Design Question Relating To A Financial Application

EDIT: Please skip over this and go to the 2nd post. After writing this, I realized it may be too complicated to comprehend, so I tried to simplify with another example which is found in the 2nd post. Or if you want you can read this one and try to understand it.


I would like to go over a design we came up with to handle a certain reporting requirement from an OLAP cube to see if this is a good approach. Also, I would like to hear about other designs that we can consider.

Here is the condensed version of the requirement.

We are measuring the performance of a bunch of portfolios. So among other dimensions, we have a portfolio dimension that holds the list of portfolios we are tracking. In a typical portfolio analysis, an analyst almost always looks at a portfolio's measures along side a benchmark portfolio's measures. A benchmark is another portfolio against which a selected portfolio is compared. So, for a given portfolio, there is a determined benchmark portfolio.

For example, an analyst might want to look at a measure called Annual_Yield for a portfolio called Portfolio_A. There will be an attribute on the member Portfolio_A that tells us what its benchark portfolio is. That attribute will basically be another member from the portfolio dimension. So, as a function of what an analyst chooses as the first portfolio (the subject portfolio), the cube should be smart enought to know what to select as the second portfolio (the benchmark portfolio).

To implement this, I am proposing the following structure for the table that will create the portfolio dimension.

PortfolioID, PortfolioName, Attr1, Attr2, ..., AttrN, BenchMarkID
1, Portfolio_A, blah, blah, ...., blah, 3
2, Porfolio_B, blah, blah, ...., blah, 3
3, Portfolio_C, blah, blah, ...., blah, null
4, Portfolio_D, blah, blah, ...., blah, 3
5, Portfolio_E, blah, blah, ...., blah, 6
6, Portfolio_F, blah, blah, ...., blah, null

So, the above data says that Portfolios A, B, and D uses Portfolio C as their benchmarks. And Portfolio E uses Portfolio F as its benchmark. Lastly, if a portfolio is used as a benchmark for other portfolios, then its BenchmarkID is null, meaning it doesn't get compared to anything, thus Portfolios C and F have null BenchmarkIDs. With this data, I would create a hierarchy in the portfolio dimension that has two levels. At the top level, there will be a portfolio and if that portfolio has a benchmark, then it will drill down to it.

With a portfolio dimension designed as such, I envision a fact table that looks like the following.

PortfolioID, SomeOtherDimIDs,....., Annual_Yield
1, , blah, ,....., 10%
2, , blah, ,....., 7%
3, , blah, ,....., 0%
4, , blah, ,....., 15%
5, , blah, ,....., 20%
6, , blah, ,....., 9%

So, going back to the original report which was looking at Annual_Yield by Portfolio_A, I would do an MDX statement such as

With Set [PortfolioAndBenchmark] As
DESCEDANTS(Portfolio_A)

Select
[Annual_Yield], [PortfolioAndBenchmark] On Columns,
Whatever Other Dimension On Rows
From
[Cube]

Does this make sense? The only concern I have is that I don't want the measures to aggregate along the portfolio hierarchy.