Actuarial consistency

[Gareth Keenan]

It's my 1000th post! I wanted to make it about something more substantial than an exam question.

As the subject indicates, I would really like to know what the community thinks about actuarial consistency. This started where I was performing a quarterly analysis, and had selected development factors as part of the process. Since the last time I had performed this analysis, there had only been an additional quarters worth of data, which generally is not much. But, being inundated with other projects, I reselected development factors from scratch(not following my prior methodology), and low and behold, the result was materially different. This does not intuitively make sense, because why should your results change materially based on one additional quarter's worth of data, when I normally look at 12 for this project. Further, I would have known prior to completing the analysis if there was a shock loss event or if there were any very large individual losses.

But it really got me to thinking, when can a practicing actuary just change their mind(to be inconsistent)? During the course of module 2, I am informed a claims adjuster is advised against stair stepping. Stair stepping is where the claims adjuster will incrementally increase the reserves without new information. The idea is that a claims adjuster should only alter the reserve amounts when there is new information like amounts have been paid, new medical estimates come in, or replacement costs are different than originally anticipated.

But can an actuary just outright change their mind?

For example, say I've been performing a different analysis for some time and I choose to fit a lognormal distribution to the data, and I produce a result. Then say another qualified actuary is called on to perform the same analysis with the same data and chooses to use a gamma distribution for the same purpose, and produces a materially different result. Assume the other actuary chooses the gamma distribution based on their experience, knowledge of theory, and what they think is most appropriate based on the prior two.

Now, am I wrong to just decide you know what, I think the gamma distribution is actually more appropriate. There is no new information, and the reasons for and against using the lognormal or gamma distribution remain unchanged, but now I just feel that I am more convinced by the arguments for the gamma distribution. Assume there are good reasons to use either distribution, so they are both defensible.

I just don't have a good way to wrap my head around how constrained I am by my previous practice. I know the ASOP on ratemaking wants to not discourage new and potentially better approaches to ratemaking, so I believe there is some precedent for trying a new approach, but trying a new approach based on new information is not what I'm asking about.

Does my above question make any sense? Any one have an opinion on when you can just, change your mind?

Cheers!

Gareth Keenan

[MountainHawk]

I do quarterly reserve analysis. Right now, my templates bring in the picks from the prior year review (so April 2012 looks at April 2011) and will light up a selection that doesn't make sense (current year < prior selection, but new selection > prior selection) for example.

I don't have it yet, but I would like to add the prior quarter as well (April 2012 looks at January 2012) and have it check the selected development * the actual 3 months development compared to the prior selection, and light up a warning if it's way off.

That said, I do often override these warnings, because I change my mind about things like the impact of a process change, or I've talked about the line with someone and there are reasons for anomalies.

It's just good practice to document when you are making changes.

[Avi]

I don't work in reserving, but FWIW, ASOP 43 ( Property/Casualty Unpaid Claim Estimates) 3.6.1 may relate to your question:

In the case when the unpaid claim estimate is an update to a previous estimate, the actuary may choose to use the same methods or models as were used in the prior unpaid claim estimate analysis, different methods or models, or a combination of both. The actuary should consider the appropriateness of the chosen methods or models, even when the decision is made not to change from the previously applied methods or models.

The idea being, I believe, that if you have a good explanation (in your experience gamma predicts as well or better than log-normal, for example) you can change, as long as it is documented.

[tommie frazier]

Originally Posted by Avi

I don't work in reserving, but FWIW, ASOP 43 ( Property/Casualty Unpaid Claim Estimates) 3.6.1 may relate to your question:


The idea being, I believe, that if you have a good explanation (in your experience gamma predicts as well or better than log-normal, for example) you can change, as long as it is documented.

+1

[whisper]

Originally Posted by Gareth Keenan

This started where I was performing a quarterly analysis, and had selected development factors as part of the process.

Development factors can have a huge impact on the analysis, because any small change can be compounded through the estimation.

But it really got me to thinking, when can a practicing actuary just change their mind(to be inconsistent)?

New information may get you to change your mind in the results.
1.) Document why you changed your mind.
2.) Discuss with the client/stakeholders about what is going on. Confirrm that what you think is going on is truly happening, and you're not reacting to some random element that has no predictive power.

Now, am I wrong to just decide you know what, I think the gamma distribution is actually more appropriate. There is no new information, and the reasons for and against using the lognormal or gamma distribution remain unchanged, but now I just feel that I am more convinced by the arguments for the gamma distribution. Assume there are good reasons to use either distribution, so they are both defensible.

If you think the argument for the gamma is better than the argument for the lognormal - then changing is appropriate. If you think the lognormal is better, you can keep it. There is a chance that two actuaries looking at the same results will come up with different conclusions.