方老师:
在“协整向量检验案例分析的问题”一课中,我想到一个问题,请帮忙解答,谢谢。
原程序:
uscn.f = lexrates.dat[,"USCNF"]##提取远期汇率
uscn.s = lexrates.dat[,"USCNS"]##提取近期汇率
uscn.ts=seriesMerge(uscn.s,uscn.f)
# determine lag length using AIC
var.fit = VAR(uscn.ts,max.ar=6,criterion="AIC")#构建VARP模型,最大 的阶是6阶,
var.fit$ar.order#提取最合适的阶是2,那么,VECM模型最大的阶是1.
coint.rc = coint(uscn.ts,trend="rc",lags=1)#提取最合适的阶是2,那么,VECM模型最大的阶是1.
#也就是lags=1,trend="rc,表示是用约束常数项的模型,即第二种模型
结果:
-----------------------------------------------------------------------------------------
> summary(coint.rc)
Call:
coint(Y = uscn.ts, lags = 1, trend = "rc")
Trend Specification:
H1*(r): Restricted constant
Trace tests significant at the 5% level are flagged by ' +'.
Trace tests significant at the 1% level are flagged by '++'.
Max Eigenvalue tests significant at the 5% level are flagged by ' *'.
Max Eigenvalue tests significant at the 1% level are flagged by '**'.
Tests for Cointegration Rank:
Eigenvalue Trace Stat 95% CV 99% CV Max Stat 95% CV 99% CV
H(0)++** 0.0970 32.4687 19.9600 24.6000 24.8012 15.6700 20.2000
H(1) 0.0311 7.6675 9.2400 12.9700 7.6675 9.2400 12.9700
---------------------------------------------------------------------------------------------
据此,我们判断uscn和uscf存在协整关系。
如果我们用:
##########
coint.nc = coint(uscn.ts,trend="nc",lags=1)
summary(coint.nc)
##
----------------------------------------------------------------------------------------
> summary(coint.nc)
Call:
coint(Y = uscn.ts, lags = 1, trend = "nc")
Trend Specification:
H2(r): No constant or trend
Trace tests significant at the 5% level are flagged by ' +'.
Trace tests significant at the 1% level are flagged by '++'.
Max Eigenvalue tests significant at the 5% level are flagged by ' *'.
Max Eigenvalue tests significant at the 1% level are flagged by '**'.
Tests for Cointegration Rank:
Eigenvalue Trace Stat 95% CV 99% CV Max Stat 95% CV 99% CV
H(0)++** 0.0680 17.6212 12.5300 16.3100 17.1061 11.4400 15.6900
H(1) 0.0021 0.5150 3.8400 6.5100 0.5150 3.8400 6.5100
Unnormalized Cointegrating Vectors:
也说明存在协整关系
----------------------------------------------------------------------------------------
问题:用nc模型检验也通过,用rc模型也通过。为什么不选nc模型而用rc模型呢?以后碰到类似的问题应该怎么选、选哪个模型做VECM呢?