Agricultural Production and

Policy Impact Modelling

(731.386)

Erwin Schmid
[email protected]

Institute for Sustainable Economic Development

Zoom-link
https://bokuvienna.zoom.us/j/64361436180
1
 Why is mathematical programming a
success?
 Real Life Models

 Mathematical insights

 Efficient algorithms, professional software packages

 Fast and cheap computers

Goals:
 Learn real-life problem and model formulations.
 Learn how to use GAMS to build these models and to do analyses.
 Learn good model building in GAMS.
 Learn how to interpret model solutions as well as to synthesize and
report model results.

2
 References

McCarl B.A., and T.H. Spreen (1999). Applied Mathematical

Programming using Algebraic Systems.

Online available:
http://agecon2.tamu.edu/people/faculty/mccarl-
bruce/books.htm

3
Book: Bruce McCarl and Thomas Spreen

4
 Course Outline:
1. Introduction to Linear Programming
2. Introduction to GAMS (General Algebraic Modeling Systems)
- Sets, Parameters and Conditions
3. Impact Analysis
- Farm Level Analysis
- Comparative Static/Dynamic Analysis
- Regional Analysis
- Sector (partial equilibrium) Analysis
4. Fixing unbounded and infeasible Models using GAMSCHK
5. Reporting Model Results

5
Course Material at BOKUlearn

6
 GAMS installation
Installation Process
Download: Visit the GAMS website (www.gams.com) and download the installation file for
Windows (windows_x64_64.exe).

Start Installation:

Run the downloaded file. Administrator rights may be required.

Choose the installation mode (standard or advanced). In advanced mode, you can
configure additional settings like start menu folder, PATH variable, and desktop icons.

Specify the installation path (e.g., C:\GAMS).

Select GAMS Studio as the primary editor if prompted during installation.

 GAMS installation II
Set up License:

Use Access Code: 47125672-d6ef-4756-95e0-b0520658d7fe

Completion: After installation, a desktop icon for GAMS Studio will be created, which you
can use to launch the application.

Using GAMS Studio

Launch: Open GAMS Studio using the desktop icon.

Create or Open Files:

Go to the file selection in the upper left corner to create new files or open existing
ones.
 GAMS installation III
Prepare Models:

Edit your models in Studio's text editor and ensure they are ready to run.

Execute:

Click the "Run" button (or press F9) to execute the model.

Analyze Results:

Navigate through the output files and check the results in the "Process Log" window.

GAMS Studio offers a user-friendly interface with

www.gams.com
Select proper platform download
Check “use advanced installation
mode”
Select folder on your computer
 License access code:

Access Code: 47125672-d6ef-4756-95e0-b0520658d7fe

 Assignment

 Develop a model and perform some analysis.

 Latest submission date: 20th June 2025
 Submission by email to: [email protected]
 max 2 pages (single space) + appendix
 Structure:
 Title, Name, Matrikelnummer, Date
 Research problem description
 Research objective
 Data and model description (the model mathematically)
 Results
 Conclusions
 Appendix with gams code 16
 Notes on Linear Programming

Based on Uwe Schneider (Uni Hamburg), and

Bruce McCarl (TAMU)

17
 Linear Programming (LP)
Definition: A linear programming problem (LP) is an optimization problem
for which:
1. We attempt to maximize (minimize) a linear function of the
decision variables. (objective function)
2. The values of the decision variables must satisfy a set of
constraints, each of which must be a linear inequality or linear
equality.
3. A sign restriction on each variable. For each variable xi the sign
restriction can either be (a) xi ≥ 0; (b) xi ≤ 0; or (c) xi is unrestricted
(objv).
Assumptions:
Proportionality: If one item brings in a profit of x, then k items bring in a
profit of kx.
Additivity: The decisions made are independent; prohibits cross-product
terms e.g. 5x1x2 which might represent interaction effects between var.
Divisibility: Decision variables can take on fractional values.
Certainty: The values of various parameters are known with certainty.
9
 Modelling Assumptions for Linear
Programming

Whenever we solve a linear programming

model, we should ask ourselves whether
these assumption hold, and whether the
solution makes sense
=> CONTEXT is KING!!!

