Parameter Estimation Algorithm for Skew-normal Joint Location and Scale Models Based on Strong Stable Convergence
In practical applications,traditional iterative algorithms such as the Newton method and EM algorithm often exhibit sensitivity to initial values.To address this issue,a robust convergence algorithm,referred to as the Upper-crossing/Solution algorithm(hereafter called the US algorithm),has been proposed.While this algorithm demonstrates strong stability in solving univariate nonlinear func-tions,it cannot be generalized to multivariate scenarios.Therefore,for multivariate situations,this paper combines the random representation of skew-normal distribution to stratify the likelihood function of the skew-normal joint location and scale models.Additionally,the MM algorithm is utilized to handle the univariate case,followed by the application of the US algorithm to construct a robust convergence algo-rithm.Finally,through Monte Carlo simulations studies and a real example analysis show that the US algorithm significantly reduces sensitivity to initial values compared to the Newton-Raphson method and markedly improves convergence stability.
万方数据
维普
