International Journal of Strategic Decision Sciences, 2(3), 1-17, July-September 2011 1

A Logit Model for Budget

Allocation Subject to
Multi Budget Sources
Saeed A. Bagloee, Parsons Overseas Limited, UAE
Christopher G. Reddick, The University of Texas at San Antonio, USA

ABSTRACT
In a complex and extended system such as a government, the proper allocation of the budget to its sub-entities
is always a major challenge. As such for cases like governments, a situation in which multiple budget sources
with different concerns available to the sub-entities is common. This study develops an applicable model for
large-scale cases in which identifying the flow of capital or budget from (multiple) sources to the sub-entities
is sought. Since the influential factors to the allocation process may be mingled with some unknown param-
eters (as well as known factors) a logit model is developed from past panel data. The logit model is based
on the concept of utility, which quantifies the advantage of approaching budget-sources for the sub-entities.
Then the budget allocation problem of logit form is written as a mathematical programming formulation for
which Successive Coordinate Descent (SCD) method is proposed as the solution algorithm. In this paper, the
proposed methodology is tested numerically. The results of this study show there is strong evidence that some
of the entities’ properties can be altered in order to achieve a better budget allocation.

Keywords: Budget Allocation, Logit Model, Mathematical Programming, Public Sector, Successive
Coordinate Descent (SCD)

INTRODUCTION complexity could be exacerbated as the size of

Budget allocation always is a major challenge
to the success of any government. There are
several practices handling these challenges
advocating by different types of stakeholders.
The variety of opinions in any political system
and lack of systematic solutions implies that the
problem is more of a general issue and differs
in various environments, with many unknown
but influential parameters. This generality and
DOI: 10.4018/jsds.2011070101
entities increases and may create a situation in
which multi budget sources are available to the
sub entities. Governments are entangled within
multi budget sources situations:
• On one hand, there are several budget
sources with different and various concerns
and criterions to release their written budget
to the relevant sub-entities. These budget
sources are focused on various fields or
purposes.

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
2 International Journal of Strategic Decision Sciences, 2(3), 1-17, July-September 2011
• On the other hand, the sub-entities may be
qualified and entitled to approach various
budget sources due to their broad, multi-
field, or multi-purpose activities.
Capital allocation in banking system can
be a clear application of the defined budgeting
problem. For instance, imagine a corporate
(sub-entity) that is interested in developing a
cotton mill. This activity may attract investment
attention from agricultural banks as well as
industrial banks; each has their own criterion
and concerns spanning interest rates to required
collateral or security policies. Thus the afore-
mentioned corporation evaluates the utility (or
disutility) of the banks in order to decide about
its borrowing portfolio. The utility is defined as
combination of different influential parameters
with various impacts. In the above case, inter-
est rates, cost of borrowing, required securities
(or needed collateral) are the most known and
influential parameters. However we believe that
the complexity of the problem as introduced
before is emerged from the fact that there are
some unknown parameters affecting the deci-
sion process, especially for government. All of
these arguments indicate usage of the discrete
choice models belong to the consumers’ theory,
in which the customers choose the appropriate
product according to their maximum expected
utility (in our cases, customers and products are
the corporate and the banks’ loan respectively).
The most widely used and well-recognized
discrete choice models are logit models. Logit
models are in form of a linear equation of various
variables (parameters) with different weights
(impact). These functions must be calibrated
according to the past observations pertaining
to the consumers’ behavior. In different cases,
panel data could be the past fiscal budget
schemes for the states. In this study, a sample
survey of a bank’s customers provides raw data
for calibration in this study.
Given the utility functions (as an exog-
enous element to the algorithm) a mathematical
programming framework as well as a solution-
finding algorithm is developed. In this study,
we have developed an intuitive utility functions
Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
for the adopted case-study. We introduce a
case-study of Bank-Loan situation, in which
the interest rates of the banks products (loan) is
a main factor in the utility of the loan-seekers.
Since one of the main objectives of this study
is to tackle the budgeting problem for real size
cases, a large scale dataset for the case-study
with highly unpredictable and unbiased char-
acteristics was developed.
As a result, the contribution of each
budget-source to the sub-entities’ needs can
be examined. Obviously in the case of budget
deficit the capital needs of all the sub-entities
will not be supplied. The budget deficit cases
can be easily addressed by introducing a dummy
budget-source associated with capacity of infin-
ity and high disutility. This makes the dummy
source the last resort for the sub-entities. A
similar structure has been introduced by Spiess
(1996), in which traffic trips are split into two
trips distinguished by intermediate park-ride
destinations.
In addition, this study provides a model
to watch the capital flow so as to identify
the bottlenecks. Easing (or tightening) some
parameters such as interest rate in the case of
corporate/bank and providing more/less budget
with the budget-sources to facilitate the capital
flow, are of the expected result of the proposed
methodology. The latter is elaborated as an
application of the methodology in the form of
a numerical test.
The paper is organized as follows. In the
second section of this paper there is a litera-
ture review. Following this the third section
demonstrates the formalized methodology.
The fourth section is dedicated to the adopted
solution algorithm. The fifth section presents
the numerical tests and discusses the implica-
tions of the solutions and results. The seventh
section summarizes and concludes the study.
LITERATURE REVIEW
One of the most perennial problems that
governments face is the optimal allocation of
budgets given scarcity of resources (Groves &
 International Journal of Strategic Decision Sciences, 2(3), 1-17, July-September 2011 3

