See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/325399780

Optimal Allocation of Funds for Loans Using Karmarkar’s Algorithm: Capital

Rural Bank, Sunyani-Ghana

Article · May 2018

DOI: 10.9734/JEMT/2018/40292

CITATIONS READS

2 130

5 authors, including:

Anthony Donkor Kwaku Darkwah

University of Energy and Natural Resources Kwame Nkrumah University Of Science and Technology
15 PUBLICATIONS 35 CITATIONS 19 PUBLICATIONS 29 CITATIONS

SEE PROFILE SEE PROFILE

Justice Kwame Appati

University of Ghana
42 PUBLICATIONS 114 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Operations View project

Computer Vision View project

All content following this page was uploaded by Anthony Donkor on 28 May 2018.

The user has requested enhancement of the downloaded file.

 Journal of Economics, Management and Trade

21(5): 1-13, 2018; Article no.JEMT.40292

ISSN: 2456-9216
(Past name: British Journal of Economics, Management & Trade, Past ISSN: 2278-098X)

Optimal Allocation of Funds for Loans Using

Karmarkar’s Algorithm: Capital Rural Bank,
Sunyani-Ghana
Anthony Donkor1*, Kwaku Darkwah2, Justice Appati3, Prosper Gakpleazi2
and Naninja Wisdom4
1
University of Energy and Natural Resources, Sunyani, Ghana.
2
Kwame Nkrumah University of Science and Technology, Kumasi, Ghana.
3
University of Ghana, Legon, Ghana.
4
Dormaa Senior High School, Dormaa, Ghana.

Authors’ contributions

This work was carried out in collaboration between all authors. Author AD designed the study,
performed the statistical analysis, wrote the protocol and first draft of the manuscript. Authors KD and
JA managed the analyses of the study. Authors PG and NW managed the literature searches. All
authors read and approved the final manuscript.

Article Information

DOI: 10.9734/JEMT/2018/40292
Editor(s):
(1) Alfredo Jimenez Palmero, Kedge Business School, France.
Reviewers:
(1) Charles Ngome Eke, Federal Polytechnic Nekede, Nigeria.
(2) J. Ramola Premalatha, Vellore Institute of Technology, India.
(3) Ramona Lacurezeanu, Babes-Bolyai University, Romania.
(4) Maura La Torre, Libera Università Mediterranea, Italy.
Complete Peer review History: http://www.sciencedomain.org/review-history/24824

Received 14th January 2018

Original Research Article Accepted 20th March 2018
Published 26th May 2018

ABSTRACT

In Ghana, the banking industry is now characterised by increasing competition and innovation. This
has made most of the banks to adopt a scientific approach to improve the quality of their loan
structure. The decline of relevant portfolio planning models especially in Ghana is attributed mainly
to the evolving dynamics of the Ghanaian banking industry where the regulatory controls have
changed with a high frequency. Due to the model used in allocating funds to various loan types, a lot
of banks had suffered substantial losses from some bad loans in their portfolio. As a result, most
banks are not able to maximize their profit on loans due to poor allocation of funds. The purpose of

_____________________________________________________________________________________________________

*Corresponding author: E-mail: [email protected];

 Donkor et al.; JEMT, 21(5): 1-13, 2018; Article no.JEMT.40292

this study is to develop a linear programming model using the Karmarkar's projective scaling
algorithm to help Capital Rural Bank Limited to maximise their profit on loans. The results from the
model showed that Capital Rural Bank Limited would be making annual loan turn-over of
GH¢5,961,300.00 which is 61.3% more than the established previously. From the study, it was
further realised that policy directions mostly influence the optimal solution and not probability of bad
debt.

Keywords: Linear programming; Karmarkar’s algorithm; maximize profit.

