Winsorizing Panel data (eliminating outlier overall or across firms)

We have a panel data consisting of about 1000 firms with each about 150 monthly observations (Returns and distress measures such as Z-score).

In order to eliminate outliers we would like to winsorize at 1st and 99th percentile.

However we are not sure if:
a) to winsorize for each point of time (and across the 1000 firms) or
b) to winsorize across across the entire panel data

We would like to know what is the most used in research and best applicable in this case.
We're happy about any hint in this case.