Trend and TrendConstant

[kennethshsu]

I wanted to start a new thread and propose that we refactor how we use trends in the package.

I believe we should have two separate functions:

First, chainladder.Trend():

• This should be an estimator, but I haven't really thought out how/what models to use. Some ideas:

1. (1+T*i): linearly, is this even useful?
2. (1+i)^T: multiplicatively (default)

Basically, we want to minimize an error function and return the factor .i_. But what error function to use? MSE?

Second, chainladder.TrendConstant():
This function should take in a few parameters:

• (array of) [start date, end date]
• (array of) factors
• how? to match the chainladder.Trend(), so linearly, and multiplicatively (default)?
• trend_to: date to trend to
  And returns:
• trend_factors, the array of factors that will be applied to the origins

Thoughts?

[jbogaardt]

I like your thought process on this one.