_Xs code in STATA and I get the correlation matrix. However, I would also like to test Variance Inflation factor (VIF) to see if any of my variables are above threshold of 10. However, I can't use vif after areg regression. I know I could use command estat vce, corr but it just provides me with another correlation matrix table, and I struggle to understand should I drop some variables or not. Is it possible to test VIF with areg regression? Thanks

[Carlo Lazzaro]

Zygimantas:
1) if, on the grounds of the existing literature, systematic error autocorrelation is more likely at -country- level, you're correct at clustering your SEs at that level;
2) -estat vce, corr- (that, unlike -estat vif- you can invoke after -areg-) gives you an idea of possible problematic correlation. That said, please note that quasi-extreme multicollineraity, as oftentimes reminded on this forum thanks to a famous reply by Clyde Schechter, is rarely an issue as long as it does not produce "weird" SEs (whatever this may mean in the light of researcher's educated expectations). For a humorous explanation of multicollinearity, see, if interested, Chapter 23 of https://www.hup.harvard.edu/catalog....=9780674175440.