You are on page 1of 18

The Engineering Economist

A Journal Devoted to the Problems of Capital Investment

ISSN: 0013-791X (Print) 1547-2701 (Online) Journal homepage: http://www.tandfonline.com/loi/utee20

Optimal Capital Structure and Financial Risk


of Project Finance Investments: A Simulation
Optimization Model With Chance Constraints

Emmanuel A. Donkor & Michael Duffey

To cite this article: Emmanuel A. Donkor & Michael Duffey (2013) Optimal Capital Structure and
Financial Risk of Project Finance Investments: A Simulation Optimization Model With Chance
Constraints, The Engineering Economist, 58:1, 19-34, DOI: 10.1080/0013791X.2012.742948

To link to this article: https://doi.org/10.1080/0013791X.2012.742948

Accepted author version posted online: 06


Nov 2012.
Published online: 06 Nov 2012.

Submit your article to this journal

Article views: 965

Citing articles: 5 View citing articles

Full Terms & Conditions of access and use can be found at


http://www.tandfonline.com/action/journalInformation?journalCode=utee20
The Engineering Economist, 58:19–34, 2013
Copyright © 2013 Institute of Industrial Engineers
ISSN: 0013-791X print/1547-2701 online
DOI: 10.1080/0013791X.2012.742948

Optimal Capital Structure and Financial Risk


of Project Finance Investments: A Simulation
Optimization Model With Chance Constraints

EMMANUEL A. DONKOR AND MICHAEL DUFFEY


Engineering Management and Systems Engineering, School of Engineering and
Applied Sciences, The George Washington University, Washington, DC

We investigate the use of chance constraints in modeling the debt financing decision
under conditions of debt heterogeneity and uncertainty. We develop a stochastic financial
model that uses simulation optimization to select an optimal mix of fixed-rate debt
instruments from different sources, with the objective of maximizing net present value
(NPV) while limiting default risk. We then use simulation to evaluate the performance
of the resulting debt policy. Numerical results from our model indicate that in a world
of uncertainty, project promoters who wish to create bankable proposals for project
financing, by limiting the probability of default, should spread debt across different
maturities.

Introduction
This article deals with the problem of making capital structure decisions under uncertainty,
when a multi-sourced debt financing strategy and project finance is used as a procurement
vehicle for financing capital investments in general and public infrastructure investments
in particular. It develops a multi-period discrete time stochastic financial model that uses
simulation optimization with chance constraints to select a portfolio of debt instruments,
each with a different tenure and interest rate, and subsequently conducts a terminating
simulation of the model to assess the performance of the optimal financing decision in
terms of the level of uncertainty in project returns and its associated default risks. Metrics
to measure these system performance variables are respectively (1) the net present value
(NPV) on one hand and (2) debt service and interest coverage on the other. Via the
simulation optimization procedure, project promoters (sponsors) will be provided with an
optimal policy regarding the amount of debt to be placed with each instrument (source) and,
subsequently, insight on the risk-return characteristics of NPV as well as the probability of
bankruptcy for each period of debt service. Using the model, project sponsors who depend
on spreadsheets for preparing bankable proposals can optimize the financing decision and
obtain information that they can use to demonstrate that the forecast performance of the
project conforms to the compromised objectives of all stakeholders.

Address correspondence to Emmanuel A. Donkor, Engineering Management and Systems


Engineering, School of Engineering and Applied Sciences, The George Washington University,
1776 G Street NW, Suite 101, Washington, DC 20052. E-mail: eadonkor@gwu.edu

19
20 E. A. Donkor and M. Duffey

The context of our model relates to the observation that inadequate budgetary allocation
of funds for infrastructure expansion, renewal, and rehabilitation is of concern to public
asset managers when internally generated funds and/or cost minimization strategies are no
longer feasible options (DeWitt 2000; Jones 1991; Liner and Gibson 1996). Under such
circumstances, project finance can become a funding mechanism of choice (Raftelis 2005).
However, the ability to raise funds to finance a capital investment project using project
finance depends on the extent to which project promoters and their financial advisors can
demonstrate to lenders and equity providers that the project is viable.
Project viability in infrastructure sectors such as energy, water, transport, and telecom
“hinges critically on how risks associated with such investments are evaluated” (Dailami
et al. 1999) during the appraisal process. A critical component of this process is the decision
on capital structure, due to its impact on project returns and default risk. From finance theory,
higher leverage results in higher project value but higher default risk. Therefore, the interests
of providers of funds are in conflict because project viability is measured differently by the
different stakeholders involved. Whereas equity holders desire acceptable returns, lenders’
interest rests on strict measures that minimize financial risk, which results from uncertainty
in the investment environment. Therefore, a key challenge faced by project promoters is
how to raise the right amount of debt that maximizes project value while minimizing default
risk. Additionally, the financing decision can become complicated when multiple sources of
debt, with different financing terms such as cost and maturity, are considered. The challenge
is being able to take advantage of the opportunities that debt diversification offers for risk
management through effective maturity matching. A logical question of interest to project
sponsors faced with this dilemma, as well as researchers, will be as follows:

How should project promoters, with the ability to raise funding from multiple
debt sources with different financing terms, for financing a capital investment
project on a non-recourse basis, choose an optimal debt portfolio that maximizes
project value but minimizes default risk when project cash flows are uncertain?

