I have a statistical data warehouse with statistical records for
offices per month:

Office (varchar(50)
Month (smalldatetime)
Collections (float)

I'd like to extract a time series of the average of the top 20
offices' collections in a resultset which is one record per month. 
e.g. the top 20 offices in a given month are taken, averaged and
related to that month:

Month, AvgTopCollections

Ideally somthing like:

SELECT Month, AVG(TOP 20 Collections)
FROM Stats
GROUP BY Month

I know how to do it for all offices and one office, but I cannot get
any one the many types of correlated subqueries I've tried to properly
parse.

The final goal is to compare a single office to company-wide average
and best-practices average over time in a simple line graph with three
lines.

Is there no way but to open a cursor?

This SQL is generated not knowing exactly what your DB structure is
like.  I'm assuming that your Stats table has an Office_ID as a
primary key and "Sales" as the number you are interested in.

SELECT Month, AVG(Sales) 
FROM Stats as s1
WHERE 
   Office_ID IN (SELECT TOP 20 Office_ID FROM Stats as s2 where
s2.Month = s1.Month ORDER BY SALES)
GROUP BY Month

The inner query in this statement selects the TOP 20 offices for a
particular month.  This query is joined to the outer query's Stats
table's month.  The outer table is grouped by

What on earth is wrong with opening a cursor?!?

There is nothing wrong with opening a cursor, and I think many times
it can be a very efficient way to solve a problem.  I was thinking
there might be a way in SQL which is easy.

For instance, SELECT COUNT(DISTINCT x) FROM y is valid SQL and useful,
yet is not immediately discernable from most documentation and saves
having to do SELECT COUNT(*) FROM (SELECT DISTINCT x FROM y).  I was
wondering if there was a way to easily get an aggregare average of a
subset within a group.