GLS log of dependent variable

[giotira]

I am running a GLS analysis on a panel dataset. The dependent variable that I am using is the yearly netflow for US equity mutual fund from 2002 to 2012.
I am using the netflows in millions but the outliers (i.e. extreme values) seem to bias the results. Unfortunately I can not use the netflows as a percentage as the size is available only at fund level and not at share class level.
I was thinking to use a logarithm measure of flows. However flows in a fund could be either positive or negative. Therefore I was thinking to use the log of the absolute value of flows and then restoring the original sign (i.e. -1mil --> -ln(1mil)). This new function reduce the effect of the outliers but it is not defined in 0.
I do not have any fund with 0 flows (or in the range between -1<x<1) during any year in the sample considered .
Can I still use the GLS approach even if the function is not continue?
If not which approach do you suggest?

In the case in which I will use the log of flows, do I have to consider to use the log of all the independent variables (i.e. log(size) log(return))?

thanks