Good evening all;
I am trying to analyse which are the key determinants of NPLs on a set of european banks in a panel data set up.
My dependent variable is the percentage of NPLs; (I am also considering to calculate the flow of NPLs since I have data on the absolute value of NPLs for each year for each bank).
As covariates I have both macroeconomic variables, common to all banks belonging to a given country, and some bank related variables, such as ROA ROE Capital Adeguacy.

I have tried to do some preliminary regression with xtabond2 using difference gmm but my results are not satisfactory; since the Hansen test rejects the validity of the instruments even when I use the full lag and the Arellano Bond test fail to rejects the presence of autocorrelation in the equation in levels.

Since these results are against all the relevant economic literature, I am thinking that I am misspecifying something and I need to decide wheter is better to use a logistic tranformation of my dependent variable or to switch to Fractional Regression following Papke and Wooldrighe.

Thank you in advance, any help will be appreciate.