1. INTRODUCTION improving the approximation of the optimal

solution [6,7,8].
Lending to firms is an essential business activity
for every commercial bank. This means that According to Prosper (2011), Panos and
loans portfolio management is the necessary Mauricio (1996) explained how interior point
activity for getting the maximum return and methods, originally invented in the context of
minimum risk from bank loans. It is fundamental linear programming, have found a much broader
to the safety and sustainability of banks. Most range of applications, including global
loan managers concentrate their effort on how to optimization problems that arise in engineering,
effectively approve loans and carefully monitor computer science, operations research, and
loan performance. It is therefore vital and prudent other disciplines. Quintana et al. [9], stated that
for financial establishments like banks to since Karmarkar's first successful interior-point
maximise the return on their loan portfolio [1,2]. algorithm for linear programming in 1984, the
According to Cohen and Hammer, what makes interest and consequently the numbers of
the task of funds allocation difficult is finding an publications in the area have increased
appropriate balance between three desirable tremendously. They reviewed and classified
objectives in loans portfolio management [3]. significant publications on interior point methods,
These objectives are; profitability, liquidity and on the practical implementation of the most
safety. Banks should decide on the distribution of successful interior-point algorithms and on their
its capital among the various types of loans, applications to power systems optimisation
which differ in duration and risk and are affected problems [9].
by the environment, the borrowers’ deposits at
the given bank as well as other factors Ferris and Philpott [10], described how
(Klaassen) [4]. Besides, banking loan decisions Karmarkar’s polynomial-time algorithm for linear
require the use of large data and substantial programming significantly outperform the simplex
processing time to be able to serve a large method. The described many numerical
number of variables and a variety of different experiments carried out by other workers in the
cases related to different customers. In this field which show a much smaller iteration count
paper, we study how a rural bank in Ghana can than the simplex method but larger
optimise the allocation of funds for loans. computational times. Some had shown that, by
using advanced numerical linear algebra and
Linear programming is frequently applied in heuristics to exploit the problem structure, it is
portfolio management. To maximise profit on possible occasionally to beat the simplex method
loans, funds must be optimally allocated to the even in terms of computation time [10].
different loan types and linear programming
models are the suitable models for optimal Kojima [11] reported that Karmarkar interior-point
allocation of resources. Dantzig [5] introduced method has been successfully extended in the
the simplex algorithm in 1947 was among the field of continuous optimization to convex
pioneers to use it to optimally solve linear quadratic programs, semi-definite programs, and
programming problem and thus can be used to more general convex problems, while, in the field
optimally allocate funds to different loan types of discrete optimization, it has been playing an
[5]. Karmarkar introduced an interior point important role in terms of the semi-definite
algorithm for solving linear programming programming relaxation of 0-1 integer and
problems. It was the first reasonably efficient nonconvex quadratic programs [11].
algorithm that solves these problems in
polynomial time. It does not follow the boundary Todd [12], showed a variant of Karmarkar's
of the feasible set as in the simplex method, but projective algorithm for linear programming can
moves through the interior of the feasible region, be viewed as following the approach of Dantzig-

2
 Donkor et al.; JEMT, 21(5): 1-13, 2018; Article no.JEMT.40292

Wolfe decomposition. At each iteration, the feasible solutions. In this paper we convert a
current primal feasible solution generates prices standard maximization problem of a bank’s loan
which are used to form a simple subproblem. portfolio allocation to Karmarkar form and solve
The solution to the subproblem is then the resulting loan portfolio optimization to
incorporated into the currently feasible solution. optimality and further provide detail analysis of
With a suitable choice of step size a constant the loan allocation structure [14,15].
reduction in potential function is achieved at each
iteration [12]. 2. METHODS

Monteiro [13], analyzed the convergence and Karmarkar’s projective scaling method, also
boundary behaviour of the continuous known as Karmarkar’s interior point LP algorithm,
trajectories of the vector field induced by the starts with a trial solution and shoots it towards
projective scaling algorithm as applied to the optimum solution.
(possibly degenerate) linear programming
problems in Karmarkar's standard form. They This algorithm addresses LP problem of the
showed that a projective scaling trajectory tends form:
to an optimal solution which in general depends
on the starting point. When the optimal solution is T
Minimize Z = C X
unique, they prove that all projective scaling
trajectories approach the optimal solution
Subject to
through the same asymptotic direction. The
AX = 0
analysis was based on the affine scaling n

trajectories for the homogeneous standard form i 1

xi  1
that arises from Karmarkar's standard form by
removing the unique non-homogeneous X 0
constraint [13].
where A is a m x n matrix of rank m, is 1 x n
T
C

Bayer and Lagarias (1989), described a vector,

geometric structure underlying Karmarkar's
X   x 1 ,...., x n  , n  2 and A and C are all real.
T
projective scaling algorithm for solving linear
programming problems. A basic feature of the
projective scaling algorithm is a vector field 2.1 Converting Linear Programming
depending on the objective function which is
Problem to Karmarkar’s Form
defined on the interior of the polytope of feasible
solutions of the linear program. The geometric
structure studied is the set of trajectories In converting the LP problem:
obtained by integrating this vector field, which we
T
call P-trajectories. They also study a related Maximize Z = C X
vector field, the affine scaling vector field, and its
associated trajectories, called A-trajectories. The Subject to
affine scaling vector field is associated to another b
linear programming algorithm, the affine scaling AX
algorithm. Affine and projective scaling vector X 0
fields are each defined for linear programs of a
special form, called strict standard form and to Karmarkar’s Form, the following steps are
canonical form, respectively. This derives basic used:
properties of P-trajectories and A-trajectories. It
reviews the projective and affine scaling Step 1:
algorithms, defines the projective and affine
scaling vector fields, and gives differential Write the dual of the given primal problem
equations for P-trajectories and A-trajectories. It
shows that projective transformations map P- Minimize Z = bW
trajectories into P-trajectories. It presents
Karmarkar's interpretation of A-trajectories as Subject to
steepest descent paths of the objective function
with respect to the Riemannian geometry AT W  C T
restricted to the relative interior of the polytope of W  0

