To get maximum performance you want a merge join between a and b and you want an index on a so you can use a stream aggregation for the group by.

This can be achieved by using the following indexes:

create nonclustered index IX_a on a

(policy, form, date, location_date, location_pin, state)

include (premium)

create nonclustered index IX_b on b

(policy, form, date, location_date, location_pin, state)

include (premprcnt)