Ledyard, 1977). There is always competition (Li, 1977). However there is no research, that
and tradeoffs in the budget allocation process we are aware, that applies the logit model to
that governments must contend with. Existing the optimal budget application.
research discusses the efficiency of the bud-
get allocation process as it deals with budget
maximizing behavior (Groves & Ledyard, 1977; METHODOLOGY
McGuire, Coiner, & Spancake, 1979; Kosenok FORMULATION
& Severinov, 2008). There also is research
A set of sub-entities and set of budget-sources
that deals with the impact of the budget on
denoted by P and Q respectively are given.
the economy as a whole (Iqbal & Turnovsky,
The utility of the sub-entities approaching to
2008). Existing research has examined the
the budget-sources may be stated as follows:
types of budget reforms and their economic
and political influence on budget allocation
(Dorta-Velazquez, Leon-Ledesma, & Perez- U pq = ∑ wi . u pq
i
+ ∈pq ∀p ∈ P , ∀q ∈ Q
Rodriguez, 2010). There is also research that i
(1)
examines the politics of the budget allocation
process (Tridimas, 2001). Finally, research ex-
amines the budget allocation among federal and where U pq is the expected utility of approach-
regional systems (Atlas, Gilligan, Hendershott, ing budget-source q for sub-entity p . The
& Zupan, 1995; Hajkowicz, 2007). utility consists of two arguments: known pa-
However, there is little research at exam- rameters and the corresponding impacts
ines the budget from multiple sources and the (å i wi . u pq
i
) and unknown factors (Îpq ) . The
optimization of the allocation (Clark, 1997;
unknown factors are aggregated as a random
Zeynalian, Jandaghi, Memariani, & Jahanshahi,
variable indicating error term of the utility
2010). Some of the early research has examined
linear programming for budgeting and capital function. u pq
i
is the i-th (dis)utility factor for
projects (Ijiri, Levy, & Lyon, 1963; Garrison, sub-entity p in approaching budget-sourceq
1971; Lee & Clayton, 1972). Measuring the associated with wi a weight or impact factor.
proper allocation of the budget is an important wi determines the fate of u pq
i
: if wi > 0
issue to research since measuring it incorrectly
can have tremendous influence on the social means u pq
i
is a welcome characteristics hence
welfare of society (Jack, 2008; Ben-David & i
u pq is “utility” otherwise it is called “disutility”.
Tavor, 2011). Given the utility of the choices, probability
This study is different and makes a contri- of selecting a specific choice out of the choice
bution to the literature by using a logit model to set may be defined as follows:
determine the optimal budget allocation. This
paper uses a simulation of budget allocation,
Prp (q ) = Pr{U pq > U pq ′ } ∀q ≠ q ′, ∀q, q ′ ∈ Q
which is consistent with some of the earlier re-
search in the field to solve a large and complex (2)
problem such as this (Crecine, 1967; Rivers &
Oxner, 1982; Dellaert, Jeunet, & Mincsovics, By substituting Equation (1) into Equation
2011). There is research that provides a gen- (2) we have:
eral specification of this model with respect
to consumer theory (Anderson, De Palma, & Prp (q ) = Pr{∈pq ′ − ∈pq < ∑ wi . u pq
i
− ∑ wi . u pq
i
}
'
Thisse, 1988), agricultural economics (Morey, i i

Rowe, & Watson, 1993), and home ownership (3)

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
4 International Journal of Strategic Decision Sciences, 2(3), 1-17, July-September 2011

Lemma 1 - Eqs(5) holds at the optimal solu-

If ∈pq ′ , ∈pq are assumed distributed
tion of Eqs(6,7
Weibull, the differences (∈pq ′ − ∈pq ) would
be of logistic distribution function and a Proof 1 - By introducing ap as lagrangian
model known as multinomial logit (here simply
coefficients (or dual variables) for con-
called logit) is derived as follows:
straints (7), the lagrangian of the above
problem is
i
exp(∑ wi . u pq )
Prp (q ) = i
(4)
∑ exp(∑ wi . u pqi ' )
i
q′

For the sake of simplicity, by omitting the (8)
error terms and assuming U pq = ∑ i wi . u pq i

By establishing the first order condition of

and the total capital need of ∀p ∈ P as G p we optimality better known as Kuhn-Tucker the
would have: aforementioned equivalency of the logit form
and the Equation (6), (7) is proved:
exp(U pq )
g pq = G p . Prp (q ) = G p .
∑ exp(U pq ′
)
q′ (9)
(5)