A recent editorial in the European Journal of Operations Research asserted that the
debate regarding the existence or otherwise of an optimal capital structure, which was
initiated by the seminal work of Modigliani and Miller in 1958, are no longer contentious
(Loistl and van der Wijst 2001). Therefore, from the viewpoint of operations research, a
mathematical program that is able to model the inherent trade-off between risk and return
when making the capital structure decision will be the obvious answer. Unfortunately,
the existing literature on project valuation under uncertainty, including capital structure
optimization, for privately financed capital investments seems to be deficient in this regard.

Literature Review
A large majority of the literature on project valuation under uncertainty with debt financing
assumes a given financing decision with homogeneous debt (see, for example, Bock and
Trück 2011; Hacura et al. 2001; Jones 1991; Wibowo and Kochendörfer 2005). Similarly,
deterministic capital structure models in the project finance domain (e.g., Bakatjan et al.
2003; Zhang 2005), as well as stochastic models (e.g., Dong et al. 2011; Yun et al. 2009),
are limited in the extent to which they consider debt heterogeneity. In the case of capital
structure optimization, the recent paper by Dong et al. (2011) provided a slight exception
to the modeling of homogeneous debt, where the authors considered a mix of debt and
Stochastic Modeling of Capital Structure 21

mezzanine capital in a stochastic dynamic programming approach. This focus on modeling


homogenous debt in the literature would seem to be at variance with actual practice. For
example, in both Dailami et al. (1999) and Finnerty (2007), a case was presented on
the Indiantown cogenerating plant, where the project was financed with a first mortgage
and tax-exempt bonds. More recently, an article in Trade Finance reported that an export
credit agency-backed project financing of Qatar’s Barzan gas field development was being
supported by 31 banks with a $6.6 billion debt package, composed partly of a $3.2 billion
term loan, a $1.35 billion direct loan, and a $1.35 billion commercial loan (Trade Finance
2011). Additional cases on the use of heterogeneous debt in the context of project finance
can be found in Finnerty (2007) and Esty (2003). Given the value of a diversified loan
portfolio for risk management purposes (Agmon et al. 1981) and the evidence on the
simultaneous use of “ . . . different types, sources and priorities of debt” (Rauh and Sufi
2010), the absence of debt heterogeneity in the literature on capital structure optimization
and risk analysis for project finance investments is regrettable. If project stakeholders are to
derive the benefits of loan diversification, a model that considers a mix of debt instruments
from multiple sources, with varied financing terms, will be invaluable. Our model fills this
gap.
Chance-constrained programming has found wide application, including capital
rationing (Gurgur and Luxhoj 2003) and time–cost trade-off decisions in project man-
agement (Ke et al. 2009; Yang 2005). Under this method, the constraints in a decision
problem fraught with uncertainty are allowed to be broken with a predetermined probabil-
ity (Birge and Louveaux 1997; Prékopa 1995; Shapiro et al. 2009). If project promoters and
stakeholders are willing to accept some level of infeasibility, then including probabilistic
constraints will serve their purpose. Given the need for decisions prior to the realization of
uncertain cash flow, and subsequent recourse action to prevent any attendant bankruptcies,
chance-constrained programming seems to be the obvious choice in modeling the optimal
capital structure decision problem when the decision has to be made under uncertainty.
Unfortunately, the existing literature is deficient in this regard.
The application of chance-constrained programming requires adherence to specific
conditions. For example, in its general use, chance-constrained programming is linearized
when the constraints can be analytically expressed in terms of the decision variables
and when the stochastic variables are normally distributed. However, for most stochastic
systems, performance variables can be nonnormal, highly nonlinear, and analytically in-
tractable. Under such circumstances, the traditional stochastic programming approaches
fail, giving room for the use of simulation optimization approaches. Extensive use of this
approach can be found in the discrete event system literature (e.g., Andradottir 2002; Better
et al. 2008; Fu 2002; Glynn 1986; Law 2007) for optimizing stochastic systems. Therefore,
for its application in optimizing stochastic financial models, analysts can benefit from not
assuming any distributional form of the endogenously determined variables that contribute
to the estimation of system performance.

Formulation of the Capital Structure Decision Problem


The nature of the decision problem considered here is to determine the amount of debt Dj
to be raised from each of five debt instruments (from multiple sources) at time t = 0 of
the planning horizon to finance construction in order to maximize project NPV and reduce
the risk of default. Let ij and τ j be the interest rate and tenure for instrument j. Assume
that the cost of investment in time t = 0 is ξ CAPEX and that the project generates after tax
cash flows of ATCFt in each operating period after construction. Here t = 1, . . . , N, N
22 E. A. Donkor and M. Duffey

is the length of the study period and the notation ξ x is used to indicate that variable x is a
stochastic input variable.

