Statistical Inference And Computation Of Change-points In Linear Models

The problem of change-point detection and estimation has always been one of hot issues in the fields of statistics,biology and econometrics.It has been widely used in all aspects of quality control in industrial production.Studying the problem of change-point in linear model is helpful to raise eligibility rate and increase production efficiency.Thereby it improves the relationship between supply and demand.This paper studies the change point problem in the linear model.We mainly consider two cases where there are multiple change-points in linear model and only a single common change-point in linear panel model.For multiple change-points linear model,the large sample properties of the fractional estimator of change points obtained by sequential estimation algorithm are evaluated by least squares method in the case where the number of change-points is known.We then relax this restriction by proposing a parametric jump information model selection criterion(JIC),and prove the consistency of the estimator of the number of change-points obtained by combining sequential estimation algorithm with JIC.Finally,Monte Carlo simulation shows that the method is effective.For linear panel model with a single common change-point,we estimate the common change-point by least squares method.When the number of sequences and the number of observations in each sequence approach infinity,we prove the consistency of the change-point estimator and obtain the corresponding convergence rate.Our theory is verified by Monte Carlo simulation.