Calendar effect and in-sample forecasting (Q2656985)

The authors consider a generalization of the chain ladder methodology. Instead of the assumption $f(x,y) = h(x) g(y)$ the authors introduce an extended version based on the density $f(x,y) = h(x) g(y) k(x+y)$ with a so called calendar effect $k$. Two main examples are outstanding liabilities in insurance ($x$ time point of claims, $y$ time delay up to reporting the claim) and fertility analysis ($x$ birth date of the mother, $y$ age of the mother at time of birth of a child). In section 2 the authors give conditions such that $h,g,k$ are uniquely determined, section 3 describes the estimation procedure, section 4 the forecasting method (by extrapolating the function $k$), section 5 contains 3 theorems of theoretical properties of the estimators. Section 6 contains case studies of the two main examples: Claims from motor business in Cyprus and analysis of fertility rates with data from Italy and US. The final section 7 contains simulation results.