Hurricane Modeling question

[chinese_girl]

I'm doing some hurricane modeling analysis for a certain states's rate review. We used EQECAT as the primary model provider and use another model for a second opinion.

My question is when the modeling results had a big difference in one state but balanced in most of other states in the east coast. What will you do with the modeling result??? (the consultants from the hurricane modeling companies didn't provide satisfaction answer to the problems, they all said their model is better than the others)

[Franchise]

Do you have any historical results for the state that you could simulate to test the two models? That might be a place to start.

[Sherwin]

I have to say that each cat model is not fully credible. The modelling companies update their models periodically, which means that the models are always wrong before a new cat appears, speaking from a special viewpoint.

Sometimes, how to use the result of a cat model depends on what your purpose is. Actually speaking, if we use the cat model to price a cat coverage, I would like to choose the conservative result from the different models(AIR, RMS, EQE, ect).

Modelling is a hard thing to measure its reasonability, since the modeller could only campare the result of the model to the historical experience data, but nobody can tell us whether the the result of the model is well fit to the future cat data. After all, the future is more concerned, rather than the history.

While, if the future could be forecasted accurately, will the insurnace still be needed? Everything in the world is certain then, no need for coverage for the future. :)

[SamCat]

We used EQECAT as the primary model provider and use another model for a second opinion.

My question is when the modeling results had a big difference in one state but balanced in most of other states in the east coast. What will you do with the modeling result???

What is the underlying reason that a second model is used? Does it stem from a lack of confidence in the EQECAT model or to confirm its validity? What would lend you to believe one model over another?

Unfortunately, due to the uncertainty involved with modeling, it is often difficult to decide which is 'better'. Of utmost importance, however, is consistency when dealing with these issues. What was done in the past? Which model was relied upon when questions arose? It would be difficult to compare future - or prior - results if different methodologies were implemented. Often, it is the changes from period to period - rather than the actual values - that are most significant.

To summarize, see what was done in the past to resolve such issues. Compare the current results based on that methodology to prior data to see whether they seem 'reasonable' (which can also be rather subjective). If significant discrepancies exist from prior results - not attributable to changes in data/input - it may be necessary to reconsider which methodology to use.