3
 Donkor et al.; JEMT, 21(5): 1-13, 2018; Article no.JEMT.40292

Step 2: Minimize M = CT V

Introduce Slack and Surplus variables to primal Subject to

and dual problems AV = 0
n
 Vi  1
Combine these two (2) problems i 1

V 0
Step 3:
2.2 Steps of Karmarkar’s Algorithm
 xi   wi  K
Introduce a bounding constraint ,
where K should be sufficiently large to include all Step 1:
feasible solutions of original problem
r 
1 n 1
 
Introduce a slack variable in the bounding Compute n n  1  and 3n .
constraint and obtain

Put k = 0 and let V 0  1 / n ,...., 1 / n 

T

 xi   wi  s  K
Step 2:
Step 4:
Y0  V0
Introduce a dummy variable d (subject to (a) Let
D 0  diag V 0 
condition d = 1) to homogenize the constraints
and replace the equations
 AD 
P   0

 1  , 1  1 , . . , 1
 xi   wi  s  K
and d = 1 with the following
equations:
C  C T D0
 xi   wi  s  Kd  0

 i  i
x w  s  d  K 
and
1 Compute

C p  I  P T PP  T

1

PC T

Step 5: Cp  0
If , , any feasible solution becomes
Introduce the following transformations so as to
an optimal solution. Stop
obtain one on the RHS of the last equation:
(b) Otherwise compute
x i  K  1  y i , i  1, 2 ,.... m  n
C
wi  K 1 ymni , i  1,2,....m  n  Y0   r
p
Y new
C p

s  K 1 y2m2n1 V1 
D 0 Y new
1 D 0 Y new

d  K  1 y2m2n2 M  C T V1

Step 6: Step 3:

Introduce an artificial variable

y
2m2n3 (to be Increase k by one and repeat Step 1 until the
minimized) in all the equations such that the sum objective function (M) value is less than or equal
of the coefficients in each homogenous equation to zero.
is zero and coefficients of the artificial variable in
the last equation is one. (Dhamija, A.K.) [16] 3. DATA PRESENTATION AND ANALYSIS
The general Karmarkar’s form is as shown Capital Rural Bank provides a flexible loan
below: payment term for all the types of loans. The

4
 Donkor et al.; JEMT, 21(5): 1-13, 2018; Article no.JEMT.40292

Management Board of the bank, that was Subject to the following policy constraints:
recording bad debts on the loans and wanted a
restructuring of the loan types. The data supplied  Total funds available for disbursement is
by the bank is summarized in Table 1. GH¢20,000,000.00

Table 1. Loans types, interest rate and x 1  x 2  x 3  x 4  x 5  x 6  20

probability of bad debt
 Salary, Funeral and Commercial loans
S/N Type of Interest Probability should be at most 60% of the total funds
loan rate of bad debt
1 Commercial 0.39 0.020 x1  x 2  x 3  12
2 Funeral 0.36 0.030
3 Salary 0.36 0.10  Housing loan should be at most 50% of
4 Susu 0.36 0.055 Salary and Funeral loans.
5 Agriculture 0.36 0.150
6 Housing 0.30 0.075  0 .5 x 2  0 .5 x 3  x 6  0
Source: Capital Rural Bank Limited, Sunyani
 The sum of Susu and Agriculture loans
Table 2 is the constructed from Table 1, columns must not be more than 40% of Commercial
2, 5 and 6 indicate the variable for loan amounts, and Housing loans.
formulas for amount of bad debt and profit
respectively.  0 .4 x1  x 4  x 5  0 .4 x 6  0

