経済学書専門出版 エコノミスト社
Top書籍情報計量経済・統計ソフトウェアLIMDEP V8.0NLOGIT3.0
ファイナンス大系
リアルオプション
Excelとその応用
統計学大系
経済学大系
e-ビジネス
計量経済学
ゲーム理論
経済数学
環境問題・環境経済学
人口学講座
ナレッジマネジメント
ビジネス書
NLP理論
複雑系経済学
経営学・商学大系
マーケティング
心理学・行動科学大系
金融工学・数理ファイナンス
マクロ経済学
法情報学
企業法学講座
経営工学大系
計量経済・統計ソフトウェア
オペレーションズ・リサーチ
会計学・簿記・税務
楽しい数学
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

NLOGIT 3.0

A New Program for FIML Estimation of Nested Logit Models

NLOGIT is a full featured, extremely flexible estimator for the nested logit model. To our knowledge, this is the only FIML estimator for this model that is available as a procedure in an integrated econometrics package,instead of as a specialized, one purpose program. NLOGIT is an add-on feature of LIMDEP and is sold separately. It is not a "stand-alone" program. Among the features of NLOGIT are:

* As part of LIMDEP, all features of the program are available.
* Trees may have up to 85 choices, 25 branches, 10 limbs, and 5 trunks.
* Inclusive value parameters are free to vary across branches and limbs, or may be constrained.
* Data may be in the form of individual choices, market shares, frequencies, or ranks. The program detects this internally.
* Variables that interact with choice specific binary variables need not be created; they are easily set up in the model specification itself. (See the example below.)
* Choice sets may vary across individuals. A simple index variable is used to indicate choice availability.
* For specification analysis, any parameter in the model may be fixed at a value you specify as part of the model.
* Marginal effects, in the form of derivatives or elasticities, are computed automatically and presented in tabular format.
* Box-Cox and log transformations may be built into the model.
* Nonnested models may be estimated as intermediate results for comparison and to obtain starting values. They may also be estimated in order to use sequential estimation procedures instead of FIML.
* Predicted probabilities and inclusive values (log sums) for all choices are kept using LIMDEP's usual setup for predicted values.
* Data format is the same as in earlier versions of LIMDEP, so no reformatting is necessary.
* Since the data set will have the format of a panel in LIMDEP's standard format, all of the special features in LIMDEP for manipulating panel data are available. For convenience, data may also be condensed into a single line.
* Several modeling frameworks are included in NLOGIT, including:
* Multinomial logit and simple discrete choice.
* Multinomial probit with up to 20 choices.
* Heteroskedastic extreme value. This model relaxes the IIA assumption in a single level model.
* Covariance heterogeneity. The inclusive value parameters can vary with attributes or characteristics.
* Random parameters logit model.


An example: NLOGIT is extremely flexible and allows many different kinds of model specifications. The core model is specified algebraically. A simple example might appear as follows, in the format of the LIMDEP command: The tree structure is

Travel

┌────┴────┐
│         │
Public       Private
│         │
┌──┴──┐   ┌──┴──┐
│     │   │     │
Bus    Train  Car    Plane


Using Version 6.0 of LIMDEP, you would fit this model sequentially with two model commands, first at the lower level to obtain the inclusive values and second at the higher level to estimate the rest of the parameters. With NLOGIT, the entire model is estimated at once, for example, with:


NLOGIT ; Lhs = Mode
; Choices = Air,Train,Bus,Car
; Tree = Travel [ Public (Train,Bus) , Private (Car,Plane)]
; MODEL: U(Bus) = Ab + Bc * Cost + Bt * Time /
U(Train) = Bc * Cost + Bt * Time /
U(Car) = Ac + Bc * Cost + Bt * Time /
U(Plane) = Ap + Bc * Cost + Bt * Time /
U(Public,Private) = <0 , C> * Income $

Note in the last line that the income variable is interacted with a branch specific constant by specifying that the coefficient on income in the utility function is zero. The choice specific dummy variables in the alternative specific utility functions are specified just by including a unique parameter in the equation. (They could be constrained to be equal for some or all of the equations just by using the same name for the parameter.) Utility functions for the alternatives can be specified completely separately, or with cross equation equality constraints imposed just by using the same symbol for the parameter in the different equations. Other options are provided for fixing parameters, providing starting values, constraining inclusive value coefficients, and so on. (There are also many shortcuts for writing out the model.)


LIMDEP V8.0のTopに戻る