The short answer is: we need both.
When we do archival accounting research, we often need to merge data across various databases, such as COMPUSTAT, CRSP, and EXECUCOMP. Stata can by no means beat SAS in this regard. SAS supports full SQL (a language specifically designed for database query), whereas Stata only has a “baby” merge functionality (really, “baby”!).
When we get all data and start to do data analysis, such as statistics, correlations, and regressions, this work can be done much more efficiently by Stata than by SAS. Not mention many handy packages developed by Stata user community, an ecosystem like Apple’s App Store.
So, the best strategy is to use different software at different research stages. The learning cost is not as high as this strategy sounds like—Stata is easy to learn (in fact, way easier than SAS). Anyone can command Stata very quickly.
I have attached a PPT on this topic in a bit more details (BBLG SAS vs STATA).
Finally, I use the following quote to conclude (thanks to WRDS.US Tutorials Series):
Even though data management and regression can be performed in SAS, some users prefer to use another package to do the ‘final’ steps. For example, SAS can be used to retrieve and manage the data. The final dataset created in SAS can then be converted for example to STATA format (using StataTrans). STATA can then be used to create the tables with descriptive statistics, correlation tables, and perform (final) regressions and other statistical tests.