Capital Rural Bank Limited decided on a loan  The sum of the Agriculture and Funeral
policy with an amount of GH¢20,000,000.00. loans should be at most 15% of the total
funds.
The policy details for the type of loans are:
x2  x5  3
 Salary, Funeral and Commercial loans
should be at most 60% of the total  The overall ratio for bad debts in all loans
funds. should not exceed 0.045
 To assist people in the area (Sunyani
Metropolis) to undertake more housing 0.025 x1  0.015 x 2  0.035 x 3  0.010 x 4  0.105 x 5
projects, the Housing loan should be at  0.030 x 6  0
most 50% of Salary and Funeral loans.
 The sum of Susu and Agriculture loans  Non-negativity
must not be more than 40% of Commercial
and Housing loans. x1  0 , x 2  0 , x 3  0 , x 4  0 , x 5  0 , x 6  0
 The sum of the Agriculture and Funeral
loans should be at most 15% of the total The loan policy model is the Standard Linear
funds. Programming maximization problem below:
 The overall ratio for bad debts in all loans Maximize
should not exceed 4.5%.
Z  0.3622 x1  0.3192 x2  0.3464 x3 
3.1 Formulation of Linear Programming
(LP) Problem 0.2852 x4  0.1560 x5  0.2025 x6

The objective function is given by interest on Subject to

profit amount less bad debt amount and the
formula is x 1  x 2  x 3  x 4  x 5  x 6  20
x 1  x 2  x 3  12
6 6
Z  I i (1  B i ) x i  B i xi  0 .5 x 2  0 .5 x 3  x 6  0
i 1 i 1  0 .4 x1  x 4  x 5  0 .4 x 6  0
x2  x5  3
Z  0.3622 x1  0.3192 x2  0.3464 x3   0.025x1  0.015x2  0.035x3  0.010x4  0.105x5  0.030x6  0
0.2852 x4  0.1560 x5  0.2025 x6 x1 , x 2 , x 3 , x 4 , x 5 , x 6  0

5
 Donkor et al.; JEMT, 21(5): 1-13, 2018; Article no.JEMT.40292

Table 2. Formulation of bad debt and profit amounts

Loan type Loan amt Interest rate Probability of Bad debt amt Profit amt
(x) (I) bad debt (B) (B x ) (1B)x
Commercial x1 0.39 0.020 0.020 x1 0.980 x1
Funeral x2 0.36 0.030 0.030 x2 0.970 x2
Salary x3 0.36 0.010 0.010 x3 0.990 x3
Susu x4 0.36 0.055 0.055 x4 0.945 x4
Agriculture x5 0.36 0.150 0.150 x5 0.850 x5
Housing x6 0.30 0.075 0.075 x6 0.925 x6
Source: Anthony Donkor (obtained from Table 1)

3.2 Converting the Linear Programming By the procedure explained in section 3 and
Problem into Karmarkar’s Form illustrated in appendix A, the primal and dual
linear programming problems are converted to
The Standard Linear Programming form is as the Karmarkar format to obtain
shown below:
Minimize M  y27
Maximize

Z  0.3622x1  0.3192x2  0.3464x3  0.2852x4  0.1560x5  0.2025x6 Subject to:

Subject to y1  y 2  y 3  y 4  y 5  y 6  y 7  20 y 26  13 y 27  0
y 1  y 2  y 3  y 8  12 y 26  8 y 27  0
x 1  x 2  x 3  x 4  x 5  x 6  20
 0 . 5 y 2  0 . 5 y 3  y 6  y 9  y 27  0
x 1  x 2  x 3  12
 0 . 4 y 1  y 4  y 5  0 . 4 y 6  y 10  2 . 2 y 27  0
 0 .5 x 2  0 .5 x 3  x 6  0
y 2  y 5  y 11  3 y 26  0
 0 .4 x1  x 4  x 5  0 .4 x 6  0
0.025 y1  0.015 y2  0.035 y3  0.010 y4 
x2  x5  3
0.105 y5  0.030 y6  y12 1.07 y27  0
 0.025x1  0.015x2  0.035x3  0.010x4  0.105x5  0.030x6  0 y13  y14  0.4 y16  0.025 y18  y19  0.3622 y26  0.2128 y27  0
x1 , x 2 , x 3 , x 4 , x 5 , x 6  0
y13  y14  0.5 y15  y17  0.015 y18  y20  0.3192 y26  1.1658 y27
 Writing the dual of the given primal
0
problem:

Minimize 20 w 1  12 w 2  3 w 5 y13  y14  0.5 y15  0.035 y18  y21  0.3464 y26  0.1186 y27  0

Subject to :
y13  y16  0.010 y18  y22  0.2852 y26  0.7248 y27  0
w1  w 2  0 .4 w 4  0 .025 w 6  0 .3622
y13  y16  y17  0.105 y18  y23  0.1560 y26  1.949 y27  0
w1  w 2  0 .5 w 3  w 5  0 .015 w 6  0 .3192

w1  w 2  0 .5 w 3  0 .035 w 6  0 .3464 y13  y15  0.4 y16  0.030 y18  y24  0.2025 y26  1.5725 y27  0
w1  w 4  0 .010 w 6  0 .2852
0.3622 y1  0.3192 y2  0.3464 y3  0.2852 y4  0.1560 y5 
w1  w 4  w 5  0 .105 w 6  0 .1560
0.2025 y6  20 y13  12 y14  3 y17  33.3285 y27  0
w1  w 3  0 .4 w 4  0 .030 w 6  0 .2025
25
w1 , w 2 , w 3 , w 4 , w 5 , w 6  0 
i 1
y i  100 y 26  75 y 27  0