Derivation of Simulation-Based Chance Constraints


Chance-constrained programming can be used to allow for some infeasibility in opti-
mization models by constraining the probability of infeasibility in a system. Equation (1)
provides a general formulation of chance constraints:

Prob[g (x, ω) < 0] ≤ p (1)

In (1), p is the maximum probability of infeasibility and the function g(.) is a per-
formance measure determined by a vector of decision variables x and a random vector ω.
For most stochastic systems, g(.) is not well defined and is highly nonlinear. In addition,
variables defined by the random vector ω are not necessarily normally distributed. This
combination of nonlinearity and lack of structure (Law 2007) precludes the use of linearized
equivalents of Equation (1), prompting the use of simulation optimization approaches.
In our specific application here, Equation (1) can be written in the form

Prob (DSCR < α1 ) ≤ p1 (2)

and

Prob (ICR < α2 ) ≤ p2 (3)

where,
EBITDA
DSCR (Debt service coverage ratio) = (4)
Debt service
EBIT
ICR (Interest coverage ratio) = (5)
Interest payment
Here, earnings before interest, taxes, and amortization (EBITDA) and earnings before
interest and taxes (EBIT) are as defined in Equations (29) and (30), respectively, in Table 1.
The parameters α 1 and α 2 are the minimum debt service coverage ratio (DSCR) and
interest coverage ratio (ICR), respectively, and p1 and p2 are maximum levels of probability
of default for debt service and interest coverage, respectively. Although the mathematical
definition of DSCR has several forms, we adopt Equation (4) from the definition in Esty
(1999).
Let DSt and It be total debt service and interest payment at time t, respectively. Then,
Equations (4) and (5) can be expanded into Equations (6) and (7) as follows:

Prob(EBI T DAt − α1 DSt < 0) ≤ p1 , ∀t := 1, . . . , max(τj ) (6)


Prob(EBI Tt − α2 It < 0) ≤ p2 , ∀t := 1, . . . , max(τj ) (7)

Equations (6) and (7) indicate that the probability of default should not exceed pk
(k = 1, 2) for each period t in which debt service and interest payments are required, up to
a period covering the maturity of the debt instrument with the highest tenure in the selected
portfolio. This feature of our model differentiates it from those in previous studies in the
sense that debt maturity structure is endogenously determined, instead of specifying it from
the onset based on the tenure of a homogeneous debt. In this formulation, debt service
Stochastic Modeling of Capital Structure 23

Table 1
Variable definitions of the model

Variable Definition Algebraic formulation Equation


HRS Hours of operation per year = 365 × 24 (15)
PRICEt Electricity price charged at PRICE0 ∗ (1 + f )t (16)
time t
PROD0 Initial production capacity 270 ∗ γ ∗ HRS (17)
Rt Harmonic decline rate ξ R,0 /(1 + ξ R,0 t) (18)
PRODt Amount of electricity PROD0 ∗ (1 − Rt ∗ t) (19)
generated and sold in
year t (MWh)
REVt Revenues PRODt ∗ PRICEt (20)
GFCt General fixed cost charged GFC0 ∗ (1 + f )t (21)
at time t
OFCt Other fixed cost charged at OFC0 ∗ (1 + f )t (22)
time t
OPEXt Operation and maintenance ξ OPEX ∗ PRODt ∗ (1 + f )t (23)
cost

CP End of life θ ξ CAPEX (24)
decommissioning cost
CF Decommissioning cost at CP ∗ (1 + f )T (25)
end of concession period
Ct Decommissioning cost at CF { (1+isis)N −1 } (26)
time t
TEt Total expenses GFCt + OFCt + OPEXt + DCt (27)
DEPt Depreciation charged at ξ CAPEX /10 (28)
time t
EBITDAt Earnings before interest, REVt − TEt (29)
taxes, and amortization
EBITt Earnings before interest EBITDAt − DEPt (30)
and taxes

m
i (1 + i )j
It Interest payment on Dj ∗ { (1j + ij )jj−1 } ∗ (31)
outstanding debt at time t j =1
{1 − (1 + ij )−(−t + 1) }

m
i (1+i )j
PRINt Principal payment on debt Dj ∗ { (1j+ ij )jj −1 } ∗ (32)
j =1
{(1 + ij )−(−t + 1) }

m
i (1 + i )j
DSt Total debt service in year t Dj ∗ { (1j + ij )jj−1 } (33)
j =1
TIt Taxable income at time t EBITt − It (34)
TAXt Tax charged at time t T ∗ TIt (35)
ATCFt After-tax cash flow EBITDAt − DSt − TAXt (36)
24 E. A. Donkor and M. Duffey

and interest payment cannot go beyond the maximum tenure in the set of instruments
considered.
In their current forms, Equations (6) and (7) require some modification for a simulation-
based implementation. We do this by using a default indicator (DI) to model the incidence of
default, which occurs whenever EBITDAt − α 1 DSt < 0 or EBITt − α 2 It < 0. We compute
the expected value of the indicator and constrain the expectation to be upper-bounded by
pk . Using binary variables {1, 0} to model the occurrence of default or otherwise, we recast
the constraints specified in Equations (6) and (7) into those indicated in Equations (9) and
(10), respectively.

Specification of the Capital Structure Decision Model


The complete simulation optimization formulation of the capital structure decision problem,
designed for spreadsheet implementation, is presented in Equations (8) through (14).
⎡ ⎤
 m  N
ATCFt ⎦
Mazimize Expected NPV = E ⎣ Dj − ξCAPEX + (8)
j =1 t =1
(1 + i)t

s.t E[DI D,t ] ≤ p1 ∀t : = 1.. max τj (9)

E[DI I,t ] ≤ p2 ∀t : = 1.. max τj (10)
m

Dj ≤ ξ CAPEX (11)
j =1

Dj ≥ 0 ∀j : = 1..m (12)

where
⎧ 
⎨ 1 if EBITDAt − α1 DSt < 0, ∀0t := 1.. max τj
DID,t = (13)

0 otherwise
⎧ 
⎨1 if EBITt − α2 It < 0, > t := 1.. max τj
DII,t = (14)

0 otherwise

The objective function in Equation (8) is random and the constraints do not make
any assumptions about the distribution of EBITDA and EBIT. Following Prékopa (1995),
the objective function maximizes the expected value of NPV. In this formulation, ξ CAPEX
indicates that capital expenditure is a stochastic input variable, as previously defined.
Equations (13) and (14) use indicator variables to model the occurrence of default in debt
service coverage and interest coverage, respectively. The expected values of these variables
take values between 0 and 1 and are constrained with an upper bound equal to p1 and p2 in
Equations (9) and (10), respectively. The total debt and nonnegativity constraints for each
debt instrument are specified in Equations (11) and (12).

