If you want the median of the date-specific changes, then Sandeep's solution works as long as both months have the same number of days.
If you want the median percent change of all the measures in June to all the measures in September, then you want the percent change on all possible combinations of June-September measures. This involves getting a Cartesian product of the measures in June and September and then getting the median of those combinations. For June-September, there would be 900 (30x30) combinations.
Substitute a PROC SQL for Sandeep's merge
CREATE TABLE combined AS
SELECT data1.BP AS BPJune,
data2.BP as BPSept
OUTER JOIN data2;