readme.txt

Rice University的Robin C. Sickles professor开发的专门用于paneldata model test and estimation 的program

Last Update: September 2, 2004


1.   The following MATLAB programs implement panel data estimation.
     PDE.m is the main program, and refer to the notes in it for the usage
     of the programs. 


2.   The estimators are briefly described in the paper, 

     "Panel Estimators and the Identification of Firm-Specific Efficiency
     Levels in Parametric, Semiparametric and Nonparametric Settings"

     by Robin C. Sickles, Department of Economics, Rice University, 2003
     
     
3.   The programs are provided for academic uses only.
     Users of the program should cite the aforementioned paper.


4.   Computed Estimators

       * Basic Panel Data Estimators
     (1) FR: Fixed and Random Effects Estimators (FIX, RND)
     (2) HT : Hausman-Taylor Estimator 

       * Semiparametric Efficient Estimators
     (3) PSS1: Regressors are correlated with effects
     (4) PSS2: AR(1) error (Within:PSS2W, GLS:PSS2G)
     (5) PSS3: Dynamic panel data model with AR(1) 

       * Time-varying Effects Estimators
     (6) CSS: Cornwell-Schmidt-Sickles Estimator (Within:CSSW, GLS:CSSG)
     (7) KSS: Kneip-Sickles-Song Estimator
     (8) BC : Battese-Coelli Estimator
     (9) DEA: Data Envelope Analysis


     Time-invariant Effects Estimators: (1)-(5)
     Time-variant Effects Estimators: (6)-(9)


5.   Notes on DEA estimator
     . DEA does not allow multiple outputs, so makes sure that X has only input 
       values


6.   Data Ordering: Firm1  Time 1-T
                    Firm2  Time 1-T
		    ...


7.   To ensure proper spacing in outputs, use a fixed-width font, 
     such as Fixedsys or Courier. We recommend to use 'Courier New' 
     font for Command Window.

     : File-Preferences-General-Font&Colors-Desktop Font-Courier New


8.   If output matrix is too large so that it does not appear in a row of 
     command window, refer to the following solution obtained from the Technical
     Support Page of MATLAB Homepage: 


     Solution Number: 30703 
     Date Last Modified: 2002-06-18 
     Product: MATLAB 6.1 ==> 6.5 
     Platform: All Builds 

     Problem Description

     Why can't I change the font size in the Command Window to allow more data 
     to appear in one row for MATLAB? 

     When the font size inside the Command Window is changed, it does not 
     increase the amount of data that will appear in a row of command window 
     output.


     Solution:

     This problem has been fixed in MATLAB 6.5 (R13). If you are using an 
     earlier version, please see the workaround below. 

     Resizing the MATLAB Command Window will force MATLAB to recognize the new 
     font size. You can change the font size of the Command Window by going to 
     the 'Preferences' menu. 


9.   If the data size is too large, you may have a memory problem in calculating
     RND, HT, CSSG, and PSS1,2,3 estimators. In that case, try to increase virtual 
     memory of Windows, or try to reduce data size. 


10. The method of truncation of the estimated effects: For the time invariant estimators
    this is easy because we may just delete top and bottom 5% of effects. 
    For the time varying estimators, it is not easy. There may exist various ways to do it. 
    The method adopted here is the following. We check the firms whose effects are 
    included in the upper and bottom 5% range 'at least at one time period'. Then 
    we delete all these firms' effects from the calculation of efficiency. Thus, 
    practically, we are deleting more than 10(5+5)% of firms from the calculation. Also, 
    note that the total number of firms deleted varys from estimators to estimators. 


11. Files included
     
    bc.m
    bootpss.m
    css.m
    cssw.m
    dea.m
    estm.m
    ht.m
    kss.m
    panel.m
    panelf.m
    pde.m
    pss1.m
    pss2.m
    pss3.m
    spearman.m

    limann.asc
    limlarge.asc
    limsmall.asc

    readme.txt
    update0409.txt
    Sickles.pdf


12. Sample Data
    U.S. Limit Bank
    number of banks: 1220
    time period: 1984-1995, 12 years, annual

    
    * variables 
    reln : real estate loans
    ciln : commercial and industrial loans
    inln : installment loans
      CD : certificate deposits
      DD : demand deposits
      OD : retail time and savings deposits
     lab : labor
     cap : capital
    purf : purchased funds


    * data in limann.asc, limlarge.asc, limsmall.asc

    columns 1-3
    log(reln)  -log(ciln/reln) -log(inln/reln) 

    columns 4-9
    -log(CD) -log(DD) -log(OD) -log(lab) -log(cap) -log(purf)


    * files 
    limann.asc: 14640 by 9, 1220 banks, total banks
    limlarge.asc: 7320 by 9, 610 banks, large size banks
    limsmall.asc: 7320 by 9, 610 banks, small size banks

    small size banks: average loan size is below 50 percentile
    large size banks: average loan size is above 50 percentile



13.  If you have any questions, please contact:

     Robin C. Sickles

     Professor
     Department of Economics
     Rice University
     6100 Main
     Houston, TX 77005, USA

     E-mail: [e-mail]
     Tel: [phone]
     Fax: [phone]

or 

     Wonho Song

     Research Fellow
     Korea Institute for 
     International Economic Policy (KIEP)
     300-4, Yomgok-dong, Seocho-gu
     Seoul 137-747, South Korea

     E-mail: [e-mail], [e-mail]
     Tel: [phone]
     Fax: [phone]