Where g pq is capital flow, means the

amount of obtained capital of sub-entity p from
budget-source q . (10)
Hereafter in order to simplify the formula-
tion without losing the generality of argument, Substituting Eqs(10) into (9) yields Eqs(5)
we denote U pq with a single terms while (note U pq = −u pq ). End of Proof 1
wi = −1 and then we discard the superscript The introduced mathematical program-
i from the utility function thus we have ming must adopt a set of new constraints to
U pq = −u pq . Adopting -u pq implies u pq as a take the capacity of each budget-source
(C q , ∀q ∈ Q ) explicitly into consideration.
disutility factor such as interest rate of approach-
ing to q applied to p . As shall be shown, This can be done by adding the following
equation (5) is equivalent to solving the fol- constraint to Eqs(5,6:
lowing mathematical programming:
∑g pq
≤ Cq q ∈Q (11)
Min ∑ g pq ( n g pq − 1 + u pq ) (6) pq

pq

S.t.: By introducing bq ³ 0 as dual variables

∑g pq
= Gp p ∈ P, q ∈ Q for Eqs(11) in additional to ap for Eqs(7) and
q
(7) arranging the Kuhn-tucker optimality condition
for the new problem (Eqs(6, 7 and 11 we have
a similar equation to Eqs(9):

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
 International Journal of Strategic Decision Sciences, 2(3), 1-17, July-September 2011 5

g pq = exp(− u pq − αp − βq ) (12)
(16)
By substituting Eqs(12) into the lagrangian
form of problem (Eqs(6, 7 and 11 the final dual
bq .∂Z / ∂bq = 0 ⇒ bq .(−∑ g pq + C q ) = 0
problem is reached as follows: p
(17)

The SDC algorithm finds the optimum

(13)
For the sake of feasibility of the above
problem the amount of needed capital of the
sub-entities must be less than the amount of
available budgets with the budget-sources:
∑G p
≤ ∑Cq (14)
p q
In the case of budget deficit
( ∑ p G p > ∑ q C q ) a dummy budget sources
with capacity of ( ∑ p G p − ∑ q C q ) associ-
ated with infinity disutility factors
(i.e. u pq = ∞) .
The optimal value of bq are the shadow
prices of the corresponding budget-sources. It
implies how worthy are the depleted budget-
sources at the end. bq can be assumed as the
attractiveness of the budget-sources.
SOLUTION ALGORITHM
Problem Eqs(13) does not have any explicit
constraints. With no worry about the convexity
of the problem it can be solved by utilizing the
Successive Coordinate Descent (SCD) algo-
rithm efficiently (Luenberger, 1984). Again the
solution by iteratively solving Eqs(15 and 16.
Eqs(17) indicates that expectedly the shadow
price bq for undepleted/depleted budget-
sources is zero/positive.
In order to facilitate the SDC running and
to get rid of the exponential terms let’s have
the following substitution:
a p = exp(−ap ) ,bq = exp(−bq ) and
Ωpq = exp(−u pq )
For the later references let’s call Wpq con-
ductivity from p to q. The less u pq (less disutil-
ity) implies more Wpq which is welcome. Thus
Wpq conveys a positive message. Now the SDC
algorithm may be formulated as follows:
Step 0 – Initialization - set iteration counter
as i := 1 and initialize betas as
bq0 = 1 ∀q ∈ Q
Step 1 – Computing budget flows g pq at it-
eration i :
First compute a pi based on current value
of bqi-1 from Eqs(15):
a pi = G p / ∑ (bqi −1 . Ωpq ) ∀p ∈ P
q

Kuhn-tucker conditions guarantee the optimum (18)

solution which is:
With respect to Eqs(12) now compute g pq :

(15) i
g pq = a pi .bqi −1 . Ωpq ∀p ∈ P , ∀q ∈ Q
(19)

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
6 International Journal of Strategic Decision Sciences, 2(3), 1-17, July-September 2011
Step 2 – Termination Condition
With respect to Eqs(16) compute error
index denoted by Errq and defined
as total amount of violation capacity
constraints at budget-source q. If
Errq = ∑ g pq
i
−C q < e, ∀q ∈ Q then termi-
p
nate the algorithm.
Step 3 – Updating
With respect to Eqs(11) dual variable bqi
is updated as follows
bqi = min (1, bqi −1 . C q / ∑ g pq
i
),
p
Step 4 – Continuation
Set i := i + 1 and go to Step 1.
Proposed Methodology Versus
Hitchcock Transportation Problem
Figure 1 outlines the structure of a case study.
The left-first column of the dots represents the
sub-entities demanding capital denoted by G1..
G1000 and the last column represents the budget-
sources associated with budget capacities C1..
C100. The two columns are linked with some
arrows associated with conductivity rates
Ω pq = exp(−upq ) . The flow of capital on the
arrows is sought. The above arrangements
reminisce very well known problem in the field
of operational research known as Hitchcock
Transportation Problem (HTP). In Hitchcock
problem simply the minimum of the total in-
curred cost or friction or disutility to the capi-
tal flow ( å pq u pq.G pq ) is sought. The opera-
tional research literatures yield very efficient
solution algorithms to tackle the Hitchcock
problem even for very large size cases (Bradley,
Hax, & Magnanti, 1977).
Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
One may asks why we did not adopt the
Hitchcock problem despite the aforementioned
easy solving advantage. To address this concerns
it worth noting that the Hitchcock problem cor-
responds to cost minimization, where the cost
or friction (or based on our terminology; disu-
tility) are exactly known which is not our case
in this study. As discussed due to complexity
of the problem so many known and unknown
factors are influencing the sub-entities’ deci-
sions. Imagine a situation in which a single
entity with need of 100 seeks the budget with
two almost identical budget sources: one with
cost of 1.0 and the other with cost of 0.9. The
Hitchcock interpretation of the situation sug-
gests to providing the entire budget needs from
the source with lesser cost even if the costs
∀q ∈ Q
difference is marginal. However we intuitively
(20) recognize that this kind of decisive and absolute
situation in fuzzy real world environment is
impossible. Instead we would like to have a
behavioral approach, in which we do not have
exact knowledge of the costs, (dis)utilities and
frictions associated with the budget seeking
activities, rather the perceived frictions (or
disutilities) are statistically distributed. This
gives rise to a logit distribution for the choice
which is not only more realistic but also gives
solutions mathematically much more stable
(i.e., a small change in the cost (as seen in the
above example 0.9 versus 1.0) would cause
small change in the flow of capital).
Having the foundation of the methodology
laid in previous sections, the way is paved to
examine the methodology. In the next section,
a large-scale case-study is introduced and
developed to test the algorithm’s merit and
convergence efficiency in facing the real world
problem is examined.
NUMERICAL TESTING
A large-scale dataset comprising 100 budget-
sources and 1000 sub-entities looking for budget
is set up. The case study has three quantitative
features:
 International Journal of Strategic Decision Sciences, 2(3), 1-17, July-September 2011 7