19
 Linear Programming, 1
Max c1 *X1 +…+ cn *Xn = z

s.t. a11*X1 +…+ a1n*Xn ≤ b1

…
am1*X1 +…+ amn*Xn ≤ bm
X1 , Xn ≥ 0

Standard form
20
 LP Tableau

x1 x2 s1 s2 Rhs

Obj c1 c2 0 0

Eq1 a11 a12 1 0 <= b1

Eq2 a21 a22 0 1 <= b2

Type + + + +

21
 Units

X1 (ha) X2 (ha) Rhs

Obj (Euro) 120 €/ha 180 €/ha

Eq1 (ha) 1 (ha/ha) 1 (ha/ha) 50(ha)

Eq2 (h) 2.2 (h/ha) 2.5 (h/ha) 100(h)

22
 Linear Programming, 2
Max c1 *X1 +…+ cn *Xn = z

s.t. a11*X1 +…+ a1n*Xn ≤ b1

…
am1*X1 +…+ amn*Xn ≤ bm
X1 , Xn ≥ 0

n +1 Variables
23
 Linear Programming, 3
Max c1 *X1 +…+ cn *Xn = z

s.t. a11*X1 +…+ a1n*Xn ≤ b1

…
am1*X1 +…+ amn*Xn ≤ bm
X1 , Xn ≥ 0

Objective function
24
 Linear Programming, 4
Max c1 *X1 +…+ cn *Xn = z

s.t. a11*X1 +…+ a1n*Xn ≤ b1

…
am1*X1 +…+ amn*Xn ≤ bm
X1 , Xn ≥ 0

n Objective function coefficients (data)

25
 Linear Programming, 5
Max c1 *X1 +…+ cn *Xn = z

s.t. a11*X1 +…+ a1n*Xn ≤ b1

…
am1*X1 +…+ amn*Xn ≤ bm
X1 , Xn ≥ 0

m Constraints
26
 Linear Programming, 6
Max c1 *X1 +…+ cn *Xn = z

s.t. a11*X1 +…+ a1n*Xn ≤ b1

…
am1*X1 +…+ amn*Xn ≤ bm
X1 , Xn ≥ 0

m*n technical coefficients (data)

27
 Linear Programming, 7
Max c1 *X1 +…+ cn *Xn = z

s.t. a11*X1 +…+ a1n*Xn ≤ b1

…
am1*X1 +…+ amn*Xn ≤ bm
X1 , Xn ≥ 0

m resource limits (data)

28
 Linear Programming Example
Max 2*X1 + 3*X2 = z

s.t. X1 + 2*X2 ≤ 10
X1 , X2 ≥ 0

29
 Feasibility Region
X2
Max 2*X1 + 3*X2 = z
s.t. X1 + 2*X2 ≤ 10
X1 , X2 ≥ 0
5

X1,X2 ≥ 0

X2 ≤ 5 – 0.5*X1

10 X1

30
 Feasibility Region
X2

Convex Set

10 X1

31
 Objective Function Isoclines - 1
X2
Max 2*X1 + 3*X2 = z
s.t. X1 + 2*X2 ≤ 10
5
X1 , X2 ≥ 0

X2 = z/3 – 2/3*X1

10 X1
z=6
X2=2–2/3*X1

32
 Objective Function Isoclines - 2
X2
Max 2*X1 + 3*X2 = z
s.t. X1 + 2*X2 ≤ 10
5
X1 , X2 ≥ 0

X2 = z/3 – 2/3 * X1

10 X1
z=6 z=15
z=0

33
Graphical Solution
X2
Max 2*X1 + 3*X2 = z
s.t. X1 + 2*X2 ≤ 10
X1 , X2 ≥ 0
5

X2 = z/3 – 2/3 * X1

z=20 X1
10

34
 Graphical Solution
X2

Is at extreme point!
5

z=20 X1
10

35
Other Extreme Points
X2

10 X1

36
 Solving Arithmetically
Max c1 *X1 +…+ cn *Xn = z

s.t. a11*X1 +…+ a1n*Xn ≤ b1

…
am1*X1 +…+ amn*Xn ≤ bm
X1 , Xn ≥ 0

37
 Convert Inequalities
Max c1 *X1 +…+cn *Xn = z
s.t. a11*X1 +…+a1n*Xn ≤ b1
…
am1*X1 +…+amn*Xn ≤ bm
X1 , X2 ≥ 0
Max c1 *X1 +…+cn *Xn +0*S1 +…+0*Sm = z
s.t. a11*X1 +…+a1n*Xn +1*S1 +…+0*Sm = b1
…
am1*X1 +…+amn*Xn +0*S1 +…+1*Sm = bm
X1 , X2 , S1 , Sm ≥ 0
38
 Arithmetic Problem, 1

