We were hoping to do exactly this, which does work in most cases. However, we do trend analysis with weeks so because the week links to our sales fact table then it doesn't represent the correct last year week.
In theory, if 1 week is filtered and the week is a dimension, then it doubles up the rows so that it would look like:
It's in general not possible to get an universal working comparison against groupings to classical calendars. Weeks are overlapping between months and years as well as month and years have different numbers of calendar-days and working-days.
In many scenarios is the resulting blurriness not essential to interpret the data and to derive appropriate actions respectively often have other factors much more impact. Nevertheless the effects could be minimized by comparing rates, maybe like:
sum(Sales) / count(distinct WorkingDays)
With more efforts specialized calendars in the direction of 445 or 454 or similar might be added and/or continuing month/week-fields are created, like:
year(Date) * 12 + month(Date) as Period
or maybe as offset like:
(year(today()) * 12 + month(today())) - (year(Date) * 12 + month(Date) as Period
whereby such measurements will increase the complexity and not each user will benefit from the advanced possibilities.