Numerical Example
We develop a simplified stochastic financial model using parameters from a hypothetical
energy investment. Among several simplifications, we adopt the traditional discounted cash
Stochastic Modeling of Capital Structure 25

Table 2
Operating and financial assumptions of the model

Variable Description Units Assumption


i MARR % 9.50
T Marginal combined tax rate % 38
f General inflation rate % 3.00
N Concession period Years 20
CAP Plant capacity MW 270
DEP Depreciation Year 10
is Interest rate for sinking fund % 4.00
γ Capacity utilization % 90
GFC0 Annual general expenses (fixed) $ $902,000
OFC0 Other costs (fixed) $ $818,000
θ Decommissioning cost % 8
PRICE0 Electricity price in Year 0 ($/kWh) $0.08
ξ R,0 Initial harmonic decline rate % Triangle (9, 10.9, 12.5)
ξ CAPEX Investment cost in Year 0 $ PERT (475.250, 510.285, 597.500)
(000,000)
ξ OPEX Annual operating and $/kWh PERT (0.026, 0.029, 0.039)
maintenance

flow analysis as used in Finnerty (2007), recognizing that various valuation models have
been proposed in the literature (Babusiaux and Pierru 2009; Esty 1999; Garvin and Cheah
2004; Petravicius 2009). We also assume independence among the stochastic variables and
periodic cash flows. The choice of an energy sector case, and the various simplifications,
are for illustrative purposes only and do not impose any limitations on the focal problems
addressed in this study. The basic concepts are applicable to other infrastructure sectors and
analysts can impose specific modeling covenants peculiar to their sectors with additional
complexity as desired. The only necessary conditions are some amount of uncertainty and
simulation-based modeling of chance constraints. The operating and financial assumptions
of the model are specified in Table 2, where the notation ξ x is used to indicate that variable
x is a stochastic input variable.
The problem concerns the evaluation of a 270 MW dry-steam geothermal power plant
to be procured using project finance. Financing will be arranged through private placements
from the fixed-rate debt market, where intermediate terms of 5 to 10 years are available.
The level and mix of debt funding is to be determined by stochastic optimization. Five
debt instruments, from different sources, with the following financing terms are under
consideration: (1) a 5-year loan at 6% interest (2) an 8-year loan at 7.5% interest (3) a
6-year loan at 6.5% interest (4) a 10-year loan at 8.0% interest, and (5) a 7-year loan at
7.0% interest. Any one of these, or a combination up to five, could be secured. Straight-
line depreciation for 10 years is considered appropriate, with zero salvage value. Both
depreciation and interest are tax deductible at the marginal combined tax rate. Construction
of the facility is assumed to be completed in Year 1 for a 20-year concession contract. The
problem assumes a constant inflation rate and makes provision for the accumulation of
funds for facility decommissioning through a sinking fund with a rate of 4.0%. Cash flows
for periods beyond the concession (e.g., equipment replacements or capital gains on the
26 E. A. Donkor and M. Duffey

sale of assets) are assumed to be zero. Uncertainty in the model is introduced through three
parameters: initial harmonic decline rate, investment cost (CAPEX) and annual operation
and maintenance cost (OPEX).
Definitions of endogenous variables controlling system output are presented in Table 1,
where the variable m, as used in Equations (31) through (33), represents the actual number
of debt instruments included in the debt portfolio decision. Electricity production in Year 0
is determined by plant capacity (CAP), number of hours of operation in a year (HRS) and
capacity utilization (γ ; see Equation (17)). Subsequently, it declines each year due to the
nature of the geothermal reservoir as specified in Equation (19). The annual decline rate
is initiated in Year 0 through the initial harmonic decline rate. Subsequent periodic rates
decline with time. This is modeled in Equation (18).
Electricity prices in Equation (16) and all fixed and variable costs are affected by
inflation. Fixed costs in Year 0 are split into annual general expenses (Equation (21)) and
other costs (Equation (22)). Annual operation and maintenance expenses are stochastic
variable costs and depend on the production level at time t (Equation (23)). The level
of annual revenue is specified in Equation (20). Each year, a payment must be made
into the sinking fund in order to have a sum large enough at the end of the concession
period for final decommissioning of the facility. Equation (24) computes, at time t = 0,
the end-of-life decommissioning cost, calculated as a percentage of CAPEX. Through the
assumed inflation rate, the future value of this cost is calculated in Equation (25). Using the
sinking fund rate, the annual savings required to accumulate this future value is estimated in
Equation (26). Total expenses, depreciation, and earnings for each period of the concession
contract are specified in Equations (27) through (30).
Equations (31) to (33) determine the financial obligations toward lenders and are
modeled in a debt module described under the next section on procedure. Periodic inter-
est payments (Equation(31)), principal payments (Equation (32)), and total debt service
(Equation (33)) are determined by the number of instruments in the selected debt port-
folio (m), the amount of the loan (Dj ), the interest rate (ij ), and the tenure (τ j ) of instru-
ment j in the portfolio. Excel functions are used to model these endogenous variables.
Finally, Equations (34) through (36) respectively compute the level of taxable income,
taxes due, and net after-tax cash flow (ATCF) for each period t. System output variables
are measured by the NPV, DSCR, and ICR, as defined in Equations (8), (4), and (5), re-
spectively. Consistent with the literature, we set the minimum debt service coverage ratio
(α 1 ) to 1.25 and the acceptable interest coverage ratio (α 2 ) to 1.5 (Bakatjan et al. 2003;
Dailami et al. 1999). Maximum probabilities of default, p1 and p2 , are both assumed to
be 0.05.
From the foregoing, it is evident that with the exception of debt service and interest
payments, all endogenous variables after time t = 0 are stochastic, making system output
stochastic in nature. Obviously, any mathematical expression in terms of the input variables,
including the decision variables, will be highly nonlinear in the latter and structurally
cumbersome.

