DW and "what if" questions

[brokenbone]

Hi all!

I'm new in data warehousing and I have already read a couple of theoretical literature about data warehousing, OLAP, multidimensional modeling etc.
I've read a lot about how DW is perfect for answering "what if" types of questions, but havent found any practical examples. I have a problem imagining it in practice. Can somebody please take a couple of minutes and show me a small and quick example on what kinds of "what if" questions and how they can be answered.

I have an example of "what if" problem and I don't know what kind of data would be needed for answering it.
Let's say,that a company takes a lot of loans in different ways. Now they would like to know, what will the balance be in future, if interest rate is x%. What about if it is y, z%?
Is this the type of question that can be answered by DW data? Does the DW have to include all the precalculated balances for all possible interest rates or is this done something like this: user enters interest rates and procedures then calculate balance?

I hope I was clear enough, my main question was about a practical example for what if question, maybe from someone that has already done something like that.

Thank you in advance!
BB

 

[johnherman]

While not directly a DW benefit, this analysis can be accomplished using the DW and DW structures. Consider this table structure.

Loan Table
LoanID
LoanDate
LoanAmt
LenderID
CustomerID
LoanPlanID <*

Loan Plan Table
LoanPlanID
InterestRate
RepayWindow
OtherTerms

You will see that you can now allow joins between the Loan table and any one or more of the Loan Plans, thus allowing "what if" analysis.

Hope this helps.


-------------------------
The trouble with doing something right the first time is that nobody appreciates how difficult it was - Steven Wright

 

[brokenbone]

John, thanks for reply.
If I understand you correctly, you mean something like I said. Loan plans must be generated, stored in DW and than user can perform "what if" analysis? I can't use some 'generation' tools where I would use interest rate as parameter or something like that?

Thank you, bye!

 

[johnherman]

You can use any analysis tool you want against the DW as long as the DW has the data to support your tool If the tool can do the Loan calculations and interest rates itself (either off the shelf or custom software), then that's great.

-------------------------
The trouble with doing something right the first time is that nobody appreciates how difficult it was - Steven Wright

 

[blom0344]

A DWH can offer you an additional component no analysis tool can offer: the fact that it will store historical data for trend analysis. The top-end BI tools will allow you to play with parameters to get different scenario's, but they cannot fake trends. You need real DWH data to bring into the picture. In fact, this part of the data-mining approach..

Ties Blom

 

[brokenbone]

Thank you all for replies!

If i understood you correctly I need all the data stored in DW, especially because computing all the payments with different interest rates would take a lot of time to compute. So when a user changes interest rate on report it just displays the data in DW and aggregates are the only thing that has to be computed on the way (that is, if aggregate tales are not stored in DW and precomputed).

Take care,
BB

 

[byk]

DW with BI will allow you to understand the change and the trend when the underlying variables changes. But this is only based on the past data.

This data can then be used for extrapolation either based on the trend analysis or using some heuristic algorithm.

Bear in mind that DW is only the storing of data which makes reporting intuitive and easier

 

[dmcmunn]

historical trend analysis is great for relatively static systems (home loans, major appliance orders, banking) because those systems change incrementally. Where they fail and breakdown totally is in finding order in highly dynamic ("chaotic") systems, such as weather, stock market, massively multi-player online gaming, webserver usage. Almost all traditional BI tools shoot for the historical trending, but few adequately address much more complex nature (multi-variate pattern identification/recognition and true perturbence modeling) referred to as data mining (as mentioned above). To work with chaotic systems will require the statistically significant samples of high quality, accurate, cross-referenced data (best provided by a data warehouse system) combined with modeling/mining applications which use combinations of neural networking, advanced multivariate statistical applications capable of identifying and presenting patterns for feedback by the expert user (for probability assignment/discard) or near-realtime interaction with a scenario-based modeling ("what-if") application capable of playing out multiple models in parallel nad learning from its successful models. But then who defines what success is in a chaotic model? For now that is still US, and not the machines, so train up young Neo!