Figure 1. A schematic outlook of the proposed structure for the problem

(i) Available capital at budget-sources, cumulative density function of exponential

(ii) Capital needs of the sub-entities and probability is as follows:
(iii) Sub-entities’ utility (or disutility) of ap-
proaching to budget-sources. 1 − e −x /l x ≥ 0
F (x ) =   (21)
 0 x < 0
The aforementioned features are speci-  
fied stochastically random, so as to avoid any
biased result of applying the methodology to In order to generate exponential random
the case study. numbers it is suffice to generate 0 £ r £ 1 a
In order to simulate a real situation, ex-
uniform random number and replace F (x ) in
ponential type of random numbers are utilized
Eqs(21) with r which results:
which yield a highly unpredictable dataset
(Leemis & Park, 2006). Exponential random
numbers bear the main property of exponen- x = −λ . n(1 − ρ) (22)
tial distribution probability function which is
memoryless, implying a highly unpredictable
Since 0 £ r £ 1 , (1 - r) also is a uniform
situation wherein the sequences of events are
random number, thus the above formulation
not depended upon the preceding events. The
can be shorten as

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
8 International Journal of Strategic Decision Sciences, 2(3), 1-17, July-September 2011
x = −λ . n ρ (23)
In Eqs(21..23), l is expected value of x ,
(l = E (x )) .
The above discussion in exponential
random numbers lays context to elaborate in
adopted procedure to specify the triple quan-
titative features:
Budget Available at
Budget-Sources and Capital
Needs of the Sub-Entities
We assume that average available budget at the
budget-sources (q ) and average budget needs of
the sub-entities (p) are 1000 and 100 unit re-
spectively. (i.e. lq = 1000 and lp = 100) .
Thus by applying Eqs(23) over 100 budget-
sources and 1000 sub-entities, the total available
budget and budget need for the sub-entities
accumulated on 92,258 and 105,778 unit re-
spectively. As explained before, a dummy
budget-source with infinity capacity of budget
is embedded as well to address the budget
deficit, which is amounted to 13,520 in the set
up dataset (i.e. 13520 = 105778 − 92258) .
Sub-Entities’ Utility
(Disutility) of Approaching
to the Budget-Sources
It is intuitively understandable that the main
factor contributing to the budget seeking enti-
ties’ decision is the interest rate at the budget
sources. Thus we first specify the interest rates
by assuming of having average of 5% and utiliz-
ing the exponential distribution:
%Rq = −5 . n r (24)
Second, we establish Fpq the friction or
disutility imposed on the sub-entities’ approach
to the budget-sources as follows:
Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
Fpq = %Rq + 2 . rpq .%Rq (25)
Where, the first term (%Rq ) suggests that
the utility on the path from p to q is highly
depended upon the interest rate at q. The second
term (2 . rpq .%Rq ) refers to the personal atti-
tudes of the budget seeking entities (repre-
sented by uniform random number rpq ) which
also is under influence of the interest rates of
the ahead budget sources. Knowing that the
expected values of rpq s is 0.5, in order to have
a unique dimension in both terms, rpq are
multiplied by 2 (i.e. 2. rpq » 1) . Since the
defined disutility in Eqs(25) is composed of
disutility on the approach to and at the ahead
budget-source (denote by subscript pq and q
respectively), we aggregated all the components
as single quantity named friction Fpq (note:
according to the previous notations, Fpq and
u pq are identical).
In order to make the dummy budget-source
as the last choice (in the case of budget deficit),
the interest rate is assumed a big number (say
å q %Rq which is %503.15).
Table 1 indicates a summary of the above
arrangements, in which the interest rates and
aggregation of the friction rates at the budget
sources have been shown.
Algorithm’s Implementation in the
Form of Matrices Operations
For the large size cases, efficient coding and
handling implementation of the algorithms
may become a serious prohibitive issue. In
this regards permutation of the implementa-
tion process to the format of matrix operations
would ease the process since the ordinary com-
mercial optimization software such as GAMS
and MATLAB are able to accommodate matrix
 International Journal of Strategic Decision Sciences, 2(3), 1-17, July-September 2011 9

