你好:我用上一楼的数据和方法,得到了结果确不同与他的,我的tsDyn是9.o版的。结果如下:
>svpdx <- read.table("data.txt", header = TRUE)
>svpdx
cpi rpi m2
1 3.4 2.9 0.26524514
2 6.4 5.4 0.31278198
3 14.7 13.2 0.37310154
4 24.1 21.7 0.34529154
5 17.1 14.8 0.29467111
6 8.3 6.1 -0.03867458
7 2.8 0.8 0.34668927
8 -0.8 -2.6 0.17245702
9 -1.4 -3.0 0.14411363
10 0.4 -1.5 -0.31433668
11 0.7 -0.8 0.87572994
12 -0.8 -1.3 0.16337324
13 1.2 -0.1 0.19679602
14 3.9 2.8 0.18161695
15 1.8 0.8 0.14387323
16 1.5 1.0 0.18886459
17 4.8 3.8 0.15787746
18 5.9 5.9 0.18875571
19 -0.7 -1.2 0.18736354
> x=svpdx$cpi
> y=svpdx$rpi
> z=svpdx$m2
> library(tsDyn)
>ndx.lstar <- lstar(x, m=3,d=1, thVar=z,control=list(maxit=3000))
(使用此命令得到的结果与你的上述结果有很大有不同,特别是gamma值= 330.0193)结果如下:
Using maximum autoregressive order for low regime: mL = 3
Using maximum autoregressive order for high regime: mH = 3
Using only first 16 elements of thVar
Performing grid search for starting values...
Starting values fixed: gamma = 100 , th = 0.2919505 ; SSE = 103.1979
Grid search selected lower/upper bound gamma (was: 1 100 ]).
Might try to widen bound with arg: 'starting.control=list(gammaInt=c(1,200))'
Convergence problem code 1. You might want to increase maximum number of iterations by setting 'control=list(maxit=1000)'
Optimized values fixed for regime 2 : gamma = 330.0193 , th = 0.2820525 ; SSE = 90.99445
> summary(ndx.lstar)
Non linear autoregressive model
LSTAR model
Coefficients:
Low regime:
const1 phi1.1 phi1.2 phi1.3
-0.7398994 1.8526579 -1.2560973 0.7310564
High regime:
const2 phi2.1 phi2.2 phi2.3
2.7868701 -1.0567428 1.0397095 -0.8612935
Smoothing parameter: gamma = 330
Threshold
Variable: external
Value: 0.2821
Residuals:
Min 1Q Median 3Q Max
-5.95810 -1.28246 -0.17115 1.28210 4.00887
Fit:
residuals variance = 4.789, AIC = 50, MAPE = 132%
Coefficient(s):
Estimate Std. Error t value Pr(>|z|)
const1 -0.739899 0.995576 -0.7432 0.457368
phi1.1 1.852658 0.257118 7.2055 5.784e-13 ***
phi1.2 -1.256097 0.581746 -2.1592 0.030836 *
phi1.3 0.731056 0.318841 2.2929 0.021856 *
const2 2.786870 2.025909 1.3756 0.168941
phi2.1 -1.056743 0.347791 -3.0384 0.002378 **
phi2.2 1.039710 0.661091 1.5727 0.115784
phi2.3 -0.861293 0.390391 -2.2062 0.027368 *
gamma 330.019336 315.129183 1.0473 0.294984
th 0.282053 0.013939 20.2354 < 2.2e-16 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Non-linearity test of full-order LSTAR model against full-order AR model
F = 0.11978 ; p-value = 0.94616
Threshold
Variable: external>
我的问题是: 1、这是多变量的LSTAR模型(平滑转换)。为什么在LSTAR中体现不同系列的不稳性问题呢,或者是不需要作平稳性检验??
2、我们知道在对作平滑转换模型前,要对基本的回归作非线性检验。这里看不到这一点;是不是不需要这一过程,或含在过程中!
3、过程中说:Performing grid search for starting values.也即晶格搜索过程;我不知道这一过程的机理,你能不能写出这个过程的程序来;如何选择的!?
4、这一转换变除X以外的变量z。如是转换变量是Z的滞后项时,又如何办法。是不是程序中也对转换烃量Z及其滞后有选择!如果没有的话,我们要作出是Z的滞后项时,又如何选择呢?
谢谢!