Time Series of Continuous Proportions

摘要: SUMMARY A vector of continuous proportions consists the some total accounted for by its constituent components. An example is world motor vehicle production Japan, USA and all other countries. We consider situation where time series data are available interest focuses on rather than actual amounts. Reasons analysing such times include estimation underlying trend, effect covariates interventions, forecasting. develop a state space model proportions. Conditionally unobserved state, observations assumed to follow Dirichlet distribution, often considered be most natural distribution simplex. The follows conjugate which introduced here. Thus model, although based does not have restrictive independence properties. Covariates, trends, seasonality interventions may incorporated in way. has worked well when applied several examples, we illustrate with components production.