Calculating Trends

[tope]

DATA - Assuming you have incurred claims each month from January 2004 to Present with Paid claims through to Present

QUESTION - How do you calculate annual trends as of current date


PSSIBLE SOLUTION - Do you calculate the PMPM for 2004 and 2005 (current) and then use completing factors to bring them to date?

OR

NEED HELP HERE- Is there another method

Basically I want to calculate annual trends as of current day lets say August 2005 because that when we have our most recent data


Any help would do

Thanks

[hopper]

If you did this (completed present month PMPM)/(present month-12 PMPM)-1 , you may as well as the evaluation actuary what trend he/she uses in setting the IBNR and use it as the trend.

[Soccerboy]

Interesting..my company used that method above (hopper) for IBNR..wonder if there are other more interesting methods out there

[Polly Nomial]

There are many ways to calculate trend and it's best not to rely on any one. I would start with 2 years of data and take pmpm2/pmpm1. For refinements, I'd look to 3, 6 or 12 month rolling averages and fit exponential curves. I suspect that you do not have claims for a large number of lives so any month by itself would not be reliable. The real trick is to look at the calculations and decide which makes sense and which doesn't.

[Dr T Non-Fan]

I suggest year-over-year 3-month rolling averages.

And do note that past trends are not correlated to future trends.

Also check for changes to exposure during trend periods.

[tope]

Originally Posted by Polly Nomial

There are many ways to calculate trend and it's best not to rely on any one. I would start with 2 years of data and take pmpm2/pmpm1. For refinements, I'd look to 3, 6 or 12 month rolling averages and fit exponential curves. I suspect that you do not have claims for a large number of lives so any month by itself would not be reliable. The real trick is to look at the calculations and decide which makes sense and which doesn't.

Sounds like a good idea,

conceptually (mathematically) how do you fit that, do you just plot your rollling averages against the prospective trends then find a pattern?

[hopper]

Do searhes in Excel for the following functions: Logest and Index. The combination of these 2 functions, when applied to your monthly or 3 month moving or 6 month moving or whatever moving, will give you the trend assuming exponential trend. Just do not forget to exponentiate your result to get to the annual trend. Of course, you can always take the log of the PMPMs and do a linear regression on them.

[Polly Nomial]

I use the GROWTH function in Excel. The y values are the PMPM's for each 3/6/12 months. The x values are just linear, ie, month 1,2,3.... When you get the fitted curve - which is nice to graph along side your data points - the trend is [(point n)/(point n-1)]^12

[hopper]

Do not use the Trend or Growth, because they all assume claims are increasing linearly.

[Dr T Non-Fan]

Do not use past trend as future trend. There isn't much alternative, though.