# Determine and filter Credibility of data

1. Dec 29, 2015

### Semidevilz

I'm doing some work and don't really know if there is a better way to proceed. I have vendor information of how much premium they bring in, and how much losses they incur. With those 2 figures, I can get a 'loss ratio' of each vendor. My job is to look through the list and Identify any 'problem' vendors. I,e consistently high loss ratios over a 3 year period etc.

The issue is that since loss ratio depends on premiums, I might have a high loss ratio just because of low premiums. So an 800% loss ratio on 80,000 premium paints a different picture than an 800% loss ratio on 80 premium.

So with that in mind, is there a way for me to filter out the data that is not credible? Should I just get an average of the premiums and then analyze only the premiums above the average?

2. Dec 29, 2015

### Stephen Tashi

Do you have only that information? Or do you also have more detailed information?

As I interpret that definition of a problem vendor, such a vendor consistently underestimates the risk of loss. There could be other types of problems with vendors. For example, a vendor might be good at assessing risk but not good at making sales.

To formulate a mathematical problem we have be specific about what is different about the two pictures. I don't know the business that is involved. Just guessing, I visualize a "agent" selling insurance policies. His skill at assessing risk is exercised each time he sells a policy, so his net result not only depends on the premium of the policy that is paid but on the number of policies he sells since he must exercise his skill on each sale. In the real world of insurance, I don't know whether an agent who sells policies actually has much choice about what to charge for them. Perhaps the agent's only choice is to put the client is one of several categories. And there might not be much subjective choice involved in doing that.

Suppose the insurance company incorrectly assesses the risk of a certain policy P1 and an agent sells a lot of this policy. Then the "fault" of the agent is doing a good job of selling the policy.

In the model, I have described, I think assessing the skill of the agent well would involve getting into the details of whether the company that writes the policies is assessing risk well. If policy P1 tends to loose money for the company regardless of which agent sells it then is a particular agent to be regarded as a problem if he sells a lot of that policy?