Hello,

I am doing a panel data analysis (random effects) on the effects of audit committee characteristics on discretionary accruals.

However, I find that many of my continous variables in dollar amount have the problem of multicollinearity (Highly correlated) such as MV to BV, total liabilities scaled by total assets, ROA and operating cash flow scaled by total assets. The correlations are as high as 0.9. VIF is as high as 300. It is unreasonable but why?

As a result, when I try to use Modified Jones Model to estimate discretionary accruals, the adjusted R-square is as high as 0.95 which does not make sense in my opinion.

Although I have tried to find the information from some forums and it says that multicollinearity among control variables is not a big problem, because of multicollinearity, my model has a problem of endogeneity.

The problem of endogeneity goes away if I remove all the control variables with muticollinearity.

However I do not want to transform the control variables because it is uncommon to have log 10 of MV/BV, log 10 of ROA or log 10 of operating cash/total assets.

Therefore, what should I do in this situation?

If I ignore the multicollinearity, there is a problem of endogeneity..... Again what should I use with it?

Thank you very much for your help.