SFM GAUSS Code

This page introduces the GAUSS programs, SFMs which can be used for various stochastic frontier models.  They have been written by Young Hoon Lee, but some programs were originally written by Seung Chan Ahn and Lee modified into this form under Ahn's permission. The SFMs estimate any linear stochastic production functions and estimate individual technical efficiencies. Please cite the source of program code.

Overview

The stochastic frontier models that the SFMs can cover are as follows;

    cross-sectional data models:

           Aigner, Lovell and Schmidt(1977) with half-normal and exponential

           Stevenson(1980) with truncated-noraml

           Almanidis, Qian and Sickles(2014) with tail-truncated half-normal

           Lee and Lee(2014) with uniform

    panel data models with time-invariant efficiency

            Battese and Coelli(1988) with half-normal and truncated-noraml

            Schmidt and Sickles(1984)

    panel data models with time-varying  

            Battese and Coelli(1992) and Kumbhakar(1991) without distributional assumption, the fixed effect treatment

            Cornwell, Schmidt and Sickles(1991)

            Lee and Schmidt(1993)

            Lee(2006)

            Ahn, Lee and Schmidt(2007)

            Lee(2010)

The SFMs are based on the GAUSS program and some SFMs require the co(constrained optimization) module which is provided by the GAUSS. All SFMs along with example data files can be downloaded below. If you obtain a copy of the SFM, please send a brief email to Young Hoon Lee at yhnlee@sogang.ac.kr. The current version of SFM is a bit crude. If you have any suggestions regarding how the SFM programs could be improved or if you find a bug, then please contact Young Hoon Lee by email.

If you have any technical questions regarding the use of these SFMs, you can contact Young Hoon Lee by email. but only brief questions (less than 5 minutes) will be answered.

Instruction

For Koreans, refer to [확률적변경모형: 이론과 응용 Stochastic Frontier Models: Theory and Application] (Sogang Unversity Press, 2014) for instructions.

For non-Koreans, refer to the working paper, "Stochastic Frontier Models Using GAUSS."  in Research Institute For Market Economy in Sogang University. 

1. Stochastic Frontier Models with Cross-Sectional Data: This code includes Aigner, Lovell and Schmidt(1977) with half-normal and exponential inefficiency distributions, Stevenson(1980) with truncated-noraml, Almanidis, Qian and Sickles(2014) with tail-truncated half-normal and Lee and Lee(2014) with uniform. All models are estimated by ML.

Short Instruction: Since this code uses the constrained optimization, users need [CO module] in order to be able to run this code.

line 51 chooses a specific model. 1: half-normal, 2: exponential, 3: truncated-noraml, 4: tail-truncated half-normal, 5: uniform

line 60-70 sets innitial values.

code: SFM_MLE_cross-section 

data:  data

2. Stochastic Frontier Models with Panel Data and Time-Invariant Efficiency: These codes include Battese and Coelli(1988) with half-normal and truncated-noraml, and Schmidt and Sickles(1984). BC88 is estimated by ML and the code for SS provides with OLS, Within and GLS with hetero/auto-adjusted variances.

Short Instruction for BC88:

Since this code uses the constrained optimization, users need [CO module] in order to be able to run this code.

line 41 deginates a specific model. 1: half-normal, 2: truncated-noraml

code: SFM_BC88_MLE 

data: MLB 

Short Instruction for SS:

No specific instructions. Please deginate data file, sample size and output/input variables.

code: SFM_SS 

data:  MLB

3. Stochastic Frontier Models with Panel Data and Time-Varying Efficiency: These codes include Battese and Coelli(1992) and Kumbhakar(1991) without any distributional assumption, Cornwell, Schmidt and Sickles(1991), Lee and Schmidt(1993).  All models are estimated by the fixed effect treatment: BC92(GMM), CSS(Within with heteo/auto adjusted variance), LS(OLS, SS, Within LS).

Short Instruction for BC92 and Kumbhakar:

line 143-149 set intrumental variables.

line 153-154 set innitial values.

code: SFM_BC92,  SFM_KUM

data:  rice  

Short Instruction for CSS:

line 37-40 set output, inputs variables.

line 42-44 set variable names.

code: SFM_CSS 

data: rice

Short Instruction for LS:

line 33-40 set output, inputs variables.

line 115 sets innitial values.

code: SFM_LS 

data:  rice

4. More Stochastic Frontier Models with Panel Data and Time-Varying Efficiency: These codes include Lee(2006, GrLS), Lee(2010, GrBC) and Ahn, Lee and Schmidt(2007, ALS07).  All models are estimated by the fixed effect treatment: GrLS(Witnin), GrBC(Witnin), ALS07(GMM).

Short Instruction for GrLS:

line 33-68 set grouping.

line 323-341 are relevant to testing for various groupings.

code: SFM_GrLS

data:  rice

Short Instruction for GrBC

line 52-69 set grouping.

line 323-341 are relevant to testing for various groupings.

code: SFM_GrBC 

data:  rice

Short Instruction for ALS07:

line 14-40 set output, inputs and instrumental varialbes.

line 50 sets number of factor.

line 90-111 set instruments

code: SFM_ALS07

data:  rice