Single Exponential Smoothing
Hi. What am I doing wrong in the screenshot below? 2018 is too high, no?
Howdy, Stranger!
Categories
- 241 All Categories
- 21 GENERAL
- 6 Practice Exam 1
- 6 Practice Exam 2
- 1 CAS Hybrid Exam
- 12 GLM.Basics
- 1 GLM.ModelBuilding
- 1 GLM.DataPrep
- 6 GLM.ModelSelection
- 4 GLM.Interactions
- 7 GLM.ModelRefinement
- 8 GLM.Validation
- 1 GLM.Comments
- 1 GLM.Variations
- Holmes.Intro
- Holmes.PenalizedRegression
- Holmes.Transformations
- Holmes.BiasAndVariance
- Holmes.LassoCredibility
- Holmes.Diagnostics
- Holmes.CaseStudy
- Holmes.Appendices
- 8 Fisher.RiskSharing
- 6 Fisher.OtherLSPlans
- 5 Fisher.AggExcess
- 19 Fisher.Visualization
- 4 Fisher.TableM
- 5 Fisher.LimitedTableM
- 7 Fisher.TableL
- 1 Fisher.Understanding
- 12 Fisher.CaseStudy
- 14 Bahnemann.Chapter5
- 10 Bahnemann.Chapter6
- 5 Bailey.Simon
- 5 Couret.Venter
- 5 NCCI.ExperienceRating
- 8 Mahler.Credibility
- 8 NCCI.Circular
- 1 NCCI.InformationalExhibits
- 1 ASOP.12
- ASOP25.Credibility
- 3 Bernegger.Exposure
- 5 Clark.Reinsurance
- 3 Clark.PropPerRisk
- 2 Clark.CasPerOcc
- 1 Clark.AggModels
- 2 Clark.PropertyCAT
- Grossi.Overview
- 1 Grossi.CATModels
- 4 Grossi.Curves
- Robertson.HazardGroups
Comments
You're not doing anything wrong. When Z is relatively small, we will end up putting more weight on the oldest year because 1-Z is large so (1-Z)^(n-1) doesn't go to 0 fast. It's very possible for the oldest year to receive a large amount of credibility.
We're not necessarily used to seeing this which is why the wiki/Mahler suggests it's common/appropriate for the last year to be an independently derived complement instead. For example, when pricing private passenger auto in a state you may use the latest three years of business in the state and then an independent countrywide or regional complement instead of adding a fourth year of state data.