带跳跃的GARCH模型MATLAB程序

function [LCL,CL] = jumpgarch(u,h,lambda,delta,theta) % Jump GARCH function evaluation procedures. jumpGARCH is a function for a % standard jump GARCH, while ARJIGARCH is for the ARJI-Garch model of Chan and % Maheu(2002), "Conditional Jump Dynamics in Stock Market Returns", JBES, vol 20, % no. 3, 377-389. The names of the components are taken from that paper. % % kmax, which is the maximum value of jumps to be used in the infinite sum, is a % global variable which needs to be set before using either of these procedures. % The default is 20. % % Note that in evaluating the likelihood, rather than use the standard % exp(-lambda) integrating constant for the Poisson, this sums the kernels of the % Poisson into wt and divides by that. In effect, this computes the log likelihood % conditional on the number of jumps being <= kmax. This should make the % estimation less sensitive to the choice of kmax. (Even if the final converged % likelihood is unaffected by a slight change in kmax, it's possible for function % evaluations for large test values of lambda to produce misleading results if the %standard integrating constant is used). % % Revision Schedule: % 08/2008 Written by Tom Doan, Estima % 本程序根据Tom Doan（2008）的Rats程序改编，在此感谢Tom Doan！ % u = current residual % h = base GARCH variance (before jumps) % lambda = Poisson intensity for jumps % deltasq = variance of Normal jump process % theta = mean of Normal jump process % 09/2011 Written by Yu Cong, Hunan University % 设置最大累加次数 kmax=20; T=size(u,1); % 初始化 wt CL wt=0.0; CL=zeros(T,1); % 计算似然函数值 for k=0:kmax jsd = sqrt(h+k*delta.^2); jp = exp(k*log(lambda)-gammaln(k+1)); wt = wt+jp; CL = CL + jp*(2*pi)^(-0.5)*exp(-0.5*((u-k*theta)./jsd).^2)./jsd; end CL = log(CL/wt); LCL = sum(CL); LCL = -LCL;