 Number of unknown variables: n+m

 Number of equations (constraints): m

 Can only solve m equations for m

unknown variables

39
 Arithmetic Problem, 2

From graphical illustration, we know

Solution occurs at extreme point

40
1 Constraint ⇒1 Non-Zero Variable

Max Objective
X2 s.t. X1 + 2*X2 + S = 10
X1 , X2 , S ≥ 0
 X1 = 0 
 
 X2 = 5 
 S=0 
5  

X1 = 1 0
 X1 = 0 
   X 2 = 0 
 X2 = 0   
 S = 10   S=0 
   

10 X1
32
 2 Constraints
X2
10 Max Objective
s.t. X1 + 2*X2 + S1 = 10
2*X1 + X2 + S2 = 10
X1 , X2 , S1, S2 ≥ 0

5 10 X1
33
 2 Constraints ⇒ 2 Non-Zero Variables
X2 Max Objective
10
s.t. X1 + 2*X2 + S1 = 10
 X1 = 0 
2*X1 + X2 + S2 = 10

 X2 = 5 
 X1 , X2 , S1, S2 ≥ 0
   X1 = 10 
 S1 = 0   3
 X2 = 10 
 S2 = 5   3
5  S1 = 0 
 
 S2 = 0 
 X1 = 5 
 
 X1 = 0   X2 = 0 
 
 X2 = 0   S1 = 5 
 
 S1 = 10 
   S2 = 0 
 S2 = 10  5 10 X1
43
 Arithmetic Problem, 3

Extreme points have only m non-zero

variables and n-m zero variables
Thus, if we knew which n-m variables are
zero, we could solve for the remaining m
variables using our m equation system

44
 Optimal solution is extreme point

 Number of non-zero variables = number of

equations

 Non-zero (positive) variables are called basic

variables

 All remaining variables are called

non-basic variables

 Non-basic variables are zero.

45
Decomposed LP in Matrix Notation

Max CBXB + CNBXNB

s.t. BXB + ANBXNB = b

XB , XNB ≥ 0
Solution of LP in Matrix Notation

BBXB = b - ANBXNB

BXB = b

XB = B-1b
 Solver tasks

 Determine which variables form basis

 Invert coefficient matrix to calculate

optimal variable values

48
 If optimal solution is an extreme
point, why not simply calculate all
extreme points and choose the one
with highest objective function
value?

49
Number of Possible Extreme Points

n!
(n − m)!⋅ m!
n .. Number of Variables
m .. Number of Equations

50
 Calculating all extreme points?

(n) (m) n!/[m!*(n-m)!]

Variables Equations Combinations Time
10 5 252 ~0 Seconds
20 10 184.756 58 Seconds
30 15 155.117.520 17 Hours
40 20 137.846.528.820 2 Years
50 25 126.410.606.437.752 2 Millenia
 Linear Program

MAX CBXB + CNBXNB

s.t. BXB + ANBXNB = b

XB , XNB ≥ 0
 Important LP Equations
=Z Cb X b + Cnb X nb

X b B −1b − B −1 Anb X nb
=

Cb B −1b − (Cb B −1 Anb − Cnb ) X nb

Z=

∂Z −1
= C=
b B U SHADOW PRICE
∂b
∂Z
− ( Cb B −1 Anb − Cnb )
=
∂X nb
REDUCED COST
 Duality

Pr imal Dual
Max CX Min U′b
s.t. AX ≤ b s.t. U′A ≥ C
X ≥0 U ≥0

54
 Complementary Slackness

 derived from duality

U (b − AX )
*' * = 0

(U A − C )X
*' * = 0

55
Unbounded Solution

56
Infeasible Solution

x2

A
B x1

57
Multiple Optima

Isocline with
highest
x2
objective
P1

P2 x1

58
Degeneracy

x2

x1

P1
59
 GAMS
General Algebraic Modeling System

60
GAMS Homepage

61
GAMS Software

62
Our Text - Model
Max 2*X1 + 3*X2 = z

s.t. X1 + 2*X2 ≤ 10
2*X1 + X2 ≤ 10
X1 , X2 ≥ 0

63
 GAMS Model

Positive Variables x1,x2;

Declare variables
Free Variables Z; and equations

Equations obj,e1,e2;
obj.. 2*x1 + 3*x2 =E= z;
e1.. 1*x1 + 2*x2 =L= 10; Assign equations

e2.. 2*x1 + 1*x2 =L= 10;

Model name (slide)
Model slide /all/;
Solve statement
solve slide using lp maximizing z;