formatted of the cases. Fortunately the solution b u d g et s o u r ces in Eq s ( 2 0 ) . ( N o te:

algorithm can be easily interpreted as a series of Ω p×q = exp(−Fpq ) ) . Also the outputs are: Gp×q
matrix calculation with simple operations. The
matrix of capital distribution and B1×q vector of
core component of the computation efforts are
beta indices with budget-sources in Eqs(26, 27.
Eqs (18) and Eqs (19) which may be written in
By having the above arrangements the algorithm
the matrix format as follows:
is launched. Due to adopting a mathematical
programming form, knowing the accuracy of
Eqs(18) ⇒ Ap×1 ⊗ ( B1×q * Ω−p×1q )−1 =G p×1 achieved solution and the computational time
(26) is important, especially when it aims to address
the large size datasets. Figure 2 depicts the
Eqs(19) ⇒ Ap×1 * B1×q ⊗ Ωp×q =G p×q algorithm’s convergence over the successive
iterations, in which error index denotes the
(27)
total amount of capacity constraint violation
(å q Errq ) .
Where -1 means matrix transposing op-
eration and *, Ä mean regular matrix multi- Figure 2 shows the proposed algorithm
rapidly converges, so that the total amount of
plication and entry-by-entry multiplication
capacity constraint violation at iteration 100
operand respectively.
stood at 0.24. Table 1 summarizes the results
As the size of the data set increases, even
of the above settings called “Initial Setting”
coding the problem can become prohibitive.
(for the interest of abbreviation only odd records
Thus, due to similarity of the proposed problem
of q are shown). The cash distribution rates
to the Hitchcock problem, we encrypted the
multiplied by the friction rates
problem as a pseudo-transportation network
can be interpreted as some form of sources’
(Figure 1) by using EMME/3 transportation
dislike (or cost or disutility) imposed on the
planning software. Since EMME/3 is meant
budget seeking entities, that we call total
for transportation applications, the coding and
source’s dislike (TSD; the less TSD the better).
encrypting of the dataset has been eased con-
Therefore the system dislike can be defined as
siderably by utilizing some specific transpor-
the aggregation of the sources’ dislike:
tation features and terminologies. Sub-entities
. According to
and budget sources were coded as origin and
destination nodes and the arrows with friction Table 1 the å q TSDq of the initial setting
are coded as link with costs in the EMME/3. scenario is 9,113. Apart from
Origin, destination, node, link and cost are the analyzing the beta rates is of great interest since
specific terminologies used in the transporta- it shows how smoothly the cash has been dis-
tion literature. In additional, EMME/3 is highly tributed among the budget-sources.
capable of handling large scale matrix opera-
tions. Therefore the algorithm was prepared as Beta Rates Analysis
a macro feature of EMME/3 which calls the
matrix calculation modules (macro in EMME/3 Figures 3 through 6 depict how the betas vary
is a set of commands). against frictions, interest rates and capacity of
the budget-sources. Figure 3 shows as the size
Executing the Algorithm of the budget-source increases (more capacity)
the beta (price of the budget-source) rapidly
The macro requires three matrices as the inputs: decreases, however a relatively strong direct
conductivity matrices Ω p×q and also Gp×1 vec- correlation between beta and capacity was
observed (R2=0.697). Figure 4 demonstrates
tor of capital needs of the sub-entities in Eqs(26,
betas versus the total frictions ended up to the
27 as well as Cq vector of available capital with
budget-sources . Total friction can

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
10 International Journal of Strategic Decision Sciences, 2(3), 1-17, July-September 2011

