ywh19860616 发表于 2012-6-12 20:12 
epoh老师,您可以参见那篇文章进行阅读。
Matlab & Winrats两者结果接近.
in Matlab:
>Launch_PSTR
Initial Conditions : Assumed Number of Thresholds r = 1 Number of Regressions = 270
Initial Conditions on (c,gamma)
5.0000 -0.4222
**************************************
*** FINAL ESTIMATION OF PSTR MODEL ***
**************************************
Final Estimation of the Model with r = 1 and m = 1 by NLS ***
Initial Conditions on (gamma,c) :
1.0e+003 *
5.5412 -0.0003
WARNING FINAL ESTIMATION: At least one estimated Location Parameter is outside the trimming for a PTR model
The location Parameter must range from -1.5713 to -0.3003 for a PTR model
RSS = 7.093 Convergence = 1
AIC = -4.644 BIC = -4.607
Estimated slope parameter of the transition function (one for for each transition function)
5.5802e+003
Estimated location parameters (per column for each transition function)
-0.2537
Estimated slope parameters (per column for each transition function)
0.0295 -0.0450
0.1180 -0.1903
Standard Errors of estimated slope parameters corrected fo heteroskedasticity (per column for each transition function)
0.0067 0.0119
0.0143 0.0482
t-statistics based on corrected standard errors (per column for each transition function)
4.4036 -3.7712
8.2272 -3.9479
%%%%%%%%%%%%%%
in Winrats:
cal(panelobs=36) 1960
open data Data_Demo_PSTR.xls
data(format=xls,org=columns) 1//1960:1 21//1995:1 y q x1 x2
source gtvd.src
@gtvd(m=||1||) y
# x1 x2
# q
ESTIMATION DU MODELE A 1 VARIABLE(s) DE TRANSITION :
Q
valeurs initiales sur c
-1.287990 -1.191639 -1.095289 -0.998938 -0.902588
-0.806237 -0.709887 -0.613536 -0.517186 -0.420835
PREMIERE ETAPE : RECHERCHE DES VALEURS INITIALES DANS LES GRILLES
selection des valeurs initiales optimales dans les grilles :
gamma=
5.000000
c=
-0.420835
objectif initial associ?= 7.28501
nombre de r嶲ressions effectu嶪s : 100
SECONDE ETAPE : OPTIMISATION BFGS A PARTIR DES VALEURS INITIALES PRECEDENTES
FIND Optimization - Estimation by BFGS
Convergence in 40 Iterations. Final criterion was 0.0000000 <= 0.0000100
Function Value 7.0932
Variable Coeff Std Error T-Stat Signif
************************************************************************************
1. GAMMA(1) 2049.1320 234127.3310 0.00875 0.99301683
2. C(1,1) -0.2549 0.0412 -6.19236 0.00000000
Linear Regression - Estimation by Least Squares
Dependent Variable DEPVARC
Panel(36) of Annual Data From 1//1960:01 To 21//1995:01
Usable Observations 748
Degrees of Freedom 744
Skipped/Missing (from 756) 8
Centered R^2 0.1532929
R-Bar^2 0.1498788
Uncentered R^2 0.1532929
Mean of Dependent Variable 0.0000000000
Std Error of Dependent Variable 0.1058994912
Standard Error of Estimate 0.0976414686
Sum of Squared Residuals 7.0931891570
Regression F(3,744) 44.8994
Significance Level of F 0.0000000
Log Likelihood 680.8262
Durbin-Watson Statistic 0.3119
Variable Coeff Std Error T-Stat Signif
************************************************************************************
1. EXPLIC_C(1) 0.029500024 0.004501028 6.55406 0.00000000
2. EXPLIC_C(2) 0.118042817 0.013961525 8.45487 0.00000000
3. EXPLICTRANSF(1) -0.044999529 0.006625826 -6.79153 0.00000000
4. EXPLICTRANSF(2) -0.190267216 0.040569272 -4.68993 0.00000325
Rappels : signification des intitul廥 des variables :
variable expliqu嶪 : Y
variables explicatives :
EXPLIC_C(1) = X1
EXPLIC_C(2) = X2
EXPLICTRANSF(1) = X1 x fonction de transition num.1
EXPLICTRANSF(2) = X2 x fonction de transition num.1
crit鋨es de s幨ection dans ce mod鋩e ?6 param鋈res
AIC = -4.64223
BIC = -4.60519