急：R软件加载程序包tsDyn问题 [推广有奖]

131楼

雁茗轩 发表于 2011-10-18 23:03:50

epoh 发表于 2010-9-30 10:23
library(tsDyn)
x

前辈您好！我想请教一下您，在STAR模型中做脉冲响应分析是不是用的bootstrap模拟方法呢？我怎么有点看不懂您在讨论中给出来的那个程序，bootstrap方法不是用来求置信区间的吗？可否详细给我说说，谢谢您啦！

132楼

epoh 发表于 2011-10-19 15:10:56

请先看两篇文献

看完有概念之后我好解释.

第一篇:先了解[TI] & [GI] 之区别

Smooth Transition Autoregressive Models -A Survey of Recent Developments.pdf
page 27/56
traditional impulse response function [TI] (38)
page 28/56
Generalized Impulse Response Function [GI] (41)

In case of the STAR model, analytic expressions for the conditional
expectations involved in the GI are not available for h > 1.
Stochastic simulation has to be used to obtain estimates of the
impulse response measures. See Koop et al. (1996) for a detailed
description of the relevant techniques.

接着再看第二篇的详细步骤:

Impulse response analysis in nonlinear multivariate models.pdf
Gary Koop , M. Hashem Pesaran, Simon M. Potter
5. Monte Carlo techniques for computation of GI
page 17-19/29

133楼

lintao2006 发表于 2011-10-19 21:44:48

epoh 发表于 2010-9-30 10:23
library(tsDyn)
x

epoch你好，请问R有实现LSTVAR的软件包吗？

134楼

epoh 发表于 2011-10-20 20:00:51

请先参考底下两个函数

function TVAR()

Multivariate Treshold Autoregressive model

function TVECM()

Treshold Vector Error Correction model (VECM)

135楼

epoh 发表于 2011-10-25 19:24:06

to:雁茗轩

请务必参考page 17-18/29

Monte Carlo techniques for computation of GI step1-7

同时执行79楼程序,才容易理解.

##########
N=60    #the maximum horizon
R=1000    #the number of replications
noh:112 #no. of histories
1. Pick a history and shock, W(t-1),v(t),by some combination of..
2. For a given horizon N, randomly sample (N + 1)x R values
of the (Kdimensional)innovation
#generate standardized shocks from lstar model
st_shock_idx=sample(seq(1:length(std_res)),R*(N+1),replace = TRUE)
length(st_shock_idx)
st_shock=matrix(std_res[st_shock_idx],R,N+1)
dim(st_shock)  # 1000 x 61
shockz=st_shock*sd_res
3. Use the first N random shocks (obtained under step 2) to
compute the realization.....
4. Use the same draw of N random shocks plus one additional
draw of the random shock to produce a realization...
5. Repeat steps 3 and 4 R times and form the averages...
###
for(i in 1:noh){
hist=hist_m[i,];
# benchmark profile
.....
.....
######compute "new" histories
histv=cbind(1,y,histv[,2],y,histv[,2])
realvb[k]=mean(histv[,4]);
}  #end k shock profile
6. Take the difference between the two averages to form a
Monte Carlo estimate of the GI,
GI[i,]=realvb-realzb;
}# end i numbers of histories