Table 1. Algorithm’s execution result for the large-scale case study

Initial setting Revised budget setting

q %R
1 12.75 195.727 496.434 4.105 127.496 703.417 3.749 180.657
3 8.497 128.707 459.675 4.267 78.87 703.001 3.847 120.58
5 0.634 9.35 338.339 4.739 4.304 694.673 4.012 8.835
7 3.146 48.153 4682.31 2.056 299.897 2569.143 2.661 164.524
9 13.151 194.786 192.628 5.059 48.248 657.302 3.823 164.646
11 1.331 19.867 374.031 4.622 9.706 699.545 4 18.167
13 3.432 53.042 98.07 5.92 6.871 705.498 3.939 49.428
15 1.397 21.281 782.358 3.881 22.558 821.539 3.836 23.689
17 9.214 139.731 1068.793 3.419 191.319 913.526 3.568 163.591
19 3.701 56.553 1144.309 3.453 86.768 955.054 3.638 72.471
21 4.569 69.002 2418.986 2.694 217.821 1508.228 3.158 135.802
23 3.628 54.133 174.411 5.339 12.408 679.242 3.984 48.309
25 10.393 155.724 291.263 4.69 60.295 665.701 3.856 137.86
27 8.831 133.996 372.481 4.479 63.44 682.913 3.877 116.309
29 6.916 103.451 3498.689 2.279 473.028 2002.061 2.829 270.632
31 1.015 15.575 1942.407 2.979 41.083 1301.542 3.384 27.543
33 6.777 101.123 951.43 3.585 125.125 873.252 3.662 114.864
35 2.422 37.24 285.408 4.869 14.004 682.623 4.001 33.515
37 0.354 5.417 67.734 6.353 0.498 738.345 3.956 5.431
39 4.737 72.293 1359.905 3.261 129.108 1041.562 3.532 98.856
41 5.39 80.396 150.198 5.456 16.108 680.11 3.937 72.941
43 0.825 12.571 863.304 3.794 14.95 852.365 3.811 14.758
45 10.368 155.644 64.251 6.207 12.973 720.414 3.782 145.473
47 15.653 238.546 988.642 3.364 298.289 867.352 3.5 261.598
49 4.986 73.59 3345.756 2.36 331.758 1934.576 2.9 191.735
51 1.429 21.289 130.67 5.671 3.695 694.187 4.006 19.636
53 13.526 207.226 741.59 3.696 196.326 780.641 3.636 206.699
55 6.749 100.417 221.887 5.036 29.373 669.381 3.936 88.616
57 0.414 6.203 897.73 3.768 7.546 866.695 3.795 7.288
59 6.615 99.956 366.385 4.535 48.104 686.074 3.913 90.052
61 4.396 65.554 229.528 5.051 20.137 674.865 3.965 59.199
63 5.682 86.445 1138.338 3.418 130.876 948.188 3.606 109.059
65 1.99 29.756 293.273 4.854 11.838 684.892 3.998 27.653
67 1.198 18.221 491.937 4.349 11.94 728.225 3.962 17.683
69 0.574 8.754 471.227 4.409 5.634 724.523 3.971 8.665

continued on following page

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
 International Journal of Strategic Decision Sciences, 2(3), 1-17, July-September 2011 11

Table 1. continued

71 0.753 11.164 593.844 4.17 8.878 759.329 3.929 11.358

73 6.986 104.86 2081.364 2.794 289.726 1350.506 3.218 188.067
75 12 180.951 701.684 3.784 160.772 770.446 3.695 176.49
77 10.3 151.448 209.87 5.029 41.329 662.596 3.871 130.551
79 5.421 81.071 1974.507 2.877 209.425 1306.281 3.295 138.488
81 1.011 15.097 594.871 4.167 12.238 759.51 3.915 15.629
83 3.141 46.819 69.314 6.27 4.372 729.931 3.92 46.029
85 10.788 159.658 986.097 3.469 206.715 877.723 3.577 183.976
87 3.649 54.421 3144.213 2.446 226.399 1843.341 2.985 132.833
89 0.607 9.12 846.215 3.823 10.664 847.036 3.814 10.681
91 2.227 33.829 89.546 6.032 4.024 713.656 3.961 32.045
93 2.069 30.102 121.978 5.733 4.791 696.716 3.982 27.367
95 1.195 17.321 1835.271 3.033 43.195 1253.962 3.419 29.507
97 0.287 4.351 126.397 5.73 0.76 698.593 4.013 4.199
99 5.462 81.43 538.223 4.174 58.503 731.962 3.871 79.585
Total* 252.586 3801.361 45307.771 211.548 4434.185 46112.243 185.495 4483.569
101** 503.156 1340 92258.357 0 18116.438 92258.357 0 18116.227
* Total of all the records including odd and even records,
**Dummy Budget-source

Figure 2. Algorithm’s rapid convergence for the large-scale case study

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
12 International Journal of Strategic Decision Sciences, 2(3), 1-17, July-September 2011

be interpreted as how much the budget-sources

b 0q
are attractive regardless of the amount of the C q1 =C q0 / 2 + . ∑C 0
/2
budget-need. Figure 4 indicates there is no ∑ q b 0q q q

direct correlation between the betas and the

(28)
total friction, since the friction in contrary to
the beta does not take the budget flow (Gpq)
Where superscripts 0 and 1 refer to the
into consideration which is shown in Figure
initial and the new tries respectively. According
5. Considering the amount of budget and fric-
to Table 1, the amount of total available budget
tion together which was called total source’s
and betas are 92258.357
dislike (TSD) and shown in Figure 5 indicates
and 415.994 respectively. The algorithm’s result
a tangible correlation as it is expected: less
of the new try has been presented in Table 1 as
dislikeness infers more amount of beta. Figure
“Revised Budget Settings”.
6 depicts a poor correlation between the betas
A look at Table 1 suggests that the new
and the interest rates, since the interest rates
betas are in the range of 3 to 4 versus the range
represent disutility at only one end of interac-
of 2 to 7 for the initial setting. This indicates
tion between seekers-and-sources whereas the
that the redistribution has yield benefits in the
betas are looking at the both ends which are
sense of bring the betas close to each other.
captured by the definition of friction (Figure
Figures 7 through 10 demonstrate this variation
5). Figures 4 and 6 commonly highlight that
and sensitivity of the betas to other aforemen-
the betas are widely dispersed the wide disper-
tioned factors clearly. Overall examining Fig-
sion of the betas which means that the system
ures 7 through 10 and comparing them with
has an unfair distribution of budget among the
Figures 3 through 6 (initial setting) points out
budget-sources. In contrast we like to see an
that the new redistribution has cut the dra-
identical price for same products (which here
matic dispersion of the betas. Figure 7 shows
is the Budget). Therefore a better distribution
more strong correlation between beta and the
of the budget among the budget-sources may
budget capacity. The others (Figures 8, 9, and
be made by taking the betas into account. For a
10) yield similar results to the corresponding
new try half of the total budget is redistributed
elements in Figures 3 through 6. On the sideline
proportional to the corresponding beta rates as
the above redistribution of the budget slightly
follows:
decreases å q TSDq (1.55%). Despite having
a major achievement in improving the betas,