6
 Donkor et al.; JEMT, 21(5): 1-13, 2018; Article no.JEMT.40292

27 Thus
 yi  1
i 1
Z  0.3622(4.8076) 0.3192(2.2220) 0.3464(5.7065)
yi  0
, i  1 , 2 , 3 ,..., 27
0.2852(2.5654) 0.1560(0.0303) 0.2025(3.9390)
3.3 Computational Procedure and Results
or Z  5.9613
Using the Karmarker methodology in section 3,
programming code was written in MATLAB using 4. DISCUSSION
a tolerance of 1.0  10 15 .
Table 3 shows the table of fund allocations to be
(IPentium M, 1.80GHz, 0.99GB RAM). The made by the management board of Capital bank
program converged at iteration 12,757 to as well as bad debt amount.
produce the final results shown below;
Using the above allocations, Capital Rural Bank
The appendix shows the programming code and
Ltd. could realize a turn-over of
manual calculation confirming the program
GH¢5,961,300.00 on loans as against GH¢
output.
3,653,570.00 made in the previous year. This
The results for the six (6) basic variables are: represents 61.3% increase in turn-over.

y 1  0 . 0476 , y 2  0 . 0220 , y 3  0 . 0565 , The Pearson correlation coefficient between the

y 4  0 . 0254 , y 5  0 . 0003 , y 6  0 . 0390 allocated amount and the bad debt amount was r
= 0.026, which shows very low correlation
between the allocated amount and the probability
Using the transformation xi  101 yi , the values
of bad debt. This shows that the allocations were
of the variables for the main Linear Programming influenced more by policy than the probability of
problem are calculated as: bad debt. Agricultural loan, with the highest bad
debt probability, was allocated the least amount
x1  101y1  101(0.0476)  4.8076 of 30,300 while salary loan with the next bad
debt probability of 0.10 obtained 5,706,500.00.
x2  101y2  101(0.0220)  2.2220
This is because 60% of the total loan funds were
allocated to salary, funeral, and commercial
x3  101y3  101(0.0565)  5.7065 loans while only 15% of the total funds were to
be used for agricultural and funeral loans.
x4  101y4  101(0.0254)  2.5654 Because funeral loan shared in both the 60%
and 15% allocations, it got a larger share from
x5  101y5  101(0.0003)  0.0303 the 60% allocation and squeezed the agricultural
loan into smaller share of the joint 15% policy
x6  101y6  101(0.0390)  3.9390
allocation. Why should the allocation be
influenced by the policy? This shows that Banks
The optimal objective function value is
with weak policy always suffer and cannot
Z  0.3622 x1  0.3192 x2  0.3464 x3  maximize their profit on loans hence policy
should be backed by scientific research before it
0.2852 x4  0.1560 x5  0.2025 x6 is put in use.

Table 3. Summary of results for the allocation of funds

Variable Loan Type Probability of Amount to be Bad debt amount

bad debt allocated (GH¢) (GHc)
x1 Commercial Loan 0.020 4,807,600.00 96,152
x2 Funeral Loan 0.030 2,222,000.00 66,660
x3 Salary Loan 0.010 5,706,500.00 57,065
x4 Susu Loan 0.055 2,565,400.00 141,097
x5 Agricultural Loan 0.150 30,300.00 4,545
x6 Housing Loan 0.075 3,939,000.00 295,425