Procedure

General
We follow three main steps in demonstrating the applicability of our model, using the
hypothetical investment case described above:
Stochastic Modeling of Capital Structure 27

Figure 1. Modules of the model.

Step 1: Develop a stochastic financial model for optimization and risk analysis.
Step 2: Optimize the financing decision of the model and collect optimization results.
Step 3: Simulate the model to assess the performance of the optimized financing decision.

These steps are accomplished using four modules linked as shown in Figure 1. In
Steps 1 and 2, we develop a stochastic financial model containing four worksheets.1 The
assumptions module contains project parameters that drive the cash flow module. The
cash flow module, used for project valuation, computes project earnings and NPV using
current dollars. Outputs from this module, in terms of NPV, debt service, and interest
payments, are fed into the optimization module, which serves as the worksheet for coding
the debt financing decision. In this module, the debt instruments and their financing terms
are specified, in addition to assumptions about the minimum debt service and interest
coverage ratios. The debt module takes the financing decision from the optimization module,
combines it with data from the assumptions module, and computes the periodic debt service
and interest payments. These feed into the cash flow module for the computation of NPV.
Uncertainty analysis, via simulation, subsequently follows the optimization stage in Step 3.
This is crucial for assessing the extent to which the optimization solution actually performs.
We simulate the system for 10,000 replications and evaluate the uncertainty in project NPV
as well as the period-to-period performance of debt service and interest coverage.

Solution Algorithm for Optimization Module


As previously alluded to, models for the optimization of stochastic systems provide less
structure due to the mathematical complexity of the relationship between the input and
output performance variables. Two features are evident from our decision model: (1) the
objective function and its associated constraints cannot be easily expressed in a discernible
mathematical form involving the input variables; and (2) consistent with the simulation
modeling literature, our formulation is highly nonlinear and analytically intractable in the
decision variables, requiring the use of non-gradient-based algorithms and methods such
as simulation optimization. These take advantage of recent developments in metaheuristic

1
The authors can be contacted for the general spreadsheet.
28 E. A. Donkor and M. Duffey

approximation approaches such as genetic algorithms, making them a powerful tool for the
optimization of stochastic systems (Better et al. 2008; Konak et al. 2006).
Though simple flowcharts of simulation optimization models can be found in Fu (2002)
and Law (2007), a detailed description of the algorithm as implemented in RiskOptimizer
(Palisade Corporation, Ithaca, NY) is provided in Palisade (2009). An example of the use
of the genetic algorithm–based solution for solving an optimal capital structure decision
problem of a build–operate–transfer investment project is contained in Yun et al. (2009).
With respect to our model, each simulation (v) is initiated by a new set of values of
the decision variables and is composed of n iterations/replications, obtained by sampling
from the distributions of the stochastic input variables. A simulation is invalid, and hence
discarded, if an iteration results in a violation of the iteration constraint specified in Equation
(11). Simulations for which the simulation constraints in Equations (9) and (10) are not
violated are counted as progress steps. Results for simulations where constraints (9) and
(10) are violated are ignored but the simulation is not terminated.

Modeling Platform
We chose a spreadsheet environment for model coding and implementation for the simple
reason that it is the main platform for financial modeling and risk analysis of capital
investment projects (Dayananda et al. 2002; Law 2007). In addition, current spreadsheet
add-ins make the modeling environment self-contained, allowing both optimization and
uncertainty analysis to be conducted within the same environment. In Steps 2 and 3 of our
procedure, we use version 5.7 of RiskOptimizer and @Risk, respectively, developed by
Palisade Inc. (Ithaca, NY), to optimize and simulate our stochastic financial model.

Numerical Results

Optimization Results
Table 3 shows the progress steps of the simulation optimization algorithm culminating in
the final optimization results. One hundred thirteen simulations were conducted, out of
which 23 were valid. These took a total of 55 minutes and 14 seconds to execute. The best
optimal NPV value was obtained in simulation 51, at an execution time of 13 minutes and
12 seconds. It took between 2,400 and 15,200 iterations for the progress steps to converge.
Our choice of 10,000 replications in Step 3 of our procedure is justifiable, given that the
51st and best simulation optimization took 6,600 iterations to converge. From the results,
the decision is to obtain the lion share of debt almost evenly from three instruments: the
7-year at 7% instrument (29%), the 5-year at 6% instrument (25%), and the 8-year at 7.5%
instrument (24%). Thirteen percent of the total debt is to be raised from the 10-year at 8%
instrument and only 9% of the debt is to be obtained from the 6-year at 6.5% instrument.
These add up to a total debt of $283.884 million and a corresponding debt ratio/capital
structure of 55.63%, computed at the most likely value of CAPEX ($510.285 million). The
optimal NPV value associated with this decision will vary from a minimum of −$53.859
million to a maximum of $137.048 million, with an expected value of $49.492 million and
a standard deviation of $28.036 million.