Figure 3. Beta and budget sources in the initial setting scenario

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
 International Journal of Strategic Decision Sciences, 2(3), 1-17, July-September 2011 13

Figure 4. Beta and budget sources’ total friction in the initial setting scenario

Figure 5. Beta and total source’s dislike in the initial setting scenario

Figure 6. Beta and budget-sources’ interest rates in the initial setting scenario

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
14 International Journal of Strategic Decision Sciences, 2(3), 1-17, July-September 2011

Figure 7. Beta and budget sources in the revised scenario

Figure 8. Beta and budget sources’ total friction in the revised scenario

Figure 9. Beta and total source’s dislike in the revised scenario

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
 International Journal of Strategic Decision Sciences, 2(3), 1-17, July-September 2011 15
Figure 10. Beta and budget-sources’ interest rates in the revised scenario
the inertia of the å TSDq q
index is high that
leads to only 1.55% improvement. This issue
was discussed in the previous section wherein
the proposed methodology was compared to
Hitchcock transportation problem. Since the
proposed methodology by adopting logit con-
cept is a behavioral and fuzzy (close-to-reality)
not a purely rigid mathematical methodology,
marginal changes in the system should not
change the whole system dramatically which
can be a case in Hitchcock problem. The
å q TSDq index can be compared to “Richter”
index to measure the earthquake in the sense
of its inertia. According to United States Geo-
logical Survey (2011) a unit difference in
magnitude of Richter is equivalent to a factor
of 31.6 in the energy released or destruction
power.
SUMMARY AND CONCLUSION
Supplying the capital needs of the budget-
seeking sub-entities in a large scale system
comprising of multiple budget sources is of
great interest to policy makers. On a government
level, budget allocation and in lesser extent, on
banking level, loan related activities can be of
such real applications. This problem has been
addressed by adopting a logit form of budget
seeking behavior of the sub-entities. Given the
Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
utility functions a mathematical programming
formulation was introduced which ensures that
the sub-entities would approach to the best fitted
budget-sources according to their best interest
(utility). The SCD algorithm is adopted as the
solution-finding algorithm to the established
mathematical programming. The methodology
and algorithms have been encrypted as a set of
simple matrix computational operations which
have helped the efficiency and easy-to-use of
the methodology even with ordinary com-
mercial software. The algorithm was tested
numerically. In order to challenge the algorithm
a large scale case with highly unpredictable
and unbiased characteristics was introduced
and developed. The case-study comprises 1000
sub-entities along with 100 budget-sources.
The algorithm was launched to the dataset
and quickly converged. Among the important
outputs, the shadow (or marginal) price of the
budget-sources called Beta can be pointed
out. The Betas contain important implication:
the more the betas are the more attractive the
budget-sources are to the customers, thus it may
yield some indication of how to (re)distribute
the capital among the budget-sources to have
a hassle-free flow of capital. This issue was
elaborated and numerically demonstrated. Iden-
tical prices for identical products-a fundamental
economic notion- was utilized to redistribute
budget among the budget-sources which led
16 International Journal of Strategic Decision Sciences, 2(3), 1-17, July-September 2011
to less friction on the way of the sub-entities’
approach toward the budget-sources.
ACKNOWLEDGMENTS
The authors are indebted to Dr Heinz Spiess,
whose insightful comments lead to the proof of
Lemma 1. The authors would like to thank the
anonymous reviewers and the editor for their
insightful comments and suggestions.
REFERENCES
Anderson, S. P., De Palma, A., & Thisse, J. F. (1988).
A representative consumer theory of the logit model.
International Economic Review, 29(3), 461–466.
doi:10.2307/2526791
Atlas, C. M., Gilligan, T. W., Hendershott, R. J., &
Zupan, M. A. (1995). Slicing the federal government
net spending pie: Who wins, who loses, and why.
The American Economic Review, 85(3), 624–629.
Ben-David, N., & Tavor, T. (2011). Measurement
of the social loss of wrong public budget allocation.
International Journal of Social Economics, 38(3),
209–217. doi:10.1108/03068291111105156
Bradley, S. P., Hax, A. C., & Magnanti, T. L. (1977).
Applied mathematical programming. Reading, MA:
Addison-Wesley.
Clark, D. J. (1977). Pressure and the division of a
public budget. Public Choice, 93(1-2), 179–195.
Crecine, J. P. (1967). A computer simulation model of
municipal budgeting. Management Science, 13(11),
786–815. doi:10.1287/mnsc.13.11.786
Dellaert, N., Jeunet, J., & Mincsovics, G. (2011).
Budget allocation for permanent and contingent
capacity under stochastic demand. International
Journal of Production Economics, 131(1), 128–138.
doi:10.1016/j.ijpe.2010.03.002
Dorta-Velazquez, J., Leon-Ledesma, J., & Perez-
Rodriguez, J. (2010). Models of municipal budget
allocation: Empirical data from Spanish munici-
palities. Public Budgeting & Finance, 30(2), 24–46.
doi:10.1111/j.1540-5850.2010.00955.x
Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
Garrison, R. H. (1971). Linear programming in capital
budgeting: The possibilities for capital budgeting
through sensitivity analysis of LP models are virtually
unlimited. Management Accounting, 52(10), 43–47.
Groves, T., & Ledyard, J. (1977). Optimal al-
location of public goods: A solution to the “Free
Rider” problem. Econometrica, 45(4), 783–809.
doi:10.2307/1912672
Hajkowicz, S. (2007). Allocating scarce financial
resources across regions for environmental man-
agement in Queensland, Australia. Ecological
Economics, 61, 208–216. doi:10.1016/j.ecole-
con.2006.10.011
Ijiri, Y., Levy, F. K., & Lyon, R. C. (1963). A linear
programming model for budgeting and financial
planning. Journal of Accounting Research, 1(2),
198–212. doi:10.2307/2489855
Iqbal, K., & Turnovsky, S. J. (2008). Intergen-
erational allocation of government expenditures:
Externalities and optimal taxation. Journal of Public
Economic Theory, 10(1), 27–53. doi:10.1111/j.1467-
9779.2008.00350.x
Jack, W. (2008). Conditioning aid on social expen-
diture. Economics and Politics, 20(1), 125–140.
doi:10.1111/j.1468-0343.2007.00325.x
Kosenok, G., & Severinov, S. (2008). Individually
rational, budget-balancing mechanisms and alloca-
tion of surplus. Journal of Economic Theory, 140(1),
126–161. doi:10.1016/j.jet.2007.07.002
Lee, S. M., & Clayton, E. R. (1972). A goal pro-
gramming model for academic resource allocation.
Management Science, 18(8), 395–408. doi:10.1287/
mnsc.18.8.B395
Leemis, L. M., & Park, S. K. (2006). Discrete-event
simulation: A first course. Upper Saddle River, NJ:
Pearson Prentice Hall.
Li, M. M. (1977). A logit model of hom-
eownership. Econometrica, 45(5), 1081–1097.
doi:10.2307/1914060
Luenberger, D. G. (1984). Linear and nonlinear pro-
gramming (2nd ed.). Reading, MA: Addison-Wesley.
McGuire, T., Coiner, M., & Spancake, L. (1979).
Budget maximizing agencies and efficiency in
government. Public Choice, 34(3-4), 333–357.
doi:10.1007/BF00225673
 International Journal of Strategic Decision Sciences, 2(3), 1-17, July-September 2011 17