7
 Donkor et al.; JEMT, 21(5): 1-13, 2018; Article no.JEMT.40292

5. CONCLUSION 3. Cohen and Hamer. Linear programming

and optimal Bank asset management
Optimizing the disbursement of funds available decision. Journal of Finance. 1972;22(2):
for loans from Capital Rural Bank Limited would 147-165.
result in the appropriate allocation of funds to 4. Klaassen P. Financial asset-pricing theory.
their customers. The disbursement is as follows: Management Science-
Commercial loan = GH¢4,807,600.00, Funeral pubsonline.informs.org; 1998.
loan = GH¢2,222,000.00, Salary loan = 5. Dantzig GB. Reminiscences about the
GH¢5,706,500.00, Susu loan = origins of linear programming.
GH¢2,565,400.00, Agricultural loan = Mathematical Programming (R.W. Cottle,
GH¢30,300.00 and Housing loan = M.L. Kelmanson, B. Korte, eds.),
GH¢3,939,000.00. The result shows that the Proceedings of the International Congress
bank would be able to make a maximum turn- on Mathematical Programing, Rio de
over of GH¢5,961,300.00 on loans as against Janeiro. 1984;105-112.
GH¢ 3,653,570.00 made in the previous year. 6. Bazara (1984) and Hamdy (1992).
This represents 61.3% increase in turnover. It Karmarkar’s Projective Algorithm.
has also been determined that fund allocation 7. Karmarkar N. A new polynomial-time
was influenced mostly by policy and not the algorithm for linear programming,
probability of bad debt. However the bank should Combinatorica. 1984;4:373–395.
include efficient loan recovery methodology in 8. Wikimedia Foundation Inc. Karmarkar’s
their policy so that the limit of 4.5% total bad debt Algorithm.
ratio will not be exceeded if not reduced. To Available:http://en.wikipedia.org/wiki/Karm
reduce bad debts, effective monitoring of loan arkar% 27s_algorithm
facilities through field visits and on regular basis (Accessed January, 2018)
and ensure that loans are in compliance with the 9. Quintana VH. Interior-point methods and
terms and conditions of the facility, and identify their applications; 2000.
potential problems for action to be taken. This 10. Ferris and Philpott. The performance of
would pre vent diversion of funds into business Karmarkar’s algorithm. Journal of the
ventures other than the agreed purposes and Operations Research Society; 1988.
also help loan officers assist customers who are 11. Kojima M. A primal-dual interior point
facing some business management problems algorithm for linear programming; 1989.
such as improper records keeping, and 12. Todd MJ. The affine-scalling direction for
overtrading that affect their business operations. linear programming is a limit of projection
scalling directions. Cornell University,
COMPETING INTERESTS Upson Hall Ithaca, New York; 1991.
13. Monteiro RDC. Interior Path following
Authors have declared that no competing primal-dual algorithms; 1989.
interests exist. 14. Bayer and Lagarias. Affine and projective
Scaling Trajectories. Brealey RA, Hodges
REFERENCES
SD. Playing with portfolios. The Journal of
1. Abbas TE, Majid S. Ank Loan Portfolio, Finance, 1972;30(1):125-134.
an MOLP Based Modeling. International 15. Portfolios - An empirical analysis. Journal
Journal of Finance and Economics. of Finance. 25(3):639-649.
2011;25:28–41. 16. Dhamija AK. Karmarkar’s algorithm for
2. Booth GG. Programming bank portfolios linear programming problem; 2012.
under uncertainty. Journal of Bank www.slideshare.net/akdhamija/karmarkars-
Research. 1972;3(4):28-40. algorithm-for-linear-programming

8
 Donkor et al.; JEMT, 21(5): 1-13, 2018; Article no.JEMT.40292

APPENDICES

APPENDIX A: Converting the LP Model of the Loan Optimization Problem Formulation into
Karmarkar’s Standard Form

The Standard Linear Programming form is as shown below:

Maximize
Z  0 . 3622 x1  0 . 3192 x 2  0 . 3464 x 3  0 . 2852 x 4  0 . 1560 x 5  0 . 2025 x 6

Subject to
x 1  x 2  x 3  x 4  x 5  x 6  20
x 1  x 2  x 3  12
 0 .5 x 2  0 .5 x 3  x 6  0
 0 .4 x1  x 4  x 5  0 .4 x 6  0
x2  x5  3
 0 . 025 x 1  0 . 015 x 2  0 . 035 x 3  0 . 010 x 4  0 . 105 x 5  0 . 030 x 6  0
x1 , x 2 , x 3 , x 4 , x 5 , x 6  0

 Writing the dual of the given primal problem:

Minimize 20 w 1  12 w 2  3 w 5

Subject to :
w1  w 2  0 .4 w 4  0 .025 w 6  0 .3622

w1  w 2  0 .5 w 3  w 5  0 .015 w 6  0 .3192

w1  w 2  0 .5 w 3  0 .035 w 6  0 .3464

w1  w 4  0 .010 w 6  0 .2852

w1  w 4  w 5  0 .105 w 6  0 .1560

w1  w 3  0 .4 w 4  0 .030 w 6  0 .2025

w1 , w 2 , w 3 , w 4 , w 5 , w 6  0

 Introduction of slack and surplus variables to the constraints of the primal and the dual
respectively and then combine them

x 1  x 2  x 3  x 4  x 5  x 6  x 7  20
x 1  x 2  x 3  x 8  12
 0 .5 x 2  0 .5 x 3  x 6  x 9  0
 0 . 4 x 1  x 4  x 5  0 . 4 x 6  x 10  0
x 2  x 5  x 11  3
 0 .025 x1  0 .015 x 2  0 .035 x3  0 . 010 x 4  0 .105 x5  0 . 030 x 6  x12  0

