The Estimation Of Panel Data Model With Multiple Structural Change-points Based On Screening And Ranking Algorithm

The panel data combines both spatial and temporal information,which can provide richer empirical information than cross-sectional data or time series data,and is widely used in economics,management,sociology,psychology and other fields.The panel data model may have multiple unknown structural change-points due to various factors,and the detection and estimation of change-points can build a relatively reasonable model to better grasp market liquidity and mitigate market risk.In this paper,we consider the estimation of structural change-points in panel data models based on a screening and ranking algorithm.The main contents of the study are as follows:1.The estimation of multiple change-points in the coefficients of a panel regression model is investigated by a screening and ranking algorithm.Firstly,the covariance estimation method is used to estimate the model coefficients.Secondly,the local statistics are constructed based on the coefficient estimator to preliminarily screen out possible change-point.Finally,the final change-point are screened out through the threshold.When the jumping degree of the change point is large,this method can provide a good estimate.When the jumping degree of the change point is small and the sample size is large,the estimation performance of the above method deteriorates.On the basis of the above local statistics and the thresholds for screening,this paper uses the information criterion to do further screening to for the better estimation results,and the consistency of the estimation of the number and locations of change points obtained by the two screening and ranking algorithms is demonstrated.The two screening and ranking algorithms method is compared with the adaptive group fused Lasso method by a Monte Carlo simulation.The results show that the proposed method outperforms the adaptive group fusion Lasso method in terms of the ratio of the number of correctly estimated change-point;for the average Hausdorff distance between the estimated change point locations and the true change point locations,the proposed method also outperforms the adaptive group fusion Lasso method when the explanatory variables are endogenous.Finally,two real examples are used to illustrate the effectiveness of the method.2.The estimation multiple change-points in interactive effects dynamic panel data are sutdied by a screening and ranking algorithm.Firstly,the coefficients are estimated using the nonlinear instrumental variable estimation method.Secondly,the local statistics are constructed based on the coefficient estimates to initially screen the possible change-point.Finally,the final change-point are screened by thresholding and the consistency of the estimation of the number and locations of change points obtained by the two screening and ranking algorithms is demonstrated.Monte Carlo simulation discusses the screening and ranking algorithm under different sample sizes and different numbers of change-point,and compares it with the penalized principal component estimation method.The simulation results show that the proposed method outperforms the penalized principal component estimation method in terms of the ratio of the number of change-point correctly estimated and the running time;for the average Hausdorff distance,the proposed method does not estimate as well as the penalized principal component estimation method,but the difference between them is not significant.Finally,the effectiveness of the method is illustrated by a real example analysis.