Day 429

My day at work started with a meeting.  It was a follow up to a meeting that I attended last month to discuss some rate changes for about 8 different states.  After the meeting was over Lee and I went to another conference room and he explained some of the mathematical theory behind how rates are chosen.  One of the big issues that every insurance company has to handle is the matter of how reliable the data is.  For instance, if a company covers 10 and 8 of them all happen to crash into each other at the same time... was that a coincidence or are 80% of the cars really going to get in wrecks?  Probably not.  It would be accurate to say that you don't have enough data to make a very good assessment.  This is a completely fair and standard procedure.  The issue becomes: what next?  Well.... in a lot of applied statistics, our goal is to be able to fit the data to a normal curve.  Normal curves are very nice and have amazing properties.  Thanks to the Mean Value Theorem, the only real challenge is to get enough data points to accurately (and objectively) state that our data does, in fact, fit a normal curve.  In our case, we need 1082 claims to say that our data is good enough.  This seemingly arbitrary number is determined when we figure out how accurate we want to make our data.  We can never 100% guarantee our data... but we can get close.  The issue is that getting 99.999% accuracy requires a lot of data that nobody has.  But industry standard says that 90% or 95% or so good enough.  That's where the 1082 applies: if we can get 1082 claims, we can guarantee that our data is 90% accurate.  It's pretty neat how the math works out, but I'll leave that for some other time (actually, I doubt I'll ever get into it).  So after all of that excitement, I went back to entering data into Excel models.
In other news, I get married in less than 9 days.  Let's hope it all comes together!


After work I came home and ate some rice.  Then I played games with friend Miles.  We won all of the things and saved the world from all the bad guys.


Then I went to Steak and Shake with friend Kelly!  It was much fun.  She was a trooper and wanted to hear about the sort of things I do at work.  So I did my best to explain some of the general concepts that were discussed in the meeting this morning.  Obviously I couldn't give actual details, but I tried to explain a bit about the reasons of why rates change and how we can't just change numbers all willy-nilly.  Then I did my best to explain some of the ideas behind "credible data" (what I was explaining up above) and how we can determine what is or is not credible.  This is one of those areas for which I have a difficult time giving good explanations because it's still so new to me that I haven't yet come up with a good explanation for people that aren't actuaries or mathematicians.  I mean... unless my saying that "standard practice assumes a normal distribution such that P{ -k*E[x] < S < k*E[x]} = p where k follows a Poisson distribution with some mean E[l] and p is (in essence) our confidence interval and k follows a stochastic process" would really fit well into a good explanation (which I don't think it does) then I need to work on something a little better.  So until I come up with something better.... we use magic.


Anyhow, dinner was fun.  We talked about work and weddings and friends and how amazing the milkshakes are at Steak and Shake.  'Twas a good time.


And now... bed.  But only two more early mornings for me!  Woo!


Good night!