Simulation Results
Forecast of Earnings and Debt Obligations. Figure 2 presents forecasts of interest payments
(Figure 2a) and debt service (Figure 2b) resulting from adopting the optimal solution
Table 3
Progress steps and results of the simulation optimization algorithm
Statistics of NPV Fixed-rate debt maturities

Simulation Elapsed Standard 5-year 8-year 6-year 10-year 7-year


no. time Iterations Result Mean deviation Min. Max. at 6% at 7.5% at 6.5% at 8.0% at 7.0%
1 0:00:14 4,400 2,079,966 2,079,966 28,140,994 (92,050,159) 86,192,831 — — — — —
2 0:00:21 3,700 11,480,674 11,480,674 28,130,298 (83,829,058) 97,990,973 — — — — 56,642,624
3 0:00:27 3,300 20,894,325 20,894,325 28,237,859 (73,635,292) 109,056,174 — — — — 112,567,899
5 0:01:22 3,400 26,299,460 26,299,460 28,230,168 (67,731,217) 113,481,898 — — 45,435,973 — 101,311,109
7 0:01:27 2,400 32,802,256 32,802,256 28,318,494 (61,629,276) 123,668,661 — — 40,892,375 — 141,828,905
10 0:01:39 3,400 37,649,929 37,649,929 28,137,106 (57,004,326) 124,688,565 — 47,913,487 36,803,138 — 127,646,014
12 0:02:08 3,200 39,590,392 39,590,392 28,385,281 (56,793,475) 127,336,281 37,591,115 43,122,138 33,122,824 — 114,881,413
14 0:02:33 1,800 43,840,155 43,840,155 28,399,191 (49,029,220) 130,947,925 33,832,003 38,809,924 29,810,542 37,923,595 103,393,272
16 0:02:42 4,700 47,648,520 47,648,520 28,275,838 (47,110,139) 135,285,434 81,045,057 41,040,915 26,829,487 34,131,236 93,053,944
19 0:02:53 1,800 48,520,410 48,520,410 28,376,642 (45,949,132) 136,450,334 80,234,607 40,630,505 26,561,193 38,405,767 92,123,405
25 0:04:54 8,200 48,967,057 48,967,057 28,141,288 (58,329,999) 136,653,612 72,211,146 68,087,001 23,905,073 34,565,190 82,911,064
29 0:05:43 15,200 49,068,805 49,068,805 28,126,013 (60,248,391) 140,682,490 72,287,599 68,508,469 23,881,168 34,530,625 82,828,153
38 0:08:12 15,200 49,388,446 49,388,446 28,093,205 (59,517,916) 137,188,773 72,005,226 68,240,858 23,787,882 36,155,977 83,485,209
41 0:08:37 7,800 49,456,958 49,456,958 28,149,339 (58,512,492) 137,217,308 71,285,174 67,558,449 26,595,836 35,794,417 82,650,357
51 0:13:12 6,600 49,492,927 49,492,927 28,036,911 (53,858,646) 137,047,661 71,285,174 67,558,449 26,595,836 35,794,417 82,650,357

29
30 E. A. Donkor and M. Duffey

Figure 2. Operating results of the model: trends in (a) interest payment and (b) debt service due.

described above. Interest payments (It ) are forecast to decrease with time as expected.
It seems to possess a smooth nonlinear profile from Year 1 through Year 10. Similar to
the interest payment profile, debt service (DSt ) is not constant. However, it has a more
dramatic profile. It remains constant from Year 1 to 5 and decreases in Year 6, at which
point the 5-year debt instrument would have been completely amortized. Additional drops
occur in Years 7 and 8, depicting the complete payment of the 7- and 8-year instruments.
It subsequently remains constant through Year 10.

Uncertainty in NPV. System performance with respect to project value is depicted in


Figure 3. The probability distribution of NPV resulting from adopting the optimal solution
described above indicates that, for a simulation of 10,000 replications, the mean value
($49.423 million), range (−$58.447 million, $130.648 million), and standard deviation
($28.161 million) of NPV are similar to those obtained from the optimization procedure,
with an associated 95% confidence interval of ($48.871 million, $49.975 million) of mean
NPV. The distribution of NPV is slightly skewed to the left, indicating that only a few
scenarios are likely to derail the project. The probability of the debt portfolio decision
resulting in a negative NPV is only 4.8%. On the contrary, there is a 5% chance that it will
exceed $93.6 million.

Figure 3. Probability distribution of NPV (color figure available online).


Stochastic Modeling of Capital Structure 31

Figure 4. System performance with respect to debt service coverage: (a) profile of earnings after
debt service (EBITDAt − α 1 DSt ) and (b) number of violations in debt service coverage (color figure
available online).

Default Risk: Debt Service Coverage. System performance with respect to debt service
default risk is presented in Figure 4. Figure 4a shows the profile of excess earnings after
debt service (EBITDAt − α 1 DSt ), whereas Figure 4b is a plot of the number of times out of
10,000 for which debt service default occurs in each period t; that is, EBITDAt − α 1 DSt < 0.
Excess earnings after debt service decreases in a nonlinear fashion up to Year 5, increases
in a stepwise manner from Year 6 to Year 9, and then drops slightly in Year 10. The total
number of debt service violations is 631, occurring in Years 3, 4, and 5. Year 5 turns out to
be the most crucial period for debt service in the life of the project. In terms of performance,
Figure 4 shows that the chances of defaulting on debt service are well contained since the
5th percentile of excess debt service barely touches the zero line and only a few violations
are likely to occur, out of 100,000 scenarios.

Default Risk: Interest Coverage. Similar performance results, in terms of satisfying re-
quirements for default risk, are obtained for interest coverage. Figure 5 shows a plot of
excess earnings after interest payment (EBITt − α 2 It ) and the associated number of interest
coverage violations with time. It has a serpentine structure contained within a horizontal
band, with a trough and a peak in Years 4 and 8, respectively, apart from the initial peak in
Year 1. Its periodic 5th percentile values are all above the 0 value line. The total number

Figure 5. System performance with respect to interest coverage: (a) profile of excess earnings after
interest coverage (EBITt − α 2 It ) and (b) number of violations of interest coverage (color figure
available online).
32 E. A. Donkor and M. Duffey

of violations is 1,839, out of 100,000 scenarios, distributed unevenly across time. Compar-
atively, the risk of interest coverage default is higher than in total debt service payments,
although its performance falls within the limits acceptable to lenders.

Conclusion
In this article, we have developed a simulation-based chance-constrained optimization
model for deciding under uncertainty the optimal composition of heterogeneous debt ob-
tained from multiple sources, and hence capital structure, for project finance investments.
By simultaneously treating project finance, chance constraints, simulation optimization,
and debt heterogeneity, we differentiate our model from those currently existing in the liter-
ature, where chance constraints are not considered and where debt is generally assumed to
be homogeneous. Our model maximizes expected project NPV while constraining periodic
debt service and interest coverage default risks within a given reliability level. Looking
at the distribution of instruments in the debt portfolio decision of our numerical results,
it will be difficult, if not impossible, to arrive at such a decision without the aid of our
chance-constrained simulation optimization model. The uneven spread of debt among the
instruments, and hence sources, demonstrates the advantage of debt heterogeneity and
implies that none of the instruments by themselves is optimal, calling into question the
optimality of the use of homogenous debt in previous studies. Our approach explicitly
models the year-by-year default risk of both debt service and interest coverage, yields
more valuable insights to project sponsors, and leads to decisions that otherwise would not
have been possible with the current approach to making the capital structure decision in
project finance investments. Thus, our model serves the needs of both project promoters
and financiers whose interest lie in adequately balancing the conflicting objectives of risk
and return in capital investment projects, especially those funded with private capital on a
non-recourse basis.

Acknowledgment
Funding for this study was provided by the Logistics Management Institute (LMI) through
the Engineering Management and Systems Engineering Department (EMSE) of the School
of Engineering and Applied Sciences, George Washington University. We wish to acknowl-
edge the intellectual support received from the informal project review team, composed of
selected EMSE faculty and assigned LMI staff.

References
Agmon, T., Ofer, A.R. and Tamir, A. (1981) Variable rate debt instruments and corporate debt policy.
Journal of Finance, 36(1), 113–125.
Andradottir, S. (2002) Simulation optimization: integrating research and practice. INFORMS Journal
on Computing, 14(3), 216–219.
Babusiaux, D. and Pierru, A. (2009) Investment project valuation: a new equity perspective. The
Engineering Economist, 54(2), 101–108.
Bakatjan, S., Arikan, M. and Tiong, R.L.K. (2003) Optimal capital structure model for BOT power
projects in Turkey. Journal of Construction Engineering & Management, 129(1), 89.
Better, M., Glover, F., Kochenberger, G. and Wang, H. (2008) Simulation optimization: applications
in risk management. International Journal of Information Technology & Decision Making, 7(4),
571–587.
Stochastic Modeling of Capital Structure 33

Birge, J.R. and Louveaux, F. (1997) Introduction to stochastic programming. Springer, New York,
NY.
Bock, K. and Trück, S. (2011) Assessing uncertainty and risk in public sector investment projects.
Technology & Investment, 2(2), 105–123.
Dailami, M.V.D., Lipkovich, I. and van Dyck, J. (1999) INFRISK: a computer simulation approach
to risk management in infrastructure project finance transactions (Policy Research Paper No.
2083). The World Bank, Washington, DC.
Dayananda, D., Irons, R., Harrison, S., Herbohn, J. and Rowland, P. (2002) Capital budgeting:
financial appraisal of investment projects, 1st ed. Cambridge University Press, New York,
NY.
DeWitt, J. (2000) The role of financial modeling in preparing water infrastructure plants. Paper read
at the 2000 Infrastructure Conference. Baltimore, Maryland, March 12–15.
Dong, F., Chiara, N., Kokkaew, N. and Wu, J. (2011) Stochastic optimization of capital structure in
privately funded infrastructure projects. Journal of Private Equity, 15(1), 36–47.
Esty, B. (1999) Improved techniques for valuing large-scale projects. Journal of Project Finance, 5,
9–25.
Esty, B. (2003) Modern project finance: a casebook, 1st ed. Wiley, New York, NY.
Finnerty, J.D. (2007) Project financing: asset-based financial engineering. John Wiley, New York.
Fu, M. (2002) Optimization for simulation: theory vs practice. INFORMS Journal on Computing,
14(3), 192–215.
Garvin, M. and Cheah, C. (2004) Valuation techniques for infrastructure investment decisions. Con-
struction Management and Economics, 22(4), 373–383.
Glynn, P.W. (1986) Optimization of stochastic systems. In Proceedings of the 18th Conference on
Winter Simulation. Wilson, J.R., Henriksen, J.O., and Roberts, S.D. (eds). ACM, Washington,
DC.
Gurgur, C.Z. and Luxhoj, J.T. (2003) Application of chance-constrained programming to capital
rationing problems with asymmetrically distributed cash flows and available budget. Engineering
Economist, 48(3), 241–258.
Hacura, A., Jadamus-Hacura, M. and Kocot, A. (2001) Risk analysis in investment appraisal based
on the Monte Carlo simulation technique. European Physical Journal B, 20(4), 551–553.
Jones, C. (1991) Risk analysis of infrastructure debt: the case of water and power investments.
Greenwood Publishing Group, Westport, CT.
Ke, H., Ma, W. and Ni, Y. (2009) Optimization models and a GA-based algorithm for stochastic
time–cost trade-off problem. Applied Mathematics & Computation, 215(1), 308–313.
Konak, A., Coit, D.W. and Smith, A.E. (2006) Multi-objective optimization using genetic algorithms:
a tutorial. Reliability Engineering & System Safety, 91(9), 992–1007.
Law, K. (2007) Simulation modeling and analysis, 4th ed. McGraw-Hill, New York.
Liner, B. and Gibson, R. (1996) Public or private? Making the right decisions. The American City
and County, 111(13), 42–50.
Loistl, O. and van der Wijst, N. (2001) Financial modelling and the theory of finance. European
Journal of Operational Research, 135, 231–232.
Modigliani, F. and Miller, M.H. (1958) The cost of capital, corporation finance, and the theory of
investment. American Economic Review, 48, 261–297.
Palisade. (2009) Guide to using RiskOptimizer: simulation optimization for Microsoft Excel. Palisade
Corporation, New York, NY.
Petravicius, T. (2009) Use of real options in project valuation. Transformations in Business & Eco-
nomics, 8(3), 69–89.
Prékopa, A. (1995) Stochastic programming, 1st ed. Kluwer Academic Publishers, Dordrecht, The
Netherlands.
Raftelis, G.A. (2005) Water and wastewater finance and pricing: a comprehensive guide, 3rd ed.
CRC Press, Boca Raton, FL.
Rauh, J.D. and Sufi, A. (2010) Capital structure and debt structure. Review of Financial Studies,
23(12), 4242–4280.
34 E. A. Donkor and M. Duffey

Shapiro, A., Dentcheva, D. and Ruszczynski, A. (2009) Lectures on stochastic programming: mod-
eling and theory. Society for Industrial and Applied Mathematics, Philadelphia, PA.
Trade Finance. (2011) Banks commit to Barzan gas. Trade Finance, 14(8), 137.
Wibowo, A. and Kochendörfer, B. (2005) Financial risk analysis of project finance in Indonesian toll
roads. Journal of Construction Engineering & Management, 131(9), 963–972.
Yang, I. (2005) Chance-constrained time–cost tradeoff analysis considering funding variability. Jour-
nal of Construction Engineering & Management, 131(9), 1002–1012.
Yun, S., Han, S., Kim, H. and Ock, J. (2009) Capital structure optimization for build–operate–transfer
(BOT) projects using a stochastic and multi-objective approach. Canadian Journal of Civil
Engineering, 36(5), 777–790.
Zhang, X. (2005) Financial viability analysis and capital structure optimization in privatized public
infrastructure projects. Journal of Construction Engineering & Management, 131(6), 656–668.

Biographical Sketches
Emmanuel A. Donkor, M.S., is currently a Fulbright doctoral candidate in the Engineering Man-
agement and Systems Engineering Department of the George Washington University, where he
specializes in the application of OR/MS tools to support quantitative modeling in risk and decision
analysis. Specific research interests include parametric cost modeling, design, simulation, and opti-
mization of spreadsheet models for decision making and capital investment planning. He is a student
member of INFORMS, ASCE, and AWWA.

Michael Duffey, Ph.D., is an associate professor in the Department of Engineering Management


and Systems Engineering at the George Washington University. He joined the faculty at GWU in
1992 after receiving his doctorate in mechanical engineering at the University of Massachusetts
(Amherst). His teaching and research interests relate to the economic design of engineering systems.
He also holds a B.S. in physics and M.S. in manufacturing engineering, both from the University of
Massachusetts, and a B.A. in history from Trinity College. He is a member of ASME.
本文献由“学霸图书馆-文献云下载”收集自网络,仅供学习交流使用。

学霸图书馆(www.xuebalib.com)是一个“整合众多图书馆数据库资源,

提供一站式文献检索和下载服务”的24 小时在线不限IP 图书馆。


图书馆致力于便利、促进学习与科研,提供最强文献下载服务。

图书馆导航:

图书馆首页 文献云下载 图书馆入口 外文数据库大全 疑难文献辅助工具

You might also like