一个金融数据挖掘问题，期待与大家一起交流讨论

娃娃我很忙

cookoo 发表于 2006-4-6 23:04
金融市场的数据分析一般都是比较弱智的, 也就是按照固定的技术指标和各种理论模型, 这些分析的作用只是辅助 ...

呵呵,那你觉得哪个行业可以用应用数学来描述,所有的因素都是数据化的

nicktoronto

What you really need is a balance scorecard which shows you a group of KPIs indicate how the company or stock are doing.

To rank a company or a stock good or not having many aspects need to be considered, that is why you need a combined 360 degree view on the company or stock, so how to do that?

Here is what I suggest:

1. You need to find out the key measurements about a company or stock: for example: How many Revenue the company can generate each year? What is the customer satisfaction level of this company, etc.? Then, quantify these measurements as true measures like: Year over Year Total Revenue in Dollars, Customer Satisfaction Scores Year over Year etc.

2. Now you need to develop some KPIs, to have a KPI you need to compare your actual vales vs. targets, so you need to understand what kind of target level come be considered good, bad or just OK. There must be some indsutry standards you can follow. For example: if the company's revenue increase every year is over $1 million US dollar that considered good, anything between 1 million to 500K is OK, basically you classify the measures into different buckets. Now your comapny's measure can be compared with these levels. But not every measure has the same weighting effect on the company's future, hence you need to have a weight factor, for example, higher customer sat level is good but it should not overweight than the revenue generating ability for obvious reasons.

3. After that, you can develop your strategic tree view to associat different KPIs with different weighting factor to rank if the company overall is good or bad, and how does it compare with others, etc.

This is just oversimplify the process, but should be able to give you a hint where to go with it.

I had done an implementation for a global fiancial company, basically the company does lots of stock, security, bonds trading on a daily basis. There could be a big effort to achieve this, but the final result will be quite impressive to your users.

Cheers,
Nick