Morey, E. R., Rowe, R. D., & Watson, M. (1993). A
repeated nested-logit model of Atlantic salmon fish-
ing. American Journal of Agricultural Economics,
75(3), 578–592. doi:10.2307/1243565
Rivers, R., & Oxner, T. (1982). A network simu-
lation for budgeting. Omega, 10(2), 135–155.
doi:10.1016/0305-0483(82)90049-4
Spiess, H. A. (1996). Logit parking choice model with
explicit capacities. Aegerten, Switzerland: EMME/2
Support Center.
Tridimas, G. (2001). The economics and politics of
the structure of public expenditure. Public Choice,
106(3-4), 299–316. doi:10.1023/A:1005250716104
United States Geological Survey (USGS). (2011).
Measure the size of an earthquake. Retrieved from
http://earthquake.usgs.gov/learn/topics/measure.php
Zeynalian, M., Jandaghi, G., Memariani, A., &
Jahanshahi, H. (2010). Designing a multi-purpose
optimization model for budget allocation using a hi-
erarchical approach. European Journal of Economics,
Finance, and Administrative Sciences, 25, 126–135.

Saeed A. Bagloee is a Senior Transportation Planner and lead modeler at Parsons Overseas
Limited in Dubai, UAE. He received his BS in Civil Engineering from the Khaje Nasir Toosi
University of Technology and his MS in Transportation Planning and Engineering from the Sharif
University of Technology in Tehran, Iran. His research interest is transportation planning and
modeling including travel demand forecasting, road network design and transit planning, and
economic analysis of transportation projects as well as econometric analysis. He has conducted
extensive research at the Institute for Transportation Studies and Research at the Sharif Uni-
versity of Technology and has presented and published his researches in several international
journals and conferences.

Christopher G. Reddick is an associate professor and Chair of the Department of Public Admin-
istration at the University of Texas at San Antonio, USA. Dr. Reddick’s research and teaching
interests are in public budgeting and information technology in public sector organizations. Some
of his publications can be found in Government Information Quarterly, Electronic Government,
the International Journal of Electronic Government Research and Public Budgeting & Finance.
Dr. Reddick recently edited the two-volume book entitled Handbook of Research on Strategies
for Local E-Government Adoption and Implementation: Comparative Studies. He is also author
of the book Homeland Security Preparedness and Information Systems, which deals with the
impact of information technology on homeland security preparedness.

Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.