 Addition of bounding constraint with slack variable s :

12 12

i 1
xi  i 1
wi  s  k

k  100
12 12

  
xi  w i  s  100
i 1 i 1

9
 Donkor et al.; JEMT, 21(5): 1-13, 2018; Article no.JEMT.40292

w 1  w 2  0 . 4 w 4  0 . 025 w 6  w 7  0 . 3622

w 1  w 2  0 . 5 w 3  w 5  0 . 015 w 6  w 8  0 . 3192

w 1  w 2  0 . 5 w 3  0 . 035 w 6  w 9  0 . 3464

w 1  w 4  0 . 010 w 6  w 10  0 . 2852

w 1  w 4  w 5  0 . 105 w 6  w 11  0 . 1560

w 1  w 3  0 . 4 w 4  0 . 030 w 6  w 12  0 . 2025

0 . 3622 x 1  0 . 3192 x 2  0 . 3464 x 3  0 . 2852 x 4  0 . 1560 x 5  0 . 2025 x 6  20 w 1  12 w 2  3 w 5

w i  0, xi  0 i  1 , 2 , 3 ,..., 12

 Homogenized equivalent system with dummy variable d :

x 1  x 2  x 3  x 4  x 5  x 6  x 7  20 d  0
x1  x 2  x 3  x 8  12 d  0
 0 .5 x 2  0 .5 x 3  x 6  x 9  0
 0 .4 x 1  x 4  x 5  0 . 4 x 6  x 10  0
x2  x5  x 11  3 d  0
 0 . 025 x 1  0 . 015 x 2  0 . 035 x 3  0 . 010 x 4  0 . 105 x 5  0 . 030 x 6  x 12  0
w 1  w 2  0 . 4 w 4  0 . 025 w 6  w 7  0 . 3622 d  0

w 1  w 2  0 . 5 w 3  w 5  0 . 015 w 6  w 8  0 . 3192 d  0

w 1  w 2  0 . 5 w 3  0 . 035 w 6  w 9  0 . 3464 d  0

w 1  w 4  0 . 010 w 6  w 10  0 . 2852 d  0

w 1  w 4  w 5  0 . 105 w 6  w 11  0 . 1560 d  0

w 1  w 3  0 . 4 w 4  0 . 030 w 6  w 12  0 . 2025 d  0

0 . 3622 x 1  0 . 3192 x 2  0 . 3464 x 3  0 . 2852 x 4  0 . 1560 x 5  0 . 2025 x 6  20 w 1  12 w 2  3 w 5  0

12 12

i 1
xi  
i 1
w i  s  100 d  0
12 12

i 1
xi  
i 1
w i  s  d  101

w i  0, xi  0 i  1 , 2 , 3 ,..., 12

 Introduction of transformations:

x i  ( k  1) y i  x i  101 y i i  1, 2 ,3 ,..., 12
w i  ( k  1) y 12  i  w i  101 y 12  i i  1 , 2 , 3 ,..., 12
s  ( k  1 ) y 25  s  101 y 25
d  ( k  1) y 26  d  101 y 26

 Using the above transformations, the system becomes:

101( y1  y 2  y3  y 4  y5  y6  y 7  20 y 26 )  0
101 ( y 1  y 2  y 3  y 8  12 y 26 )  0
101 (  0 . 5 y 2  0 . 5 y 3  y 6  y 9 )  0
101 (  0 . 4 y 1  y 4  y 5  0 . 4 y 6  y 10 )  0
101 ( y 2  y 5  y 11  3 y 26 )  0
101(0.025 y1  0.015 y 2  0.035 y 3  0.010 y 4  0.105 y 5  0.030 y 6  y12 )  0

10
 Donkor et al.; JEMT, 21(5): 1-13, 2018; Article no.JEMT.40292

101( y13  y14  0.4 y16  0.025 y18  y19  0.3622y 26 )  0

101( y13  y14  0.5 y15  y17  0.015y18  y 20  0.3192y 26 )  0

101( y13  y14  0.5 y15  0.035y18  y 21  0.3464y 26 )  0

101( y13  y16  0.010 y18  y 22  0.2852y 26 )  0

101( y13  y16  y17  0.105y18  y 23  0.1560y 26 )  0

101( y13  y15  0.4 y16  0.030 y18  y 24  0.2025y 26 )  0

101(0.3622y1  0.3192y 2  0.3464y 3  0.2852y 4  0.1560y 5  0.2025y 6  20 y13  12 y14  3 y17 )  0

