用LIMDEP软件中 极大似然估计法估计参数,当分布函数书对数线性形式时是顺利的,没有error,如下:
Maximize ; Labels = CONS,C_BID
; Start = B ; Maxit = 100
; Fcn =
BVH=LGP(-DOT[Z]-C_BID*LOG(BID2U))|
BVL=LGP(-DOT[Z]-C_BID*LOG(BID2D))|
BVM=LGP(-DOT[Z]-C_BID*LOG(BID1))|
R1*R2*Log(1-BVH)
+R1*(1-R2)*Log(BVH-BVM)
+(1-R1)*R2*Log(BVM-BVL)
+(1-R1)*(1-R2)*Log(BVL)
$
Entering iterative search for function optimizers.
Initial iterations to improve starting values
NOTE: Convergence in initial iterations is rarely
at a true function optimum. This may not be a
solution (especially if initial iterations stopped).
Exit from iterative procedure. 1 iterations completed.
Check convergence values shown below.
Gradient value: Tolerance= .1000D-05, current value= .3009D-06
Function chg. : Tolerance= .0000D+00, current value= .1419D+03
Parameters chg: Tolerance= .0000D+00, current value= .1905D-06
Smallest abs. param. change from start value = .
At least one parameter did not leave start value.
Normal exit from iterations. Exit status=0.
Exit from iterative procedure. 1 iterations completed.
Exit status appears above.
Exit status for this model command is .0
但是当分布函数是线性模型F(B;θ)=Λ(α+βi *xi-β_bid*B)=1⁄(1+exp(-(α+βi* xi-β_bid*B)))时,
Maximize ; Labels = CONS,C_BID
; Start = B
; Maxit = 100
; Fcn =
BVH=LGP(-DOT[Z]-C_BID*BID2U)|
BVL=LGP(-DOT[Z]-C_BID*BID2D)|
BVM=LGP(-DOT[Z]-C_BID*BID1)|
R1*R2*Log(1-BVH)
+R1*(1-R2)*Log(BVH-BVM)
+(1-R1)*R2*Log(BVM-BVL)
+(1-R1)*(1-R2)*Log(BVL)
$
跑出来一篇error:
Error: 590: Obs.= 1 Cannot compute function: Logminus
Error: 137: Iterations: function not computable at crnt. trial estimate
Error: 802: Cannot compute function at start values. Exit status=4.
Function= .70187257136D+02, at entry, .00000000000D+00 at exit
Error: 590: Obs.= 1 Cannot compute function: Logminus
Error: 137: Iterations: function not computable at crnt. trial estimate
Error: 143: Models - estimated variance matrix of estimates is singular
Error: 447: Current estimated covariance matrix for slopes is singular.
求问这些error是为什么……新手第一次发帖,有什么不对的地方请指出来……求大牛解答T。T毕业论文很焦虑啊