|
|
|
 |
|
|
| LIMDEP V8.0 |
|
 LIMDEP Windows 95/NT版
スペック
LIMDEP7.0の機能と特徴
NLOGIT3.0
Applications
LIMDEP V8.0のTopに戻る
|
Applications
LIMDEPの応用分野
A PROGRAM TO FIT A PROBIT MODEL
When a model is not contained in LIMDEP's menu of procedures, an alternative method is to write the iterative program using LIMDEP' s programming tools. The following general procedure would estimate the parameters of any specified probit model:
proc=probit(x,y)
calc ; k= col(x)
matrix ; a = [k_0]
create ; z = 2 * y-1
label ; 100
create ; vi = a'x
; li = log(phi(z*vi))
; gi = z*n01(z*vi)/phi(z*vi)
; hi = gi*(vi+gi)
calc ; l = sum(li)
matrix ; g = x'gi
; h = < x'[hi]x >
; e = h * g
calc ; list ; converge = e'g
matrix ; a = a + e
goto ; 100 ; converge > .0001
matrix ; stat (a,h)
endproc
namelist; program = one,gpa,tuce,psi
execute ; proc=probit(grade,program)
|
$
$ number of variables
$ starting values
$ will be convenient
$ beginning of iteration
? argument
? log-li(i)
? 1st derivative
$ 2nd derivative
$ log-l
? gradient
? inverse of hessian
$ change vector
$ convergence rule
$ update for iteration
$ iterate or exit
$ report results
$ procedure now defined
$ variables on the rhs
$ execute the procedure
|
MARGINAL EFFECTS IN A BOX-COX MODEL
This application computes the impact of different functional forms on a marginal effect in the probit model. In the probit model, the income variable is transformed by different values for the Box-Cox transformation. The MLE of the marginal effect of income on the probability of loan default is plotted against lambda. (Greene and Seaks, Applied Economics, 1994.)
read ; data set is entered ... $
create ; lincome = 0 $ (place holder)
namelist; x=one, family, grad, lincome$
matrix ; { i = 0 }; lambda=[20_0]; effects=[20_0]$
proc
create ; lincome = income @ 1$
probit ; lhs=default; rhs=x$
matrix ; xbar = mean (x) $
calc ; i=i+l; m_e = b(4) * nOl (xbar'b)
*xbr(income)@(l-1) $
matrix ; lambda (i) = I; effect(i) = m_e $
endproc
execute ; 1 = -.50, 1.41, .10$
mplot ; lhs = lambda; rhs = effect; fill; grid $
|
COMPUTING BINARY CHOICE MODELS
The following obtains parameters for probit and logit models, displays marginal effects for the models, then produces an output table that allows easy comparison of the two models:
probit ; lhs = grade; marginal effects
; rhs = one,gpa,tuce,psi $
logit ; lhs = grade; marginal effects
; rhs = one,gpa,tuce,psi $
review
|
Partial derivatives of E[y] = F[*] with respect to the
vector of characteristics. They are computed at the
means of the xs. Observations used for means are all obs.
Variable Coefficent Standard Error z = b/s.e. p[|Z| > z]
constant -2.4447 0.75885 -3.222 0.00127
gpa 0.53335 0.232246 2.294 0.02177
tuce 0.16969E-01 0.27120E-01 0.626 0.53150
psi 0.46791 0.184764 2.494 0.01265 |
Binary Choice Models
Probit Model Logit Model
Variable Parameter t-ratio Parameter t-ratio
constant -7.4522 -2.93 -13.0213 -2.64
gpa 1.6258 2.34 2.8261 2.24
tuce 0.0517 0.62 0.0952 0.67
psi 1.4263 2.40 2.3787 2.23
log-l -12.8188 -12.8896
log-l(0) -20.5917 -20.5917
|
|
| LIMDEP V8.0のTopに戻る |
|
|
