Mortality Forecasting with an Age-Coherent Sparse VAR Model

This paper proposes an age-coherent sparse Vector Autoregression mortality model, which combines the appealing features of existing VAR-based models, to forecast future rates. In particular, proposed model utilizes a data-driven method determine autoregressive coefficient matrix, and then employs rotation algorithm in projection phase generate forecasts. estimation phase, age-specific improvement rates are fitted VAR with dimension reduction algorithms such as elastic net. projected assumed follow short-term fluctuation component long-term force decay, will eventually converge age-invariant mean expectation. The age-invariance guarantees projections. is generalized multi-population context computationally efficient manner. Using single-age, uni-sex data UK France, we show that able more reasonable projections, well accurate out-of-sample forecasts than popular models under various settings. Therefore, expected be alternative insurance demographic analyses.