I’m using the association rules algorithm from Analysis Services 2005 to figure out things like: On days that INTC rises significantly, MSFT often rises significantly too. However, I don’t intend to use that knowledge to simply buy MSFT when I notice INTC going up. There’s not much value in that in itself. The value arises in realizing there is a connection between the fortunes of MSFT and INTC. So, if there is a situation where INTC announces its quarterly earnings before MSFT, I’ll have a good clue that MSFT’s quarter will be similar. This won’t work well on stocks as well-known as these two. If INTC announced a great quarter, MSFT will immediately start rising. But perhaps there are pairs that go relatively unnoticed.
I can also create a web of relationships using these pairs. For example, hypothetically, if I see INTC and MSFT go up and down together very often, and I see that INTC and HP also go up and down together very often, there lies a small web: MSFT-->INTC-->HP.
This is the market basket analysis approach. In the normal use of market basket analysis, analysts want to know which products are purchased together (ex: milk and cereal, pizza and beer, steak and Cabernet Sauvignon) in the same "shopping basket". That way, the products could be placed near each other or promotion campaigns could be engineered for the two products. For more on market basket analysis (using OLAP), see Amir Netz's classic article, Analysis Services: DISTINCT COUNT, Basket Analysis, and Solving the Multiple Selection of Members Problem.
Source: Eugene Asahara
No comments:
Post a Comment