12 24
101 (  y i   y i  y 25  100 y 26 )  0
i 1 i  13
12 24
101 (  y i   y i  y 25  y 26 )  101
i 1 i  13

 The system is further simplified to become:

y1  y 2  y 3  y 4  y 5  y 6  y 7  20 y 26  0 y 1  y 2  y 3  y 8  12 y 26  0
 0 .5 y 2  0 .5 y 3  y 6  y 9  0
 0 . 4 y 1  y 4  y 5  0 . 4 y 6  y 10  0
y 2  y 5  y 11  3 y 26  0
0.025 y1  0.015 y 2  0.035 y 3  0.010 y 4  0.105 y 5
 0.030 y 6  y12  0
y13  y14  0 .4 y16  0 .025 y18  y19  0 .3622 y 26  0

y13  y14  0 .5 y15  y17  0 .015 y18  y 20  0 .3192 y 26

0
y13  y14  0 .5 y15  0 .035 y18  y 21  0 .3464 y 26  0

y13  y16  0 .010 y18  y 22  0 .2852 y 26  0

y13  y16  y17  0 .105 y18  y 23  0 .1560 y 26  0

y13  y15  0 .4 y16  0 .030 y18  y 24  0 .2025 y 26  0

0 .3622 y1  0 .3192 y 2  0 .3464 y 3  0 .2852 y 4  0 .1560 y 5 

0 .2025 y 6  20 y13  12 y14  3 y17  0

25

i 1
y i  100 y 26  0
26


i 1
yi  1

yi  0
, i  1, 2 , 3 ,..., 26

An artificial variable y27 is introduced and the Karmarkar’s form is :

Minimize M  y27

Subject to:

y 1  y 2  y 3  y 4  y 5  y 6  y 7  20 y 26  13 y 27  0 y 1  y 2  y 3  y 8  12 y 26  8 y 27  0
 0 . 5 y 2  0 . 5 y 3  y 6  y 9  y 27  0

11
 Donkor et al.; JEMT, 21(5): 1-13, 2018; Article no.JEMT.40292

 0 . 4 y 1  y 4  y 5  0 . 4 y 6  y 10  2 . 2 y 27  0
y 2  y 5  y 11  3 y 26  0
0.025 y1  0.015 y 2  0.035 y 3  0.010 y 4  0.105 y 5  0.030 y 6  y12  1.07 y 27  0
y13  y14  0.4 y16  0.025 y18  y19  0.3622y26  0.2128y27  0

y13  y14  0.5 y15  y17  0.015 y18  y20  0.3192y26  1.1658y27  0

y13  y14  0.5 y15  0.035 y18  y21  0.3464y26  0.1186y27  0

y13  y16  0.010 y18  y22  0.2852y26  0.7248y27  0

y13  y16  y17  0.105 y18  y23  0.1560y26  1.949 y27  0

y13  y15  0.4 y16  0.030 y18  y24  0.2025y26  1.5725y27  0

0.3622 y1  0.3192y2  0.3464y3  0.2852y4  0.1560y5  0.2025y6  20 y13  12 y14  3 y17  33.3285y27  0

25


i 1
y i  100 y 26  75 y 27  0
27


i 1
yi  1

yi  0
, i  1 , 2 , 3 ,..., 27

12
 Donkor et al.; JEMT, 21(5): 1-13, 2018; Article no.JEMT.40292

APPENDIX B: MATLAB Code to Solve the Karmarkar’s Standard form in Appendix B

The following Matlab Code for Karmarkar’s Algorithm was written by the authors and was used to
solve the Karmarkar Formulation for the Loan Optimization Problem

% A is m x n matrix
% C is a column vector
% E is a row vector with 1’s as its entries
% k is the number of iterations
% tol is the tolerance
A=input(‘Enter the matrix A:’)
C=input(‘Enter C:’)
E=input(‘Enter E:’)
mn =size(A)
m=mn(1)
n=mn(2)
V1= ([E]/n)'
I=eye (n)
r=1/sqrt(n*(n-1))
α= (n-1)/(3*n)
V= ([E]/n)'
tol=10^(-15)
k=0
while(C'*V>tol)
D=diag(V)
T=A*D
P= [T; E]
Q=C'*D
R= (I-P'/(P*P' )*P)*Q'
RN=norm (R)
Y=V1-(r*α)*(R/RN)
V= (D*Y)/(E*D*Y)
M=C'*V
k=k+1
end
_________________________________________________________________________________
© 2018 Donkor et al.; This is an Open Access article distributed under the terms of the Creative Commons Attribution License
(http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium,
provided the original work is properly cited.

Peer-review history:
The peer review history for this paper can be accessed here:
http://www.sciencedomain.org/review-history/24824

13

View publication stats