Dear Stata members
Let us say that my dependent variable is Net credit, derived as a difference between credit granted less credit availed scaled by total assets((Credit granted-credit availed)/Total assets)). My independent variable of interest is uncertainty proxied by rolling standard deviation of yearly GDP. I ran the panel regressions with some controls. My results indicate that
1)There exists a positive relationship between(significant too) Net credit and Uncertainty.
However, when I redo the panel regressions with
  1. Credit granted scaled by total assets as my dependent variable and uncertainty as my independent variable; coefficient is insignificant and +ve
  2. Credit availed scaled by total assets as my dependent variable and uncertainty as my independent variable; coefficient is significant and -ve.
I may be wrong but think this can happen because in the first case(net credit as dependent variable), it is a relationship between net credit and rolling std dev of GDP alone, hence the nature of the relationship will be a particular one(also with one residual). However, when I run the regression independently with credit granted and credit availed separately, all the covariates vary separately in these regressions hence a difference can happen. Can someone help me here to find a better explanation(theoretically and econometrically