64
 GAMS Commands and Functions

Sets ... setting of model dimensions and elements

Parameter ... for parameter specification and calculation
Scalar ... scalar specification

Table ... data input and output

$ Conditions ... for calculations, model equations & variables
The Model
Variables ... defining decision variables and obj-variable
Equations ... defining equations (constraints & objective func.)
etc.
65
Sets in GAMS

66
Tables in GAMS

67
 external table input and output reading
from prn files:
$INCLUDE 'D:\WPR\Lehre\AgSectResMod\Chapter_1\Tab2.prn'

from xls files:

Parameter Tab3(Lines,Columns);
needs to be written in
one line!!!
$CALL GDXXRW D:\WPR\Lehre\AgSectResMod\Chapter_1\Tab3.xls
output=D:\WPR\Lehre\AgSectResMod\Chapter_1\Tab3.gdx Par=Tab3 Rng=Tab!a1:d22 Rdim=1 Cdim=1
$GDXIN D:\WPR\Lehre\AgSectResMod\Chapter_1\Tab3.gdx
$LOAD Tab3
$GDXIN

output into xls files

Execute_Unload "D:\WPR\Lehre\AgSectResMod\Chapter_1\Tab3.gdx",Lines,Columns,Tab3;
Execute 'GDXXRW.EXE D:\WPR\Lehre\AgSectResMod\Chapter_1\Tab3.gdx needs to be written
o=D:\WPR\Lehre\AgSectResMod\Chapter_1\Tab4.xls par=Tab3 Rng=Tab1!a1:d22'; in one line!!!
68
 $ Conditions
In ordinary algebra we have ways of controlling the
terms considered in indexed expressions. In
particularly, if i had 40 elements and we whished to
sum elements 1 to 20 we would write
20

X = ∑ Zi
i=1

GAMS introduces that qualifications on indexes like i via

the use of conditionals ($ and if) or special sets.

69
 $ Conditions

GAMS conditionals can be applied in several settings:

To control whether an equation is calculated

x$y = 10+z;
To control whether a term appears in an equation
x = 10+z$y
To control whether a model equation is set up
constraint$y.. z =L=10;
To control whether a term is included in a model equation
constraint.. z+x$y =L= 10;

70
 $ Conditions
To control inclusion of terms in sum, loop, smax and other
commands indexing over a set
x = sum(i$y, z(i));
loop (i$y, x=x+z(i));
To control whether a set of code is executed
if(y,x=10);
e.g., x$y=3;
z$y=4;
can become
if(y,
x=3;
y=4);

71
 $ Conditions
right versus left hand conditionals
Conditional placement affects speed and results
1) x(i,j,k)$data(i,j,k)=1+y(i,j,k);
2) x(i,j,k)=(1+y(i,j,k))$data(i,j,k);
The first will only go through cases where the data array
contains non-zeros entries. it will be faster!
If x(i,j,k) has numbers where data(i,j,k) does not, those numbers
will be retained!!!
The second will cover all cases and will zero cases of x that do
not correspond to non-zero data. The answer will be different
and the statement will likely take longer to execute (GAMS has
been introducing internal optimisation to make it perform
better in such cases)!!!
72
 Parameter Calculation

Calculate from table:

mean
mean per lines
mean per columns

standard deviation 1 2
Std = (xi − x )
standard deviation per lines n

standard deviation per columns

73
 Example: Feed Valuation Model
Valuation of non-market goods (e.g. hay, silage, green
fodder, etc.) by nutrient and energy contents (MJ, RP)
of comparable feeds (e.g. barley, soybean, etc)
Model:
min cx
s.t.
Ax ≥ b
x≥0

74
 Example: Data

Calculate the value of 1 kg hay using barley, wheat and

soybeans as comparable goods
MJ/kg gRP/kg €/kg
Barley 12.6 88 0.179
Soybeans 13.1 460 0.375
Wheat 13.6 112 0.164
Hay 9.3 117 ?
straw 5.6 - ?

75
 Model Development

 Develop mathematical structure

 Implement in GAMS
 Declare and assign sets
 Declare parameters
 Declare variables
 Declare equations
 Assign equation structure

76
 LP Tableau

x1 x2 x3 Rhs

Obj 0.179 0.375 0.164 = Z min

(€/kg) (€/kg) (€/kg)
Eq1 12.6 13.1 13.6 ≥ 9.3
(MJ/kg) (MJ/kg) (MJ/kg) (MJ)
Eq2 88 460 112 ≥ 117
(gRP/kg) (gRP/kg) (gRP/kg) (gRP)
Type + + +

77