Normally when doing a study (any study not specifically for SWR), you would ideally decide beforehand as to what data was going to be used based on your knowledge of how it was collected, representativeness, structural changes in what you are studying, and potential for errors. Picking a data set after looking at the results is very, very, bad methodology.
Personally, I would lean toward using the longer period as it has more data and is a superset of the other.
Another thing to note, is that if you change the inputs slight but get very different results then your system is unstable. This means that you can't really rely too much on any results.