向量误差修正模型案例分析
生成数据
set.seed(12345) u1<-rnorm(500) u2<-arima.sim(list(ar=0.6),n=500) #生成模拟的一阶自回归模型 u3<-arima.sim(list(ar=.4),n=500) y1<-cumsum(u1) #生成随机游走序列 y1 y2<-0.4*y1+u2 y3<-0.8*y1+u3 #调用urca包中的ca.jo()对时间序列y1 y2 y3进行Jonhansen协整检验 #2.Jonhansen协整检验 library(urca) data<-data.frame(y1=y1,y2=y2,y3=y3) #将变量组织为数据框
## ca.jo(x, type = c("eigen", "trace"), ecdet = c("none", "const", "trend"), K = 2,spec=c("longrun", "transitory"), season = NULL, dumvar = NULL) 注意这里只是用默认设置。
model.vecm<-ca.jo(data)
head(model.vecm@x) #ca.jo使用S4方法,故用@提取变量
y1 y2 y3
[1,] 0.5855288 -0.31135095 -1.0377854
[2,] 1.2949948 0.59430322 -0.5116634
[3,] 1.1856915 1.28751444 -0.1316301
[4,] 0.7321943 1.64792194 0.7132483
[5,] 1.3380818 0.09367809 1.3288343
[6,] -0.4798742 -0.61468043 0.1199645
#使用slotNames()显示模型包含的全部对象类型
slotNames(model.vecm)
[1] "x" "Z0" "Z1" "ZK" "type" "model" "ecdet"
[8] "lag" "P" "season" "dumvar" "cval" "teststat" "lambda"
[15] "Vorg" "V" "W" "PI" "DELTA" "GAMMA" "R0"
[22] "RK" "bp" "spec" "call" "test.name"
summary(model.vecm)
######################
# Johansen-Procedure #
######################
Test type: maximal eigenvalue statistic (lambda max) , with linear trend
Eigenvalues (lambda):
[1] 0.222707791 0.167079305 0.007684667
Values of teststatistic and critical values of test:
test 10pct 5pct 1pct
r <= 2 | 3.84 6.50 8.18 11.65
r <= 1 | 91.04 12.91 14.90 19.19
r = 0 | 125.47 18.90 21.07 25.75
Eigenvectors, normalised to first column:
(These are the cointegration relations)
y1.l2 y2.l2 y3.l2
y1.l2 1.0000000 1.000000 1.0000000
y2.l2 -0.2355148 -5.064504 -0.1799248
y3.l2 -1.1315152 1.143660 -0.1993207
Weights W:
(This is the loading matrix)
y1.l2 y2.l2 y3.l2
y1.d 0.05151358 0.002693258 -0.008416933
y2.d 0.11164178 0.075923301 -0.002918384
y3.d 0.51768302 -0.015197036 -0.006078055
从统计检验值可以看出,在r为2时接收原假设,即认为协整向量的秩为2
使用cajorls()估计VECM模型的系数矩阵
cajorls(model.vecm,r=2) #估计VECM模型 ,cajorls(z, r = 1, r # eg.number = NULL),其中r为协整向量的秩
$rlm
Call:
lm(formula = substitute(form1), data = data.mat)
Coefficients:
y1.d y2.d y3.d
ect1 0.05421 0.18757 0.50249
ect2 -0.02577 -0.41081 -0.04496
constant 0.08575 0.28405 -0.02490
y1.dl1 0.01915 0.22922 0.47143
y2.dl1 0.01735 -0.38180 0.04342
y3.dl1 -0.03389 -0.05120 -0.58422
$beta
ect1 ect2
y1.l2 1.000000e+00 0.0000000
y2.l2 2.775558e-17 1.0000000
y3.l2 -1.242478e+00 -0.4711494
计算结果中,第一部分(rlm)给出误差校正矩阵、常数项及差分解释变量的估
计值;第二部分为标准化后的协整向量矩阵。
估计的误差修正模型为:
VECM模型转化为水平VAR模型
library(vars)
model.var<-vec2var(model.vecm,r=2) #获取与VECM模型等价########的VAR模型估计
model.var
Coefficient matrix of lagged endogenous variables:
A1:
y1.l1 y2.l1 y3.l1
y1 1.0191535 0.01734796 -0.03389437
y2 0.2292213 0.61819733 -0.05119624
y3 0.4714272 0.04342308 0.41578278
A2:
y1.l2 y2.l2 y3.l2
y1 0.03505334 -0.04312019 -0.02131386
y2 -0.04165626 -0.02900446 0.01170232
y3 0.03105876 -0.08837964 -0.01892923
Coefficient matrix of deterministic regressor(s).
constant
y1 0.08574980
y2 0.28405415
y3 -0.02490038
计算结果表明,与VECM模型等价的VAR模型估计为: