The Interaction of Technology Choice and Financial Risk Management: An

Integrated Risk Management Perspective

Onur Boyabatlı
Technology and Operations Management, INSEAD, 77305 Fontainebleau, France
onur.boyabatli@insead.edu

L. Beril Toktay
College of Management, Georgia Institute of Technology, 30308 Atlanta, GA
beril.toktay@mgt.gatech.edu

Abstract

This paper analyzes the integrated operational and financial risk management portfolio of a
firm that determines whether to use flexible or dedicated technology and whether to undertake
financial risk management or not. The risk management value of flexible technology is due to
its risk pooling benefit under demand uncertainty. The financial risk management motivation
comes from the existence of deadweight costs of external financing due to capital market imper-
fections. Financial risk management has a fixed cost, while technology investment incurs both
fixed and variable costs. The firm’s limited budget, which depends partly on a tradable asset,
can be increased by borrowing from external markets, and its distribution can be altered with
financial risk management. In a parsimonious model, we solve for the optimal risk manage-
ment portfolio, and the related capacity, production, financial risk management and external
borrowing levels, the majority of them in closed form. We characterize the optimal risk man-
agement portfolio as a function of firm size, technology and financial risk management costs,
product market (demand variability and correlation) and capital market (external financing
costs) characteristics. Our analysis contributes to the integrated risk management literature
by characterizing the optimal risk management portfolio in terms of a more general set of op-
erational and financial factors; providing the value and limitation of operational and financial
risk management by explicitly modeling their costs and benefits; demonstrating the interactions
between the two risk management strategies; and relating our theoretical results to empirical
observations.

Key Words: Risk Management, Capacity Investment, Flexibility, Financing, Operational

Hedging.

September 2006

Electronic copy of this paper is available at: http://ssrn.com/abstract=947451

1 Introduction

This paper is about integrating operational and financial risk management and characterizing the
drivers of the optimal integrated risk management portfolio. The two means of risk management are
motivated by the existence of different market imperfection costs and utilize different tools. On the
operational side, firms are exposed to demand and supply uncertainties in product markets. These
uncertainties, which we call forms of product market imperfection, impose supply-demand mismatch
costs. To manage these costs, firms rely on different types of operational flexibility that provide a
better response to product market imperfections and counterbalance the effect of supply-demand
mismatch costs. On the financial side, firms do not always have sufficient internal cash flows to
finance their operations and depend on external capital markets to raise funds. The transaction
costs in capital markets (bankruptcy costs, taxes, underwriter fees, agency costs etc.), which are
forms of capital market imperfection, impose deadweight costs of external financing on firms. To
manage these costs, firms rely on different types of financial instruments written on tradable assets
with which their cash flows are correlated. These financial instruments engineer the internal cash
flows of firms to meet their optimal investment needs and counterbalance the effect of external
financing costs.
Despite responding to two different types of market imperfection, operational and financial
risk management interact with each other: The choice of operational risk management has im-
plications for financial risk management and vice versa. Therefore, operational and financial risk
management should be viewed as constituting an integrated risk management portfolio. In practice,
most corporate-level risk management programs of non-financial firms focus only on financial risk
management (Bodnar et al. 1998). At the same time, a number of large non-financial firms are
becoming more interested in operational solutions to manage their risk exposures (Business Week
1998). Due to the existence of both product and capital market imperfections in practice, using
both risk management tools – and doing so in an integrated fashion – is important.
The academic literature on risk management has largely documented the value and effectiveness
of each risk management tool in isolation. Relatively little progress has been made in understanding
their interactions and the main drivers of an optimal integrated risk management portfolio. The
objective of this paper is to enhance our understanding of integrated risk management. Our main
contributions are to model and analyze an integrated risk management problem that (i) yields
structural results about the characteristics and drivers of an optimal risk management portfolio; (ii)
provides managerial guidelines that can be used in designing risk management programs; and (iii)
can be used to generate hypotheses that account for operational and product market characteristics
to a greater extent than the existing empirical risk management literature.

Electronic copy of this paper is available at: http://ssrn.com/abstract=947451

 To this end, we model a budget-constrained manufacturer who produces and sells two prod-
ucts. Product demands are random, which is the product market imperfection, and correlated. The
firm chooses between flexible and dedicated technologies that incur fixed and variable costs, and
determines the capacity level of the chosen technology. Because of its risk pooling benefit, the
flexible technology is the firm’s operational risk management tool. The firm’s limited budget par-
tially depends on a perfectly tradable asset. The firm can relax its budget constraint by borrowing
from external markets, but borrowing incurs external financing costs that originate from capital
market imperfections. Forwards written on the asset price can be used as the firm’s financial risk
management tool to alter the budget distribution and help counterbalance the effect of external
financing costs. The fixed and variable investment costs of flexible technology are higher than
those of dedicated technology, and financial risk management has a fixed cost. Therefore, it may be
undesirable to use these tools despite their value. In this rich but parsimonious model, we answer
the following research questions:

1. What is the optimal risk management portfolio of the firm (defined as choosing flexible versus
dedicated technology, and engaging in financial risk management or not) as a function of firm
size, technology and financial risk management costs, product market conditions (demand
variability and correlation) and capital market conditions (external financing costs)?

2. What are the fundamental drivers of the optimal risk management portfolio?

3. Are financial and operational risk management complements or substitutes?

4. What are the consequences of the interaction between financial and operational risk manage-
ment? What is the effect of financial risk management on operational decisions?

5. Can our results be used to support or refine existing empirical research?

We derive the optimal integrated risk management portfolio and the related capacity, produc-
tion, financial risk management and external borrowing levels, the majority of them in closed form.
Our analysis reveals that there are three fundamental drivers that explain the optimal portfolio
choice: the robustness of the optimal capacity investment level to product market conditions, the
level of reliance on external financing and the opportunity cost of financial risk management. These
drivers work in opposite directions for large and small firms due to differences in their borrowing
needs under financial risk management. As a result, the size of the firm is highly relevant – the
same underlying conditions lead to different optimal portfolio choices as a function of firm size.
Conversely, it may be optimal for small and large firms to choose the same optimal portfolio for

2
entirely different reasons. These results generate managerial insights and guidelines for designing
an integrated risk management program.
Our analysis clearly illustrates the intertwined nature of operational and financial risk manage-
ment strategies. We show that firms can use financial risk management for speculative purposes
with flexible technology, whereas they may prefer to hedge with dedicated technology. The reason is
that firms with a limited internal budget can optimally increase their asset risk exposure to cover the
higher fixed cost of flexible technology and invest in capacity to generate revenue. We demonstrate
that engaging in financial risk management may induce the firm to change its technology decision;
flexible technology and financial risk management can be complements or substitutes. This is a
direct consequence of the difference between each technology regarding the counterbalancing value
of financial risk management with respect to external financing costs.
We relate our theoretical findings to empirical observations concerning risk management prac-
tices of firms. Our results provide theoretical support for some observations and highlight additional
trade-offs in others. For example, we establish that the value of financial risk management increases
in external financing costs only for large firms and not for small firms. This is in contrast to existing
understanding that this is true for any firm. We show that if firms use financial instruments only
for hedging purposes, it is optimal for small firms to not undertake financial risk management;
existing arguments attribute this observation only to the fixed cost of establishing a financial risk
management program. The distinction we make between large and small firms, and our results
related to the effect of technology and product market characteristics on the risk management
portfolio provide new hypotheses that can be tested empirically.
We note that all of the results1 obtained are analytical and are valid for any demand and asset
price distribution with positive and bounded support. With these results, we contribute to the
growing operations management literature that incorporates financial considerations in operational
decision making. In the next section, we provide more detail about how our work contributes to
the existing literature. In §3, we describe the model and discuss the basis for our assumptions.
§4 analyzes the optimal strategy of the firm, culminating in a characterization of the optimal risk
management portfolio. §5 and §6 flesh out the results of the previous section to describe the impact
of various factors on the optimal portfolio choice. We analyze the value and effect of integrated
decision making by comparing with the non-integrated benchmark in §7. In §8, we discuss the
robustness of our results to our assumptions. §9 concludes.

3
2 Literature Review

In this section, we review the streams of literature related to our paper and delineate our contribu-
tions to each stream. The operations management literature has documented the risk management
value of operational flexibility. Starting with the influential studies of Huchzermeier and Cohen
(1996), Cohen and Huchzermeier (1999) and Kouvelis (1999), this stream delineates the value of
various operational flexibilities (e.g. technology flexibility, geographical diversification, postpone-
ment) in the firm’s network structure, referred as operational hedges, in managing demand-side
product market imperfections (Van Mieghem 2003, 2006, Aytekin and Birge 2004, Kazaz et al.
2005). We refer the reader to Boyabatlı and Toktay (2004) for a recent review of papers in this
stream. A number of papers take this analysis further and study the interaction between different
operational flexibilities of firms (Bish and Wang 2004, Goyal and Netessine 2005, Chod et al. 2006a,
Dong et al. 2006). This stream of papers (often implicitly) assumes perfect capital markets and
hence there are neither deadweight costs of external financing nor any value for financial risk man-
agement. We demonstrate the effect of external financing costs and financial risk management on
the value of operational risk management, and document several interactions between operational
and financial risk management.
The finance literature on risk management, in turn, focuses on financial risk management (e.g.
forwards, options, etc.) and typically does not consider product market imperfections and opera-
tional risk management. The majority of this literature i) provides different explanations for the
existence of financial risk management that are based on different types of capital market imper-
fections; or ii) focuses on the optimal use of financial instruments in a variety of settings. Since the
focus of these papers is financial risk management, the interactions between the two risk manage-
ment strategies are not studied. We refer the reader to Fite and Pfleiderer (1995) for a review of
the first stream and Brown and Toft (2001) for a review of the second.
There are a few theoretical papers that study the firm’s integrated risk management portfolio
choice. In operations, Chod et al. (2006b) and Ding et al. (2005) analyze the interaction between
financial risk management and different types of operational flexibility, where financial risk manage-
ment is motivated by the risk aversion of the decision maker. Chod et al. (2006b) analyze whether
financial risk management complements or substitutes operational flexibility. They demonstrate
that this depends on whether the optimal flexibility level increases or decreases with financial hedg-
ing. We show that financial and operational risk management can again be either complements or
substitutes under external financing, but the driver is firm size. Ding et al. (2005) is closest to our
paper in terms of its research objective. They study the integrated operational (postponement)
and financial risk management (currency options) decisions of a multinational firm and delineate

4
the value of each risk management strategy under demand and exchange rate uncertainty. In a
numerical study, they show that engaging in financial risk management alters the robustness of op-
erational decision variables (capacity) with respect to demand variability and changes the strategic
decision variables (global supply chain structure). We demonstrate similar results analytically. In
addition, we analyze the effect of external financing costs, demand correlation and firm size on
the optimal risk management portfolio. Incorporating the costs of each risk management strategy
enables us to also explore the limits of their use.
In finance, Mello et al. (1995) and Chowdry and Howe (1999) model a multinational firm that
has sourcing flexibility (sourcing from both domestic and foreign production facilities is possible)
and that uses financial instruments to manage the exchange rate risk. These papers demonstrate
the value of sourcing flexibility in conjunction with financial risk management. The focus of these
papers is mainly financial risk management, and they do not consider a detailed representation of the
firm’s operations. Our analysis generates a number of insights about integrated risk management
in a more detailed model of firm operations.
All of these papers assume that financial risk management is costless, in which case financial
risk management is trivially included in the optimal risk management portfolio since it has positive
value. In contrast, the fixed cost of financial risk management (e.g. software and personnel costs)
can be a deterrent in practice. Motivated by this observation, we incorporate a positive fixed
cost for engaging in financial risk management. This makes whether to engage in financial risk
management or not a nontrivial question. The answer to this question goes beyond a boundary
invest/do not invest decision divorced of the other decision variables: Under a budget limit and
external financing costs, the effective cost of financial risk management is larger than its fixed cost
because the firm may need to borrow an additional amount as a result of incurring this fixed cost.
Therefore, engaging in financial risk management has an impact on the level of other decisions
variables. Similarly, the fixed cost of the technology investment has a subtle effect on the optimal
portfolio. These interactions add interesting dimensions to the optimal risk management portfolio.
In contrast to the theoretical finance research, the empirical finance literature has paid more
attention to operational risk management, as reviewed in Smithson and Simkins (2005). This
literature either statistically or qualitatively attributes a number of empirical observations to the
firm’s operational risk management capabilities, which we discuss these observations in detail in §5
and §6. We contribute to this stream in a number of ways: We provide theoretical support for some
empirical observations and delineate additional trade-offs in some others; we provide alternative
explanations to some observations that are based on the interplay between the two risk management
strategies; and we identify potential future empirical research avenues.

5
 In summary, our major contribution is to the integrated risk management literature. We con-
tribute to this literature by i) characterizing the optimal risk management portfolio in terms of
a more general set of operational and financial factors; ii) providing the value and limitation of
each risk management strategy by explicitly modelling the costs and benefits of each strategy; iii)
demonstrating the interactions between the two risk management strategies; and iv) relating our
theoretical predictions to empirical observations.
Note that we have made a distinction between papers that augment the financial risk manage-
ment analysis with operational risk management versus operational decisions only. Up to this point,
we focused on the former, which involves a type of flexibility that can be used for risk management
(and subsumes a number of operational decisions). The latter focuses only on operational decisions
in analyzing financial risk management.
In the latter stream, we highlight Froot et al. (1993) from the finance literature since their
modelling of the financial risk management motive is the same as in our paper. The authors
use a concave increasing investment cost function to capture the operational dimension. They
demonstrate that financial risk management adds value by generating sufficient internal funds to
finance operational investments when there exist deadweight costs of external financing. We ex-
tend their framework by formalizing the operational investments (by incorporating product market
characteristics, and technology and production decisions), and by imposing a cost for financial risk
management. We illustrate that some of their predictions continue to hold, whereas some change
due to the interplay between financial and operational decisions.
In the operations literature, Birge (2000), Chen et al. (2004), Gaur and Seshadri (2005), and
Caldentey and Haugh (2005, 2006) document the value of financial risk management when the
operating cash flows are correlated with a financial index. The financial risk management rationale
is the risk-aversion of the decision maker in the latter three papers. Among these papers, we
can link our paper to Caldentey and Haugh (2005) who motivate financial risk management by
imposing a budget constraint on the firm, but without the possibility of external financing. This
can be viewed as a special case of our model: When the external financing cost is sufficiently high,
the firm never borrows. The external borrowing feature of our model is an important determinant
of the risk management portfolio: the reliance on external borrowing determines the technology
choice and the value of financial risk management with each technology.
Finally, our work is related to two other streams in operations management. The stochastic
capacity investment literature analyzes the question of flexible versus dedicated technology choice
with demand-side (uncertain demand) and supply-side (unreliable supply) product market imper-
fections. We refer readers to Van Mieghem (2003) for an excellent review and to Tomlin and Wang

6
(2005) for a specific focus on the supply-side imperfection. As highlighted in Van Mieghem (2003),
stochastic capacity models (often implicitly) assume perfect capital markets. We demonstrate that
under financing frictions, there exist additional trade-offs in technology choice: the level of reliance
on external financing and the value of financial risk management with each technology.
A second stream relaxes the perfect capital market assumption and models the firm’s joint
financial and operational decisions (Lederer and Singhal 1994, Buzacott and Zhang 2004, Babich
and Sobel 2004, Xu and Birge 2004 and Babich et al. 2006). The primary focus of these papers is
to analyze the effect of external financing costs and the financing decision on operational decisions.
They demonstrate the value of integrated financing and operational decision making. We extend
the interaction argument in these papers by considering another facet of financial decisions, finan-
cial risk management. Our analysis reveals that the effect of external financing costs are largely
dependent on the value of financial risk management and that technology choice is a key determi-
nant of the firm’s reliance on external markets: the higher investment cost of flexible technology
requires higher external financing levels than dedicated technology.

3 Model Description and Assumptions

We consider a monopolist firm selling two products in a single selling season under demand un-
certainty. The firm chooses the technology (dedicated versus flexible), the capacity investment
level and the production level so as to maximize expected shareholder wealth. Differing from the
majority of traditional stochastic technology and capacity investment problems, we model the firm
as being budget constrained, where the budget partially depends on a hedgeable market risk. We
allow the firm to undertake financial risk management to hedge this market risk, and to borrow
from external markets to augment its budget. After operating profits are realized, the firm pays
back its debt; default occurs if it is unable to do so.
We model the firm’s decisions as a three-stage stochastic recourse problem under financial
market and demand risk. In stage 0, the firm chooses its integrated risk management portfolio.
The firm decides its technology choice (flexible or dedicated), whether to engage in financial risk
management, and if so, its financial risk management level under demand and financial market
risk. In stage 1, the financial market risk is resolved and the financial risk management contract
(if any) is exercised; these two factors determine the internal cash level of the firm. The firm
then determines the level of external borrowing and makes its capacity investment using its total
budget (internal cash and borrowed funds). In stage 2, demand uncertainty is resolved and the
firm chooses the production quantities for each product. Subsequently, the firm either pays back
its debt or defaults. In the remainder of this section, we define the firm’s objective and discuss

7
the assumptions concerning each decision epoch in detail. We discuss the robustness of our results
with respect to the majority of these assumptions in §8.

Assumption 1 The firm maximizes the expected (stage 2) shareholder wealth by maximizing the
expected value of equity. The shareholders are assumed to be risk-neutral and the risk-free rate rf
is normalized to 0. Shareholders have limited liability.

The main goal of corporations is to maximize shareholder wealth. The expected shareholder wealth
is a function of the expected cash flows to equity of the firm and the required rate of return of the
shareholders. By assuming the risk neutrality of shareholders, we focus on maximizing the expected
equity value of the firm. The required rate of return is the risk-free rate, which is normalized to
0 by assumption. Although the shareholders are risk-neutral, the existence of external financing
costs creates an aversion to the downside volatility of the internal cash level in stage 1: The firm
may be forced to underinvest in capacity at low internal cash level realizations because of external
financing costs. This creates a motivation for undertaking firm-level financial risk management
activities (Froot et al. 1993).

3.1 Stage 0

In this stage, the firm determines its technology choice T ∈ {D, F }, whether to use financial
risk management, and if so, the financial risk management level HT under financial market and
demand uncertainty. The flexible technology (F ) has a single resource that is capable of producing
two products. The dedicated technology (D) consists of two resources that can each produce a
single product.

Assumption 2 Technology T has fixed (FT ) and variable (cT ) capacity investment costs. The
fixed cost of the flexible technology is higher than that of the dedicated technology; FF ≥ FD . The
variable capacity investment cost of the two dedicated resources are identical. Both technologies are
sold immediately at the end of the selling season at a reduced price of γT FT where γT is the salvage
rate and 0 ≤ γT < 1. The firm commits to technology in this stage whose fixed cost is incurred in
stage 1.

Since flexible technology is generally more sophisticated than dedicated technology, the fixed cost
of flexible technology is assumed to be higher. The stage 0 commitment of the firm to technology
choice can be justified by the lead time of the acquisition (if outsourced) or the development time
(if built in-house) of the technology. When the technology is resold, because of depreciation and
liquidation costs, the fixed cost of the technology cannot be fully retrieved (γT < 1).

8
Assumption 3 The firm uses a loan commitment contract to finance its capacity investment and
to cover the fixed cost of the committed technology. The terms of the contract are known at stage
0, while borrowing takes place at stage 1.

Loan commitment is a promise to lend up to a pre-specified amount at pre-specified terms. In

practice, most short-term industrial and commercial loans in the US are made under loan commit-
ment contracts (Melnik and Plaut 1986). At stage 0, the firm owns the right to a loan contract
that can be exercised in stage 1. We discuss the characteristics of the loan commitment contract
in Assumption 6 of stage 1.

Assumption 4 At stage 0, the firm has rights to a known internal stage 1 endowment (ω0 , ω1 ).
Here, ω0 represents the cash holdings and ω1 represents the asset holdings of the firm. The asset is
a perfectly tradeable asset that has a known stage 0 price of α0 and random stage 1 price of α1 . The
random variable α1 has a continuous distribution with positive support and bounded expectation α1 .

With this assumption, in stage 0, the firm knows that the value of its endowment will be ω0 + α1 ω1
in stage 1, where α1 is random; this is the financial market risk in our model. This representation is
consistent with practice: In general, firms hold both cash and tradable assets on their balance sheet,
such as a multinational firm that has pre-determined contractual fixed payments denominated in
both domestic and foreign currency, or a gold producer that produces a certain level of gold that
is exposed to gold price risk. In these examples, the asset price α1 represents the exchange rate
and the gold price in stage 1, respectively. Although the cash and the asset holdings are certain,
the price of the asset makes the stage 1 value of the internal endowment random. The firm can use
financial risk management tools to alter the distribution of this quantity.

Assumption 5 The firm uses forward contracts written on asset price α1 to financially manage
the market risk. There is a fixed cost FF RM of engaging in financial risk management that is
incurred in stage 0 by transferring the rights of the firm’s claims ω0 and ω1 , in proportions β and
1 − β. Forward contracts are fairly priced. We restrict the number of forward contracts HT such
that the firm does not default on its financial transaction in stage 1.

Forward contracts are the most prevalent type of financial derivatives used by non-financial firms
(Bodnar et al. 1995). The fixed cost of financial risk management (FF RM ) includes the costs of
hiring risk management professionals, and purchasing hardware and software for risk management;
it is independent of the number of forward contracts used. In a recent survey, non-financial firms
report this fixed cost as the second most important reason for not implementing a financial risk
management program (Bodnar et al. 1998). Since we focus on loan commitment contracts and

9
the firm can borrow from external markets only at stage 1, FF RM is deducted in stage 0 from
the firm’s stage 1 endowment by transferring the rights of the claims ω0 and ω1 with β and 1 − β
(1−β)FF RM
proportions respectively. In other words, rights for βFF RM of the cash holdings and α0
of the asset holdings are transferred in stage 0. This leaves the firm with a stage 1 endowment
.
of (ω0F RM , ω1F RM ) = (ω0 − βFF RM , ω1 − 1−β
α0 FF RM ). The firm can only engage³ in financial risk
´
management if these quantities are non-negative, or equivalently, if FF RM ≤ min ωβ0 , α1−β
0 ω1
. Since
the firm is exposed to external financing costs in stage 1, there is an opportunity cost associated
with FF RM : The firm has lower internal cash in stage 1 and may need to borrow more from
external markets after paying for FF RM . The fair-pricing assumption ensures that the firm can
only affect the distribution of its budget in stage 1 – and not its expected value – by financial risk
h F RM i
ω
management. We restrict the feasible set of forwards to the range − 0α1 , ω1F RM . Within this
range of forwards the firm never defaults on its financial transaction in stage 1. This ensures that
we can use default-free prices in forward transactions.

3.2 Stage 1

In stage 1, the market risk α1 is resolved. The value of the firm’s internal endowment and the
exercise of the financial contract (if any) determine the firm’s budget B. In this stage, the firm can
raise external capital if the budget is not sufficient to finance the desired capacity investment. The
firm determines the amount of external borrowing and the capacity investment level under demand
uncertainty.

Assumption 6 With the loan commitment contract, the firm can borrow up to credit limit E from
a unit interest rate of a > rf = 0. The face value of the debt eT (1 + a) is repaid out of the firm’s
assets in stage 2. The firm has physical assets of value P (e.g. real estate) that are pledged to the
creditor as collateral. The loan is secured (fully collateralized), i.e. E(1 + a) ≤ P . The physical
assets are illiquid; they can only be liquidated with a lead time. The value of the physical assets P
is sufficient to finance the budget-unconstrained optimal capacity investment level of the firm. The
salvage value of technology (γT FT ) cannot be seized by the creditor among the firm’s assets. Any
possible costs that may be incurred in the borrowing process by the creditor (e.g. fixed bankruptcy
costs) are charged ex-ante to the firm in a.

We assume that the loan commitment is fully collateralized by the firm’s physical assets P , i.e.
E(1 + a) ≤ P , since most bank loans are secured by the company’s assets (Weidner 1999) and
modelled as such (Mello and Parsons 2000). Although the loan is fully collateralized, if the firm’s
final cash position is not sufficient to cover the face value of the debt, the firm cannot immediately
liquidate the collateral assets to repay its debt since the physical assets are illiquid. Under limited

10
shareholder liability, this leads to default, in which case the creditor can seize these physical assets,
liquidate them and use their liquidation value to recover the loan. The salvage value of technology
is assumed to be non-seizable; the creditor cannot use the salvage value to recover the face value of
the loan. We also assume that the creditor’s transaction costs associated with default (e.g. fixed
bankruptcy costs) are charged to the firm ex-ante in the unit borrowing cost.
A positive unit financing cost (a > 0) and a credit limit less than the value of the collateralized
asset (E < P ) can be interpreted as the deadweight costs of external financing that arise from
capital market imperfections: If the capital markets are perfect (i.e. there are no transaction costs,
default related costs, information asymmetries), then the contract parameters are determined such
that the loan is fairly valued in terms of its underlying default exposure. Since we focus on a
collateralized loan, in the absence of default-related deadweight costs, there is no risk for the
creditor associated with default. Consequently, in perfect capital markets, the fair unit financing
cost of the loan commitment contract would be the risk-free rate (a = 0), and the credit limit would
be the value of collateralized physical asset (E = P ). If there are capital market imperfections,
then a > 0 and E < P would be obtained in a creditor-firm interaction. Therefore, although we
assume that they are exogenous parameters in this paper, a positive unit financing cost (a > 0)
and a credit limit less than the value of the collateralized asset (E < P ) can be interpreted as
capturing the deadweight costs of external financing that arise from capital market imperfections.
This parallels the assumptions in Froot et al. (1993) who take the external financing costs as
exogenous and state that they can be argued to arise from deadweight costs associated with capital
market imperfections.
In a creditor-lender equilibrium, the (endogenous) contract parameters need not be identical
for each technology. In §8, we discuss conditions under which our results with identical contract
parameters are valid in a general equilibrium setting, and refer the reader to Boyabatlı and Tok-
tay (2006) for an analysis of equilibrium contract (a∗T , ET∗ ) for each technology in a creditor-firm
Stackelberg game.
To conclude, we note that our external financing cost structure provides a parsimonious model
that is consistent with real-life practices; allows us to implicitly capture capital market imperfections
and enables us to preserve tractability.

3.3 Stage 2

In this stage, demand uncertainty is resolved. The firm then chooses the production quantities
(equivalently, prices) to satisfy demand optimally. If the firm is able to repay its debt from its final
cash position, it does so and terminates by liquidating its physical assets. Otherwise, default occurs.

11
In this case, because of the limited liability of the shareholders, the firm goes to bankruptcy. The
cash on hand and the ownership of the collateralized physical assets are transferred to the creditor.
The firm receives the remaining cash after the creditor covers the face value of the debt from the
seized assets of the firm.

Assumption 7 Price-dependent demand for each product is represented by the iso-elastic inverse-
1/b
demand function p(qi ; ξ1 ) = ξi qi for i = 1, 2. Here, b ∈ (−∞, −1) is the constant elasticity of
demand, and p and q denote price and quantity, respectively. ξi represents the idiosyncratic risk
component. (ξ1 , ξ2 ) are correlated random variables with continuous distributions that have positive
support and bounded expectation (ξ 1 , ξ 2 ) with covariance matrix Σ, where Σii = σi2 and Σij = ρσ1 σ2
for i 6= j and ρ denotes the correlation coefficient. (ξ1 , ξ2 ) and α1 have independent distributions.
The marginal production costs of each product at stage 2 are 0.

4 Analysis of the Firm’s Optimal Risk Management Portfolio

In this section, we describe the optimal solution for the firm’s technology choice, and the levels of
financial risk management, external borrowing, capacity investment and production. A realization
of the random variable s is denoted by s̃ and its expectation is denoted by s. Bold face letters
represent vectors of the required size. Vectors are column vectors and 0 denotes the transpose
operator. Vector exponents are taken componentwise. xy denotes the componentwise product of
0
vectors x and y with identical dimensions. We use the following vectors throughout the text: ξ =
0
1 , K2 )
(ξ1 , ξ2 ) (product market demand), KF = KF (flexible capacity investment) and KD = (KD D
(dedicated capacity investment). P r denotes probability, E denotes the expectation operator, χ(.)
. . S
denotes the indicator function with χ($) = 1 if $ is true, (x)+ = max(x, 0) and Ω01 = Ω0 Ω1 .
Monotonic relations (increasing, decreasing) are used in the weak sense otherwise stated. Table
1 summarizes the decision variables. Table 6 that summarizes other notation and all proofs are
provided in Appendix A. We solve the problem by using backward induction starting from stage
2.

Stage Name Meaning

Stage 0 T ∈ {D, F } Technology choice, dedicated or flexible
HT Number of forwards with technology T
Stage 1 eT Borrowing level with technology T
KT Capacity investment level with technology T
Stage 2 QT Production quantity with technology T

Table 1: Decision variables by stage

12
4.1 Stage 2: Production Decision

In this stage, the firm observes the demand realization ξ̃ and determines the production quantities
QT 0 = (qT1 , qT2 ) within the existing capacity limits to maximize the stage 2 equity value.

Proposition 1 The optimal production quantity vector in stage 2 with technology T ∈ {D, F } for
given KT and ξ̃ is given by
KF −b
Q∗D = KD , Q∗F = ξ̃ .
˜−b ˜
ξ1 + ξ2−b

Since the unit production cost is zero, the firm optimally utilizes the entire available capacity.
With dedicated technology, the optimal individual production quantities are equal to the available
capacity levels for each product. With flexible technology, the firm allocates the available capacity
KF between each product in such a way that the marginal profits for each product are equal.

4.2 Stage 1: Capacity Choice and External Financing

In this stage, the firm exercises the forward contract HT (if the firm has already decided to engage in
financial risk management at stage 0) and observes the asset price α̃1 . With fair pricing, the strike
price of the forward is equal to α1 . The stage 1 budgets with and without financial risk management
. .
are therefore BF RM (α̃1 , HT ) = ω0F RM + α̃1 (ω1F RM − HT ) + α1 HT and B−F RM (α̃1 ) = ω0 + α̃1 ω1 ,
respectively. We henceforth suppress α̃1 and HT and denote the available budget realization by
B̃ ∈ [0, ∞). For given B̃ and T , the firm determines the optimal capacity investment level K∗T (B̃)
and the optimal external borrowing level e∗T (B̃).

Proposition 2 The optimal capacity investment vector K∗T (B̃) and the optimal external borrowing
level e∗T (B̃) for technology T ∈ {D, F } with a given budget level B̃ are

 .

 K0T if B̃ ∈ Ω0T = {B̃ : B̃ ≥ cT 10 K0T + FT }



 1 . 0 1 0 0
 KT if B̃ ∈ ΩT = {B̃ : cT 1 KT + FT ≤ B̃ < cT 1 KT + FT }

K∗T (B̃) = . cT , cT 10 K1 + FT − ET ≤ B̃ < cT 10 K1 + FT }
K1T if B̃ ∈ Ω2T = {B̃ : B̃ ≥ B T T
(1)



 KT if B̃ ∈ Ω3 = . cT ≤ B̃ < cT 10 K1 + FT − ET }

 T { B̃ : B T


 4 . c
0 if B̃ ∈ ΩT = {B̃ : 0 ≤ B̃ < BT }

³ ´+ ³ ´
bT .
e∗T (B̃) = cT 10 K∗T (B̃) + FT − B̃ χ B̃ > B (2)

cT is the unique budget threshold for technology T ∈ {F, D}

Here, χ(.) is the indicator function and B
such that the firm optimally does not borrow (e∗T (B̃) = 0) and does not invest in capacity (K∗T (B̃) =
0) for B̃ ≤ BcT .

13
The explicit expressions for the capacity vectors in the proposition are given in (28) in the proof. K0T
is the optimal capacity investment in the absence of a budget constraint (the “budget-unconstrained
optimal capacity”). If the budget realization is high enough to cover the corresponding cost FT +
cT 10 K0T (B̃ ∈ Ω0T ), then K∗T (B̃) = K0T with no borrowing. Otherwise, for each budget level
B̃ ∈ Ω1234
T , the firm determines to borrow or not by comparing the marginal revenue from investing
in an additional unit of capacity over its available budget with the marginal cost of that investment
including the external financing cost, (1 + a)cT . For B̃ ∈ Ω1T , the budget is insufficient to cover K0T ,
and the marginal revenue of capacity is lower than its marginal cost. Therefore, the firm optimally
does not borrow, and only purchases the capacity level KT that fully utilizes its budget B̃. For
B̃ ∈ Ω23
T , the marginal revenue of capacity is higher than its marginal cost (1 + a)cT . Therefore,
the firm optimally borrows from external markets to invest in capacity. K1T is the optimal capacity
investment with borrowing, in the absence of a credit limit (the “credit-unconstrained optimal
capacity”). If the budget realization and the credit limit can jointly cover its cost, K1T is the
optimal capacity investment; otherwise, the firm purchases the capacity level KT that fully utilizes
its budget and its credit limit. For B̃ ∈ Ω4T , the firm must borrow to be able to invest in technology,
but the total cost of the capacity that can be purchased with the remaining B̃ + eT − FT cannot
be covered by the expected revenue it generates for any eT . Therefore, the firm optimally does not
borrow and does not invest in capacity. Appendix B characterizes B bT and provides a closed-form
expression for a subset of parameter values.
The optimal external borrowing level e∗T (B̃) is such that the firm borrows exactly what it needs
to cover its capacity investment. Since production is costless, the firm does not incur any further
costs beyond this stage. Moreover, since the face value of the debt is always deducted from the
firm’s assets, the firm cannot transfer wealth from the creditor to shareholders by borrowing more
money than what is needed for its capacity investment. Therefore, the firm only borrows for funding
the capacity investment, which yields (2).
The optimal expected (stage 1) equity value of the firm with a given budget level B̃, πT (B̃), is
obtained in closed form (Equation 34 in Appendix A).

cT , ∞). It is not
Corollary 1 πT (B̃) strictly increases in B̃ for B̃ ≥ 0, and is concave in B̃ on [B
concave in B̃ on [0, ∞).

As we will see in 4.3.1, this structure has implications for the optimal financial risk management
level.

14
4.3 Stage 0: Financial Risk Management Level and Technology Choice

In this stage, the firm decides on the technology choice T ∈ {D, F }, whether to engage in financial
risk management (FRM) and if so, the financial risk management level HT , the number of forward
contracts written on the stage 1 asset price α1 . The optimal expected (stage 0) equity value Π∗ (W)
0
as a function of the internal (stage 1) endowment W = (ω0 , ω1 ) is
© ª
Π∗ (W) = max Λ−F RM , ΛF RM , ω0 + α1 ω1 + P . (3)

Here, ΛF RM and Λ−F RM denote the expected (stage 0) equity value of the better technology
with and without financial risk management (FRM), respectively, where ΛF RM is calculated at the
optimal risk management level HT∗ . In (3), the firm compares these equity values with ω0 +α1 ω1 +P ,
the expected (stage 0) equity value of not investing in any technology. §4.3.1 derives HT∗ , §4.3.2
characterizes the optimal technology choice with and without FRM, and §4.3.3 characterizes the
solution to (3). This characterization is valid for any continuous α1 and ξ distribution with positive
support and bounded expectation.

4.3.1 Financial Risk Management

The expected direct gain from the financial contract is 0 due to the fair pricing assumption. At the
same time, financial risk management affects the distribution of the stage 1 budget BF RM (α1 , HT ),
which is used to finance the firm’s capacity investment after paying for the fixed cost commitment.
In choosing HT , the goal of the firm is to engineer its budget to maximize the expected gain
from the technology commitment made in stage 0. When HT > 0 (HT < 0), the firm decreases
(increases) its exposure to the asset price risk α1 . Following Hull (2000, p.12), we refer to the
first case as financial hedging, and to the second as financial speculation. We call HT = ω1F RM full
ω0F RM
hedging because it isolates the budget from the underlying risk exposure. We call HT = − α1 full
speculation because it maximizes the firm’s asset risk exposure within the feasible range of forward
contracts. Proposition 3 characterizes HT∗ .

Proposition 3 There exists a unique technology fixed cost threshold F T such that

(i) If FT ≤ F T , then the firm fully hedges (HT∗ = ω1F RM ).

(ii) If FT > F T then

F RM
bT , then full speculation is optimal (H ∗ = − ω0
1. if ω0F RM + α1 ω1F RM ≤ B T α1 );
nn b −ω F RM
o © ªo
bT , H ∗ ∈ B
2. if ω0F RM + α1 ω1F RM > B T HT < T α10 ∪ ω1F RM and is distribution
dependent.

15
The structure of πT is key to these results. If πT is a concave function of the available budget
B̃ on [0, ∞), then full hedging is optimal. This follows by Jensen’s inequality: For concave πT ,
E [πT (BF RM (α1 , HT ))] ≤ πT (E[BF RM (α1 , HT ]) = πT (ω0F RM + α1 ω1F RM ), the equity value under
full hedging. However, πT is not concave if Ω4T 6= ∅, i.e. if there is a budget range in which the firm
would not invest in capacity in stage 1 despite having made the technology investment in stage 0.
This happens when the fixed cost of the technology investment is too high to leave sufficient funds
for a profitable capacity investment.
Below the fixed cost threshold F T , Ω4T = ∅, πT is concave, and full hedging is optimal. Above
this threshold, HT∗ depends on the expected value of the internal (stage 1) endowment ω0F RM +
αω1F RM , which is also the budget available to the firm under full hedging. When this value is
lower than BcT , the firm would optimally not invest in capacity if it were to fully hedge. Instead,
the firm optimally chooses to increase its exposure as much as possible so as to maximize the
probability of realizing high-budget states in which it is able to invest in capacity and generate
revenue from its technology investment. (This also increases the probability of realizing low-budget
states, but the outcome in those states does not change - no capacity investment is optimal.) For
ω0F RM + α1 ω1F RM > BbT , the optimal risk management level is distribution dependent and a full
characterization is not possible without making further assumptions.

4.3.2 Technology Choice

We now turn to the technology selection problem with and without financial risk management. The
choice T ∗ between flexible versus dedicated technology is determined by a unit cost threshold that
makes the firms indifferent between the two technologies.

Proposition 4 For given technology cost parameters (FT , γT ) and financing cost scheme (a, E),
and under the financial risk management level HT∗ for each technology, there exists a unique variable
cost threshold cF (cD , H∗ ) such that when cF < cF (cD , H∗ ) it is more profitable to invest in flexible
technology (T ∗ = F ). Without financial risk management, there is a parallel threshold cF (cD , 0).
These thresholds increase in cD , FD , γF and demand variability (σ), and they decrease in FF , γD
and the demand correlation (ρ)2 . With symmetric fixed costs and salvage rates,
 ·³ ´− 1 ¸ − b+1
1

−b −b b
 E−b ξ1 + ξ2 
cF (cD , H∗ ) = cF (cD , 0) = cSF (cD ) = cD 

 ≥ cD , (4)
E [ξ1 ] + E [ξ2 ] 
−b −b

where the equality only holds if the product markets are deterministic (σ = 0), or the product
markets are perfectly positively correlated (ρ = 1) and ξ has a proportional bivariate distribution.

16
 The comparative statics results developed here are used in §6 to analyze the drivers of the firm’s
optimal risk management portfolio. The threshold cSF (cD ) is independent of unit financing cost a,
credit limit E, and engaging in financial risk management. Although these factors do have an
effect on the equity value of each technology, the differential value of this effect is never sufficient to
induce the firm to alter its technology decision. This threshold is independent of α1 and valid for
any distribution of ξ. The threshold cSF (cD ) is a variant of the mix flexibility threshold in Chod et
al. (2006a), and has the same structure. It is interesting to note that the same threshold structure
is valid despite the existence of external financing costs and financial risk management policy in
the symmetric cost case.
Due to the risk pooling benefit of flexible technology, we have cSF (cD ) ≥ cD . Proposition 4
shows that there is no risk pooling benefit (cSF (cD ) = cD ) only if the product market demand is
deterministic, or the multiplicative demand uncertainty is perfectly positively correlated and it has
a proportional bivariate distribution (ρ = 1, σ1 = kσ2 and ξ 1 = kξ 2 for k > 0). Flexible technology
can have risk pooling value even if the product markets are perfectly positively correlated. This
observation is in the spirit of Proposition 6 in Van Mieghem (1998), which is based on the price-
differential of two products in a price-taking newsvendor setting. In our case, the value comes from
the fact that for non-proportional bivariate distributions, the optimal production quantities with
the flexible technology in stage 2 are state dependent such that there is still value from production
switching at different ξ realizations.

4.3.3 Optimal Portfolio Choice

The cost thresholds developed in Proposition 4 reveal which technology is more profitable with and
without financial risk management, but we need several more elements to fully characterize the
solution to (3). Four more cost thresholds achieve this purpose. These thresholds are summarized
in Table 2 and derived in the Appendix A.
The “algorithm” to solve (3) is as follows: We use the variable cost thresholds derived in
Proposition 4 to determine the optimal technologies yielding ΛF RM and Λ−F RM . Using the fixed
technology cost thresholds F −F
T
RM
and F FT RM , if we determine that not investing in any technology
dominates either exactly one or both of ΛF RM and Λ−F RM , (3) is solved. Otherwise, we need to
compare ΛF RM and Λ−F RM . If the same technology is optimal in both cases, then the fixed
financial risk management cost threshold F TF RM is used to determine whether FRM or no FRM
is optimal with that technology and (3) is solved. If different technologies are optimal with and
without FRM, then cT (c−T , HT∗ , 0) is used to determine the optimal solution. This completes the
characterization of the optimal portfolio. The next three sections highlight and discuss a series of

17
 Threshold Usage
cF (cD , 0) Comparison between technologies without engaging in FRM
∗
cF (cD , H ) Comparison between technologies with optimal FRM
cSF (cD ) Comparison between technologies with symmetric FT and γT
F −F
T
RM
Comparison between investing in T without FRM and not investing in any technology
FF T
RM
Comparison between investing in T with FRM and not investing in any technology
F TF RM Comparison between FRM and no FRM with technology T
cT (c−T , HT∗ , 0) Comparison between technology T with FRM and the other technology (−T ) without FRM

Table 2: Thresholds used in solving for the firm’s optimal strategy. The first three were derived in
Proposition 4 and the last four are derived in Propositions 11, 12 and 13 in the Appendix.

insights that can be obtained from this analysis.

5 Observations Concerning the Optimal Risk Management Port-

folio

In this section, we make several observations about the structure of the optimal risk management
portfolio and its managerial implications. We start with an observation that illustrates the limits
of the value of each risk management strategy.
F D
Corollary 2 If capital markets are perfect, F F RM = F F RM = 0: financial risk management
has no value. If product markets are perfect, and absent a fixed cost or salvage value advantage,
cF (cD , H∗ ) = cF (cD , 0) = cD : flexible technology has no value.

Without capital market imperfections, the firm is not exposed to deadweight costs of external
financing, as discussed in Assumption 6. In this case, financial risk management does not have any
value. This is consistent with the decoupling of operational and financial decisions in perfect capital
markets (Modigliani and Miller 1958). If there is no demand uncertainty (Σ = 0), the product
markets are perfect, and the firm is not exposed to supply-demand mismatch costs. Absent a fixed
cost or salvage value advantage, flexible technology does not have any value. Observation 2 confirms
our intuition about the risk management role of each strategy in counterbalancing the effects of
costs that originate from product and capital market imperfections.

Corollary 3 The firm can optimally speculate with forward contracts. Flexible technology can
trigger speculative behavior.

While firms frequently use financial derivatives for hedging purposes, Bodnar et al. (1998)
document that some firms take speculative positions with financial derivatives. Froot et al. (1993)

18
show that speculation may indeed be optimal when there is an external financing cost and the
return on the operational investments and the risk variable are statistically correlated. They also
conclude that in the absence of such correlation, the firm optimally fully hedges. In Proposition
3, we prove that the full-hedging conclusion need not hold if there are fixed costs of technology
investment: Firms with limited expected internal endowment may optimally speculate to be able
to invest in capacity. The majority of empirical papers assume that firms use financial derivatives
for hedging purposes (Geczy et al. 1997). Observation 3 illustrates that such an assumption can
be problematic in industries with fixed cost requirements.
It is interesting to note that speculation can be triggered by investment in flexible technology.
The higher investment cost of flexible technology induces the firm to speculate while it uses forward
contracts for hedging purposes with dedicated technology. This illustrates the intertwined nature
of the integrated risk management portfolio. Engaging in operational risk management (flexible
technology) may have a structural effect (going from hedging to speculation) on financial risk
management.
Firms may limit their usage of financial risk management to hedging only, since speculation
is typically not viewed as a desired strategy. Non-speculative use of financial risk management
imposes a hedging constraint on the feasible set of forwards by imposing HT ≥ 0, which yields the
following outcome:

Proposition 5 If the firm uses forward contracts for hedging purposes only, then the firm optimally
may not engage in financial risk management even if it is costless (FF RM = 0).

The intuition of this result is similar to the full speculation case above, obtained in the case of low
expected internal endowment value. The firm is better off by leaving the exposure to asset price
as high as possible (this corresponds to HT∗ = 0) to be able to invest in capacity. Empirical studies
unanimously demonstrate more widespread usage of financial risk management among large firms,
and this observation is attributed to the fixed costs of establishing a financial risk management
program (Allayannis and Weston 1999). Proposition 5 proposes another possible explanation: the
no-speculation constraint on financial derivative usage. With this constraint, small firms (that have
low internal endowments) do not engage in financial risk management.
In a recent empirical study, Guay and Kothari (2003) find no significant usage of financial risk
management among non-financial firms, and suggest that these firms may be using operational
hedges instead to manage their risks. We observe that indeed, firms can rely only on operational
hedges in an integrated risk management framework.

Corollary 4 Any risk management portfolio can be optimal. Financial risk management is not a
panacea. Firms can rely only on flexible technology for risk management purposes.

19
 If financial risk management was costless, it would always be in the optimal risk management
portfolio. Our analysis finds two reasons why firms may not use financial risk management: i)
Its fixed cost is high. Since non-financial firms do not have as much expertise as financial firms
in financial risk management, its effective fixed cost could be higher for them, which provides
support for the observed difference in usage. ii) The firm limits itself to only hedging even if
it is costless. Thus, not only the investment cost of financial risk management, but also the
interplay between financial and operational decisions is important in determining the optimal risk
management portfolio. The firm should evaluate financial risk management as an integral part of
the firm’s overall investment strategy. The next section provides guidelines about optimal portfolio
selection.

6 Characteristics of the Optimal Risk Management Portfolio

In this section, we delineate the main drivers of the optimal risk management portfolio and analyze
the interplay between financial and operational risk management. In §6.1, we relate the optimal
risk management portfolio to firm, industry, technology, product market (demand variability and
correlation) and capital market (external financing frictions) characteristics. We then analyze the
interaction between operational and financial risk management strategies in §6.2. For this analysis,
we proxy the firm size using the level of internal (stage 1) endowment. In particular:

Definition 1 The firm is defined to be small (large) if the firm borrows (does not borrow) from
external markets with flexible technology and full hedging, ω0F RM + α1 ω1F RM ∈ Ω2F (Ω0F ).

The finance literature qualitatively refers to small and large firms according to the degree to which
they are affected by external financing frictions. This definition formalizes this concept in the
context of our model. We parameterize the internal (stage 1) endowment as (λω0 , λω1 ) and the
fixed technology costs as FD = F , FF = F + δ with δ ≥ 0. For tractability, we impose some
parameter restrictions.
ω0 cT 10 K1
T (1−ab) cT K1
T (1+a)
Assumption 8 Let β = ω0 +α0 ω1 , γT = 0, E ≥ −(b+1)a , FT ≤ F T = −(b+1)a , and FT <
F −F
T
RM
.

These assumptions ensure the following: FF RM ≤ ω0 + α0 ω1 , so that financial risk management is

feasible, and undertaking financial risk management or not can be optimal. If the firm engages in
financial risk management, it optimally fully hedges; this rules out cases where the optimal financial
risk management level cannot be uniquely characterized. The firm is not constrained by the credit
limit, so the effective financing friction is the unit financing cost a. Finally, the optimality of not
investing in either technology is ruled out.

20
6.1 Comparative Statics Results

We define ∆T as the value of financial risk management (FRM) with technology T :

. £ ¡ ¢¤
∆T = E πT BF RM (α1 , ω1F RM ) − E [πT (B−F RM (α1 ))] . (5)

To investigate the main drivers of the optimal portfolio choice, we carry out comparative statics
analysis on the variable cost thresholds cF (cD , H∗ ) and cF (cD , 0), and on ∆T . The results below
hold locally such that Assumption 8 and the defining regions for small and large firms are not
violated.

Proposition 6 (Technology Choice) With symmetric fixed technology costs (FF = FD ), cF (cD , H∗ )
and cF (cD , 0) are invariant to the unit financing cost (a), the fixed costs of both technologies (F ) and
the internal endowment (λ) of the firm. With asymmetric fixed costs (FF > FD ), cF (cD , H∗ ) and
cF (cD , 0) decrease in the fixed costs of both technologies and the unit financing cost, and increase
in the internal (stage 1) endowment of the firm.

With symmetric fixed costs, the technology ordering is independent of financing cost, fixed
costs and internal (stage 1) endowment. With asymmetric fixed costs, since flexible technology
has a higher investment cost, any increase in costs (fixed cost, financing cost) favors the dedicated
technology; a decrease in costs (such as an increase in the internal (stage 1) endowment), favors
the flexible technology.

Proposition 7 (Value of FRM) The value of FRM increases in the external financing cost (a) for
large firms. For small firms, the value of full hedging increases (decreases) in the external financing
cost at low (high) levels of FF RM . For large (small) firms, the value of FRM increases (decreases)
in the fixed cost of technology (F ) and the demand variability (σ), and decreases (increases) in the
internal (stage 1) endowment (λ) and the demand correlation (ρ).

We now explain the drivers of Proposition 7 by grouping the results that have similar intuition.
Since with Assumption 8, the firm optimally fully hedges with financial risk management, we refer
to the firm engaging (not engaging) in financial risk management as the hedged (unhedged) firm.
The effect of external financing cost. Financial risk management is valuable since it reduces
risk exposure and hence the expected borrowing level. At the same time, it is costly, and there
is an opportunity cost for engaging in FRM: the firm may even need to borrow additional funds
to finance its operational investments. These two drivers combine to determine how an increase
in financing cost impacts the financial risk management decision of the firm. For large firms, the
hedged firm – by Definition 1– does not borrow at all, while the unhedged firm is adversely affected

21
from increasing financing costs. Therefore, the value of financial risk management increases in the
financing cost. For small firms, this trade-off depends on the fixed cost of financial risk management.
For low fixed costs, the value of financial risk management increases in financing costs; at high fixed
costs, the opposite occurs.
The effect of fixed technology cost and internal (stage 1) endowment. The proof of the
proposition reveals that there is one fundamental driver that explains both comparative statics
results: the level of reliance on external financing, as summarized in Table 3. A firm’s reliance
on external financing increases as the fixed investment cost F increases and the internal (stage 1)
endowment level λ decreases. By Definition 1, the large hedged firm does not need to borrow and
the large unhedged firm borrows in some budget realizations. Therefore, increasing the reliance
on external financing adversely affects the unhedged firm while not affecting the hedged firm. We
conclude that for large firms, the value of FRM increases as the need for external financing increases.
Since the small hedged firm, by Definition 1, always borrows and the small unhedged firm only
borrows in some budget realizations, increasing the reliance on external financing adversely affects
the unhedged firm, but it affects the hedged firm even more. We conclude that for small firms, the
value of FRM decreases as the need for external financing increases.

Case Borrowing level Increasing reliance on external financing

Large unhedged firm Borrows in some states Increases the value of FRM since
Large hedged firm Does not borrow the unhedged firm borrows more in expectation
Small unhedged firm Borrows in some states Decreases the value of FRM since
Small hedged firm Borrows in all states the hedged firm borrows more in expectation

Table 3: Increasing the reliance on external financing has the opposite effect on the value of financial
risk management for large and small firms. A firm’s reliance on external financing increases as the
fixed investment cost F increases, and it decreases as the internal (stage 1) endowment level λ
increases.

The effect of demand correlation and demand variability. These two factors have an effect
on the firm only with flexible technology. The proof of the proposition reveals that there is one
fundamental driver that explains these two comparative statics results: the marginal change in the
optimal investment level with changes in these factors, as summarized in Table 4. A firm’s optimal
investment level decreases as the demand variability decreases or the demand correlation increases.
The small unhedged firm borrows only in some budget realizations, while the small fully hedged firm
always borrows. As a result, the small hedged firm employs a more conservative investment policy
(the capacity investment level is lower at each state) than the unhedged firm since its exposure to
external financing costs is higher. Consequently, a similar change in variability or correlation alters

22
the small hedged firm’s optimal investment policy to a lower extent than the unhedged firm’s; its
optimal investment level is more robust to changes in these factors. Therefore, while a reduction in
the optimal investment level (due to a decrease in variability or an increase in correlation) adversely
affects the small hedged firm, it affects the small unhedged firm even more. We conclude that for
small firms, the value of FRM increases as the optimal investment level decreases. For large firms,
the opposite result holds. This follows from parallel arguments based on the fact that the large
unhedged firm needs to borrow in some budget realizations, while the large hedged firm does not.

Case Borrowing level Reduction in the optimal investment level at each budget state
Large unhedged firm Borrows in some states Decreases the value of FRM since the hedged firm’s
Large hedged firm Does not borrow optimal investment is less conservative and less robust
Small unhedged firm Borrows in some states Increases the value of FRM since the hedged firm’s
Small hedged firm Borrows in all states optimal investment is more conservative and more robust

Table 4: A reduction in the optimal investment level at each state has the opposite effect on the
value of financial risk management for large and small firms. A firm’s optimal investment level
decreases as the demand variability decreases or the demand correlation increases.

Synthesis. Table 5 summarizes the main drivers of each optimal portfolio choice for large and small
firms by combining Propositions 4, 6 and 7 for technologies with asymmetric fixed cost (FF > FD ).
By definition, if the variable cost thresholds increase in a parameter, flexible technology is preferred
under a larger set of conditions as that parameter increases, and we say that “flexible technology is
favored.” Similarly, if ∆T increases in a parameter, we say “financial risk management is favored.”
While not exact, this usage captures the direction of change. For example, high demand variability
and low demand correlation favor investing in flexible technology and undertaking financial risk
management for large firms. This is how Table 5 is constructed. We note that the capital intensity
of an industry can be captured by keeping the internal endowment level constant and altering the
fixed technology costs. With a given internal endowment level, a sufficiently high (low) fixed cost
implies a small (large) firm according to our definition. Therefore, our results about small and large
firms can be interpreted as being relevant for capital intensive and non-capital intensive industries,
respectively.
The main message of Table 5 is that the size of the firm is key to optimal portfolio choice. As
explained earlier, the three fundamental drivers behind the optimal portfolio choice (opportunity
cost of financial risk management, level of reliance on external financing, and robustness of the
optimal capacity investment level to variability and correlation) work in opposite directions for
small and large firms. Therefore, different size firms may choose the same optimal portfolio for
entirely different reasons.

23
 Portfolio Choice Large Firms Small Firms
High demand variability High internal endowment
F with FRM Low demand correlation Low technology fixed costs
Low financing costs with low FF RM
Low internal endowment Low demand variability
D with FRM High technology fixed costs High demand correlation
High financing costs High financing costs with low FF RM
High internal endowment High demand variability
F without FRM Low technology fixed costs Low demand correlation
Low financing costs Low financing cost with high FF RM
Low demand variability Low internal endowment
D without FRM High demand correlation High technology fixed costs
High financing costs with high FF RM

Table 5: Main Drivers of the Optimal Risk Management Portfolio with Asymmetric Fixed Tech-
nology Costs.

Table 5 is for asymmetric fixed technology costs. With symmetric fixed costs, it follows from
Proposition 4 that the technology ordering is independent of changes in any parameter. Therefore,
changes in parameter levels only affect the choice between undertaking FRM or not. Consequently,
all the conditions in Table 5 that favor flexible or dedicated technology with FRM and without
FRM for a given firm size favor using FRM and not using FRM, respectively. We conclude that
the technology cost characteristic is also key to the optimal portfolio structure.
We now relate our theoretical findings to the associated empirical literature. The financial
risk management literature relates the value of financial risk management to underlying exposure,
growth opportunities and size of firms (Allayannis and Weston 1999). Our results demonstrate
that the value of financial risk management also depends on the product market and technology
characteristics, and that there are subtle differences between large and small firms.
Gay and Nam (1998) say that firms with higher investment opportunities that are exposed to
higher external financing frictions and lower levels of cash make greater use of financial derivatives.
We show (in the proof of Proposition 7) that the effect of cash ω0 is the same as the effect of
internal (stage 1) endowment: A lower internal (stage 1) endowment increases the value of hedging
for small firms, but not for large firms. Therefore, our results support their argument for small
firms, but not for large firms.
The financial risk management literature hypothesizes that the value of financial risk manage-
ment increases as financing frictions increase by invoking the counterbalancing effect of financial

24
risk management with respect to external financing frictions (Mello and Parsons 2000). Our results
support this argument for large firms, but not for small firms. The key is how much the firm needs
to borrow after undertaking financial risk management.

6.2 The Interaction of Operational and Financial Risk Management

We first investigate whether flexible technology and financial risk management are substitutes or
complements in an integrated risk management framework. They are defined to be substitutes if
the firm invests in flexible technology when the firm is not allowed to use financial risk management
and switches to dedicated technology when the firm engages in financial risk management; they are
called complements if the switch is from dedicated to flexible technology.

Proposition 8 Flexible technology and financial risk management can be complements or substi-
tutes. Small (large) firms tend to substitute (complement) flexible technology with financial risk
management.

The main driver of Proposition 8 is the value of financial risk management with each technology.
Flexible technology is more expensive, so it is more exposed to external financing costs. The use
of financial risk management allows large firms to secure a budget level sufficient to eliminate bor-
rowing. Thus, large firms complement flexible technology with financial risk management in their
integrated risk management portfolio. Small firms need to borrow to invest in flexible technology,
even using financial risk management, but may not need to borrow for dedicated technology if they
use financial risk management. In other words, the value of financial risk management is higher
with dedicated technology. This explains why flexible technology and financial risk management
are substitutes for small firms.
Interestingly, the empirical literature also finds mixed results on this question, albeit in other
contexts. Geczy et al. (2000) document complementarity between operational (physical storage)
and financial means of risk management among natural gas pipeline firms. In a multinational con-
text, Allayannis et al. (2001) find that financial and operational (geographical diversification) risk
management tools are substitutes. In a different framework, Chod et al. (2006b) provide another
theoretical justification for these mixed empirical results by focusing on the effect of financial risk
management on the optimal flexibility level of the firm. They demonstrate that financial risk man-
agement is a complement (substitute) to operational flexibility when the optimal flexibility level
increases (decreases) with financial hedging.
We next analyze whether the value of operational risk management (defined as the expected
(stage 0) equity value difference between flexible and dedicated technologies) is more or less robust

25
to changes in product and capital market conditions when financial risk management is undertaken.
Robust strategies are preferable because they perform well under a wider range of parameters, and
can be implemented with more confidence.

Proposition 9 For large (small) firms, the value of operational risk management is less (more)
robust to changes in product market conditions (ρ, σ) and more (less) robust to changes in capital
market conditions (a) with financial risk management than without.

The proof of the proposition reveals that the robustness with respect to product market conditions
is linked to the value of FRM with flexible technology. The value of operational risk management
is more or less robust with respect to correlation if the value of FRM decreases or decreases in
correlation, respectively. This is valid for small and large firms, respectively, as we discussed in
§6.1. Robustness with respect to variability follows from a similar argument. Robustness with
respect to the unit financing cost is determined by the difference between the value of FRM with
flexible and dedicated technologies: The value of operational risk management is more robust to
changes in a if the value of FRM with flexible technology increases more rapidly than the value of
FRM with dedicated technology in response to an increase in a.
Proposition 9 again illustrates the intertwined nature of operational and financial risk manage-
ment strategies: Engaging in financial risk management has the opposite impact on the robustness
of the value operational risk management with respect to product and capital market conditions.

7 Value and Effect of Integrated Decision Making

Sections 5 and 6 analyzed the properties of the optimal integrated risk management portfolio and
its drivers. In practice, firms may not take an integrated approach to these decisions; operational
and financial risk management decisions may be taken independently. In this section, we focus on
the value and effect of integrated decision making. We relax the restrictions of Assumption 8 and
focus on general parameter settings.
If we ignore its effects on operational decisions, financial risk management does not have any
value because forward contracts are investments with zero expected return. For this reason, we
take no FRM as the non-integrated benchmark. Since the non-integrated benchmark is no FRM,
the results of this section can also be interpreted as the effect of engaging in FRM on the firm’s
performance and optimal decisions. The effect of FRM on the optimal expected capacity investment
and external borrowing level is ambiguous:

Proposition 10 Engaging in financial risk management can increase or decrease the optimal ex-
pected capacity investment and the optimal expected borrowing levels.

26
Since financing frictions negatively impact the stage 1 capacity investment level at each budget
state, and the firm uses FRM to counterbalance the effect of financing frictions, one may expect
that with FRM, the firm’s expected borrowing level would be lower and the expected capacity
investment level would be higher than without. On the other hand, if there is cost associated with
engaging in financial risk management (FF RM > 0), the firm has less internal endowment to invest
in capacity at each budget state, and has to borrow additionally to compensate for FF RM . In the
proof of Proposition 10, we illustrate that even if FRM is costless, the optimal expected capacity
investment can decrease and the expected borrowing level can decrease. This is a direct consequence
of the joint optimization in external borrowing and capacity levels. The fundamental driver of this
result is the marginal profit of the capacity investment in the joint optimization problem as we
discussed in §4.2.
Proposition 10 shows the dependence of capacity investment on financial risk management. We
now analyze the effect of engaging in financial risk management on the technology choice:

Corollary 5 The firm may make different technology decisions with and without financial risk
management.

In their numerical analysis, Ding et al. (2005) demonstrate that financial risk management can alter
more strategic operational decisions (global supply chain structure) than the capacity investment
levels. Observation 5 is in line with their conclusion. We analytically prove that the technology
choice of the firm may be altered by engaging in FRM. The direction of change in technology choice
is determined by the value of FRM with each technology. Proposition 8 is an example for such
changes and provides the intuition with some restrictions on the parameter levels.
The analysis above illustrates the effect of integrating risk management decisions on the firm’s
decisions. We now analyze the value of such integration as a function of firm size. To separate
the value of integration from the cost of FRM, we use FF RM = 0. Here, our definition of a large
firm is the same as Definition 1, but our definition of a small firm is slightly more restrictive. We
refer to firms with very limited expected internal endowment value that optimally fully speculate
with FRM as small firms. Since under the conditions of Assumption 8, these firms fully hedge with
FRM, the new definition is consistent with Definition 1 and corresponds to a subset of small firms
in §6 that have a significantly low expected internal endowment value.

Corollary 6 The value of integration is low for small firms with low cash levels (ω0 ) and large
firms with high cash levels. If the firm uses financial risk management only for hedging purposes,
the value of integration is higher for large firms than for small firms.

The value of integration is equivalent to the value of engaging in FRM. Since large firms with high

27
cash levels are not significantly exposed to external financing frictions without FRM, the value
of FRM, and hence the value of integration is low. In the extreme case, a cash level sufficient
to finance the budget-unconstrained optimal investment level completely removes the exposure to
external financing frictions and FRM has no value. For small firms with low levels of cash, the
ω0
additional benefit of full speculation (HT∗ = α1 ) over not using FRM (HT∗ = 0) is low. In the
extreme case, if the small firm does not have any cash (ω0 = 0), then FRM has no value.
When the firm uses financial risk management only for hedging purposes, it follows from Propo-
sition 5 that small firms optimally do not engage in FRM. In this case, integration has no value.
Large firms generally fully hedge with FRM, therefore integration has value for them. In a nu-
merical analysis not reported here, we observe a similar pattern without imposing the hedging
constraint.

8 Robustness of Results to Model Assumptions

In this section, we investigate the robustness of our results to the assumptions presented in §3.
Non-identical and exogenous financing costs. We assumed a unique external financing cost
structure (a, E). The firm can be exposed to a different external financing cost structure (aT , ET )
with each technology T ∈ {D, F }. All the analytical results of §4 continue to hold by replacing
(a, E) with (aT , ET ) where a lower unit borrowing cost is associated with a higher credit limit. The
main insights of the paper do not change except that the technology with lower aT and higher ET
is favored in the optimal risk management portfolio.
Endogenous financing costs. In this paper, we focus on a partial equilibrium setting where
the financing costs are exogenous and identical for each technology. In a general equilibrium
setting, the financing cost for each technology is determined by the interaction between the firm
and a creditor. In Boyabatlı and Toktay (2006), we derive the equilibrium level of secured loan
commitment contracts (a∗T , ET∗ ) for each technology in a creditor-firm Stackelberg game using a
similar firm model. We show that the borrowing terms will be independent of technology choice
when the creditor has limited information about the firm and the technologies, there is no credible
way of information transmission, and the creditor bases its assessment of default probability on
the same cash flow distribution of the firm for any technology. These conditions are relevant for
bank financing where banks rely on the credit history of the firm for credit risk estimation and do
not have operational expertise. All of the results in this paper are valid in the general equilibrium
sense under these conditions. We refer the reader to Boyabatlı and Toktay (2006) for a detailed
treatment of endogenous financing costs.
Unsecured loan commitment contracts. If the firm uses unsecured loan commitment contracts

28
(P = 0), the firm only receives the salvage value of the non-pledgable technology in the default
states. The limited liability of the shareholders left-censors the stage 2 equity value distribution at
0. The expected (stage 1) equity value is calculated using conditional expectations with respect to
default and non-default events. The probability of default depends on the capacity investment level,
external borrowing level and the risk-pooling value of the technology choice. At stage 1, similar to
secured lending, the firm optimally borrows so as to finance the optimal capacity investment level.
In a single-product price-taking newsvendor setting, Babich et al. (2006) provide conditions under
which the expected (stage 1) equity value is unimodal (though not concave) in capacity. With
two products and endogenous pricing, the optimal capacity investment level is very hard to solve
and becomes intractable for flexible technology because of the dependence on default regions with
bivariate product market uncertainty. In our paper, the effect of limited liability is inherent in
the financing cost structure (a, E). When the capital market imperfection costs are default-related
(e.g. bankruptcy costs), if there were no limited liability then the creditor would be sure to recoup
the face value of the loan and default-related costs from the shareholders’ personal wealth. With
such a riskless loan, the cost of the loan would be the risk-free rate (a = 0) and the firm could raise
sufficient funds to finance the budget unconstrained capacity level (E = P ).
If we allow unsecured lending in our setting, we conjecture that the optimal capacity investment
level would be lower: The marginal cost of borrowing is less than 1+a because of the default, which
should induce the firm to borrow more and invest more in capacity. Structural results related to
financial risk management are expected to hold. How the technology choice would change is not
clear because of the dependence on default regions. The arguments in this section are also relevant
for i) partially secured lending (P is positive but not sufficient to finance the budget unconstrained
capacity investment), and ii) secured lending with default-related costs deducted from the firm’s
seized assets by the creditor in the case of default.
Positive production cost at stage 2. Let y denote the unit production cost for both products
with either technology. With y > 0, the optimal production vector at stage 2 is limited by the cash
availability of the firm in addition to the physical capacity constraints. In this case, the literature
often uses a clearing-pricing strategy for tractability that fully utilizes the physical capacity (see
for example, Chod and Rudi 2005). If we assume a clearing-pricing strategy, the firm optimally
borrows so as to fully utilize the physical resource in stage 2 and all the results of our paper continue
to hold by replacing cT with cT + y.
If we focus on the optimal pricing policy with y > 0, the optimal production vector with flexible
(dedicated) technology is state dependent and has a complex form that is characterized by a two-
region (six-region) partitioning of the demand space (ξ1 , ξ2 ) with respect to capacity constraints3 .

29
The optimal capacity level is lower than the y = 0 case, and accounts for the state-dependent
optimal production vector. With flexible technology, the firm optimally borrows the exact amount
required for the full utilization of the physical resource. With dedicated technology, the optimal
borrowing level is such that the physical resources are never fully utilized. Financial capacity has a
risk-pooling benefit with dedicated technology because the firm can allocate the financial resource
to each physical capacity contingent on the demand realization. Because of this additional risk-
pooling benefit of dedicated technology, flexible technology is more adversely affected from y > 0
compared to y = 0. With y > 0, the majority of the insights and the structural results obtained
with y = 0 remain valid. The results concerning the product market characteristics (ρ, σ) are among
the few exceptions. Similar to flexible technology, the value of dedicated technology decreases in ρ
and increases in σ. This is a direct consequence of the declining risk-pooling value of the financial
capacity. The optimal technology choice as a function of product market conditions is not clear in
this setting.
Seizable salvage value of technology. We assume that the creditor cannot seize the salvage
value of the technology in case of default. If the salvage value of the technology is offered as
an additional collateral, then the creditor can seize the technology. With exogenous financing
costs, seizable technology does not have any impact on the results of this paper. With endogenous
financing costs and immediate liquidation of technology, collateralizing the technology reduces the
default risk and hence external financing costs in equilibrium. Different salvage values of the
technologies have a significant impact on the technology choice in equilibrium as we discuss in
Boyabatlı and Toktay (2006).
Fixed cost of technology is incurred at stage 0. If the firm incurs the fixed cost of technology
at the time of commitment (at stage 0), then this fixed cost is deducted from the firm’s internal
stage 1 endowment (ω0 , ω1 ) in the same way as FF RM . With this assumption, the firm always
optimally fully hedges with financial risk management; hence Observation 3 and Proposition 5 do
not hold. All the other results remain valid. The same conclusions hold in the absence of technology
fixed costs (FF = FD = 0).

9 Conclusions

This paper analyzes the integrated operational and financial risk management portfolio of a firm
that determines whether to use flexible or dedicated technology and whether to undertake financial
risk management or not. The risk management value of flexible technology is due to its risk
pooling benefit under demand uncertainty. The financial risk management motivation comes from
the existence of deadweight costs of external financing. Financial risk management has a fixed cost,

30
while technology investment incurs both fixed and variable costs. The firm’s limited budget, which
depends partly on a tradable asset, can be increased by borrowing from external markets, and its
distribution can be altered via financial risk management.
In a parsimonious model, we solve for the optimal risk management portfolio, and the related
capacity, production, financial risk management and external borrowing levels, the majority of
them in closed form. We characterize the optimal risk management portfolio as a function of firm
size, technology and financial risk management costs, product market (demand variability and
correlation) and capital market (external financing costs) characteristics.
We find that three fundamental drivers explain the optimal portfolio choice: the robustness of
the optimal capacity investment with respect to product market characteristics, the level of reliance
on external financing and the opportunity cost of financial risk management. Our results provide
managerial insights about the design of integrated operational and financial risk management pro-
grams. A firm that operates in highly variable or highly negatively correlated product markets
should use flexible technology with financial risk management if the firm has sufficiently high inter-
nal endowment (large firm); and without financial risk management if the firm has limited internal
endowment (small firm). For large firms with low (high) external financing costs, flexible technol-
ogy with financial risk management (dedicated technology without financial risk management) is
the best risk management portfolio. For small firms, the insights related to technology choice under
high and low external financing costs continue to hold but the firm should only use financial risk
management if the fixed cost of financial risk management is sufficiently low.
Our analysis clearly shows the intertwined nature of operational and financial risk management
strategies and illustrates their subtle interactions. For example, operational and financial risk
management can be complements or substitutes depending on the firm size. Flexible technology and
financial risk management tend to be substitutes for small firms and complements for large firms.
The fundamental driver of this result is the difference in the value of financial risk management
with each technology. We also show that the firm’s use of financial instruments for speculative
reasons can be triggered by choosing the higher cost flexible technology.
Our analysis extends the modelling framework of Froot et al. (1993) by formalizing operational
investments and imposing a cost for financial risk management. With our more detailed operational
model, some of their findings do not continue to hold. For example, firms can optimally use financial
risk management for speculative purposes even if the returns from operational investments are
independent from the financially hedgable risk variable. The driver of this result is the fixed cost
of technology. In addition, we show that firms may choose not to use financial risk management
due to its cost when resources are limited. The effective cost of financial risk management is larger

31
than its fixed cost because of the existence of operational investments: After incurring the fixed
cost of financial risk management, the firm may need to borrow additional funds to finance its
operational investments, which imposes an opportunity cost on the firm. These results enhance our
understanding of the effect of operational factors in risk management and underline the importance
of integrated decision making.
This paper brings constructs and assumptions motivated by the finance literature into a classical
operations management problem. In turn, we provide theoretical support for some observations
made in the empirical finance literature and highlight additional trade-offs in some others. For
example, we establish that the value of financial risk management increases in external financing
costs only for large firms and not for small firms. This is in contrast to the existing understanding
that this is true for any firm. There is evidence that large firms use financial instruments more
frequently than small firms. This observation is attributed to the fixed cost of establishing a
financial risk management program. Our analysis proposes another explanation that is based on the
hedging constraint sometimes imposed in practice: If firms are allowed to use financial instruments
for hedging purposes only, it is optimal for small firms to not undertake financial risk management
even if it is costless.
Our paper opens new empirical avenues. The existing literature on risk management typically
does not capture operational aspects such as characteristics of different technologies and product
market characteristics. As demonstrated by our analysis, these can have a significant effect on
the risk management portfolio and generally have opposite effects for large and small firms. The
distinction we make between large and small firms (or equivalently, between capital intensive and
non-capital intensive industries), and our results related to the effect of technology and product
market characteristics on the risk management portfolio provide new hypotheses that can be tested
empirically. For example, we expect to see that large firms engage in financial risk management less
frequently than small firms in highly positively correlated markets. We also expect to see a positive
relation between fixed technology costs and the frequency of engaging in financial risk management
for large firms and a negative relation for small firms.
In §8, we discussed the implication of relaxing some of our assumptions. Other interesting re-
search directions remain. For example, this paper focuses on a monopolistic firm. In an integrated
risk management framework, strategic risk management has not received much attention. Goyal
and Netessine (2005) analyze the value of flexible technology under product market competition. It
would be interesting to incorporate financial risk management decisions of the firm in this compet-
itive setting. The financially hedged firm may invest in more costly flexible technology whereas the
non-hedged competitor may not because of external financing frictions. Financial risk management

32
will certainly have a non-trivial impact on the equilibrium of the game. Dong et al. (2006) take a
step in this direction by modeling operational flexibility and financial risk management decisions
of a global firm facing a local competitor that can only respond by setting its production quantity.
We assume an exogenous external financing cost structure. Technology characteristics can
affect the external financing costs in equilibrium; this occurs if the lender has information about
the firm’s technology options and the ability to assess their operational and collateral value. In this
case and with loan commitment contracts, the financing cost structure would depend on the firm’s
likelihood of borrowing and the default risk conditional on the borrowing level. Flexible technology
has higher costs, and requires more external borrowing than dedicated technology; but the risk-
pooling value of flexible technology decreases the default risk. The different collateral values of
each technology bring another facet to this interaction. It is interesting to analyze which effect
dominates under what conditions. The broader question is whether firms should use flexible versus
dedicated technology in imperfect capital markets. We analyze these issues in a companion paper
(Boyabatlı and Toktay 2006).

Notes
1 With the exception of sensitivity results with respect to demand variability and correlation: These
results require formalization of demand variability and correlation via specific distributional or
structural (using stochastic orderings) assumptions.
2 To capture the effect of demand correlation and variability, we use different measures that are
commonly used in the literature. Throughout the paper, by “an increase in demand variability,”
we refer to any one of the following cases: i) ξ has a symmetric bivariate lognormal distribution
0
and σ monotonically increases, ii) ξ with independent marginal distributions is replaced with ξ
0 0
with independent marginal distributions such that ξ = ξ and ξi is stochastically more variable
0
than ξi for i = 1, 2, or iii) ξ with σ = 0 is replaced with ξ with σ 6= 0. By “an increase in
demand correlation,” we refer to any one of the following cases: i) ξ has a bivariate lognormal
0
distribution and ρ monotonically increases, ii) ξ is replaced with ξ which dominates ξ according
0
to the concordance ordering, or iii) ξ with ρ 6= 1 is replaced with ξ with ρ = 1. The details of the
analysis can be found in the proof.
3 The proofs for the stage 2 optimal production vector for each technology with y > 0 are available
upon request.

33
Acknowledgement

This research was partly supported by the INSEAD Booz-Allen-Hamilton research fund. We thank
Paul Kleindorfer and the participants of the BCTIM Risk Management Mini-Conference in Wash-
ington University, St. Louis, MO, 2005 for their insightful comments.

References

Allayannis, G., J. Ihrig, J., J. Weston. 2001. Exchange-rate hedging: Financial vs operational
strategies. American Economic Review Papers&Proceedings 91 391–395.

Allayannis, G. J. Weston. 1999. The use of foreign currency derivatives and industry structure. G.
Brown, D.H. Chew, eds. Corporate Risk: Strategies and Solutions. Risk Books. 329–338.

Aytekin, U., J. Birge. 2004. Optimal investment and production across markets with stochastic
exchange rates. Working Paper, The University of Chicago Graduate School of Business.

Babich, V., M.J. Sobel. 2004. Pre-IPO operational and financial decisions. Management Science
50 935–948.

Babich, V., G. Aydin, P. Brunet, J. Keppo, R. Saigal. 2006. Risk, financing and the optimal
number of suppliers. Working paper, University of Michigan.

Bish, E., Q. Wang. 2004. Optimal investment strategies for flexible resources, considering pricing
and correlated demands. Operations Research 52 954–964.

Birge, J. R. 2000. Option methods for incorporating risk into linear capacity planning models.
Manufacturing and Service Operations Management 2 19–31.

Bodnar, G. M., G. Hyat, R.C. Marston. 1995. 1995 Wharton survey of derivative usage by US
non-financial firms. Financial Management 25 113–133.

Bodnar, G. M., G. Hyat, R.C. Marston. 1998. 1998 Wharton survey of financial risk management
by US non-financial firms. Financial Management 27 70–91.

Boyabatlı O., L.B. Toktay. 2004. Operational Hedging: A Review with Discussion. Working Paper
2004/12/TM, INSEAD, Fontainebleau, France.

Boyabatlı O., L.B. Toktay. 2006. Capacity investment in imperfect markets: The interaction of
operational and financial decisions. Working Paper, INSEAD.

Brown, G., K.B. Toft. 2001. How firms should hedge. The Review of Financial Studies 15 1283–
1324.

34
Business Week. 1998. Perils of the Hedge Highwire. 26 October 74–77.

Buzacott, J.A., R.Q. Zhang. 2004. Inventory management with asset-based financing. Management
Science 50 1274–1292.

Caldentey, R., M. Haugh. 2005. Supply Contracts with Financial Hedging. Stern School of Business
Working Paper, New York University, NY.

Caldentey, R., M. Haugh. 2006. Optimal control and hedging of operations in the presence of
financial markets. Mathematics of Operations Research 31 285–304.

Chen, X., M. Sim, D. Simchi-Levi, P. Sun. 2004. Risk aversion in inventory management. Working
Paper, MIT.

Chod, J., N. Rudi. 2005. Resource flexibility with responsive pricing. Operations Research 53
532–548.

Chod, J., N. Rudi, J.A. Van Mieghem. 2006a. Mix, time and volume flexibility: Valuation and
corporate diversification. Working Paper, Kellogg School of Management, Northwestern Uni-
versity.

Chod, J., N. Rudi, J.A. Van Mieghem. 2006b. Operational flexibility and financial hedging: Com-
plementarity or substitution effects. Working Paper, Kellogg School of Management, North-
western University.

Chowdhry, B., J.T. Howe. 1999. Corporate risk management for multinational corporations:
Financial and operational hedging policies. European Finance Review 2 229–246.

Cohen, M.A., A. Huchzermeier. 1999. Global supply chain management: A survey of research and
applications. S. Tayur, M. Magazine, R. Ganeshan, eds. Quantitative Models for Supply Chain
Management, Kluwer Academic Publishers.

Ding, Q., L. Dong, P. Kouvelis. 2005. On the integration of production and financial hedging
decisions in global markets. forthcoming in Operations Research.

Dong, L., P. Kouvelis, P. Su. 2006. Operational hedging strategies and competitive exposure to
exchange rates. Working paper, Olin School of Business.

Fite, D., P. Pfleiderer. 1995. Should firms use derivatives to manage risks?. W.H. Beaver, G.
Parker, eds. Risk Management Problems and Solutions, McGraw-Hill International Editions.
139–169.

Froot, K., D. Scharfstein, J. Stein. 1993. Risk management: Coordinating corporate investment
and financing policies. Journal of Finance 48 1629–1658.

35
Gaur, V., S. Seshadri. 2005. Hedging inventory risk through market instruments. Manufacturing
and Service Operations Management 7 103–120.

Gay, G.D., J. Nam. 1998. The underinvestment problem and corporate derivatives use. Financial
Management 27 53–69.

Geczy, C., B.A. Minton, C. Schrand. 1997. Why firms use currency derivatives. Journal of Finance,
52 1323–1354.

Geczy, C., B.A. Minton, C. Schrand. 2000. Choices among alternative risk management strategies:
evidence from the natural gas industry. Working paper, University of Pennsylvania.

Goyal, M., S. Netessine. 2005. Capacity investment and the interplay between volume flexibility
and product flexibility. Working paper, The Wharton School, University of Pennsylvania.

Goyal, M., S. Netessine. 2006. Strategic technology choice and capacity investment under demand
uncertainty. forthcoming in Management Science.

Guay, W., S.P. Kothari. 2003. How much do firms hedge with derivatives?. Journal of Financial
Economics 70 423–461.

Huchzermeier, A., M.A. Cohen. 1996. Valuing operational flexibility under exchange rate risk.
Operations Research 44 100-113.

Hull, J.C. 2000. Options, futures & other derivatives. Fourth edition, Prentice Hall.

Jin Y., P. Jorion. 2006. Firm value and hedging: Evidence from U.S. oil and gas producers. Journal
of Finance 61 893–919.

Kazaz, B., M. Dada, H. Moskowitz. 2005. Global production planning under exchange-rate uncer-
tainty. Management Science 51 1101–1119.

Kouvelis, P. 1999. Global sourcing strategies under exchange rate uncertainty. S. Tayur, M.
Magazine, R. Ganeshan, eds. Quantitative Models for Supply Chain Management, Kluwer
Academic Publishers.

Lederer, P.J., V.R. Singhal. 1994. The effect of financing decisions on the choice of manufacturing
technologies. International Journal of Flexible Manufacturing Systems 6 333–360.

Mello, A. S., J.E. Parsons. 2000. Hedging and liquidity. The Review of Financial Studies 13
127–153.

Mello, A. S., J.E. Parsons, A.J. Triantis. 1995. An integrated model of multinational flexibility
and financial hedging. Journal of International Economics 39 27–51.

Melnik, A., S. Plaut. 1986. Loan commitment contracts, terms of lending and credit allocation.

36
 Journal of Finance 41 425–435.

Modigliani, F., M. Miller. 1958. The cost of capital, corporation finance and the theory of invest-
ment. American Economic Review 48 261–297.

Smithson, C., B.J. Simkins. 2005. Does risk management add value? A survey of evidence. Journal
of Applied Corporate Finance 17 8–17.

Tomlin, B., Y. Wang. 2005. On the value of mix flexibility and dual sourcing in unreliable
newsvendor networks. Manufacturing and Service Operations Management 7 37–57.

Van Mieghem, J.A. 1998. Investment strategies for flexible resources. Management Science 44
1071–1078.

Van Mieghem, J.A. 2003. Capacity management, investment and hedging: Review and recent
developments. Manufacturing and Service Operations Management 5 269–302.

Van Mieghem, J. A. 2006. Risk-Averse Newsvendor Networks: Resource Flexibility, Sharing, and
Hedging. Working Paper, Kellogg School of Management, Northwestern University.

Weidner, D. 1999. Borrowers may save more with secured bank loans. American Banker 164 4.

Xu, X., J. Birge. 2004. Joint production and financing decisions: Modelling and analysis. Working
Paper, Industrial Engineering and Management Sciences, Northwestern University.

37
A Appendix A

Name Meaning
(ω0 , ω1 ) cash and asset holdings of the firm, called the firm’s endowment
β proportion of FF RM deducted from cash holdings of the firm
α0 stage 0 price of tradable asset
(FT , cT ) fixed and variable capacity costs of technology T
γT salvage rate of fixed cost of technology T
FF RM fixed cost of financial risk management (FRM)
B stage 1 budget
(a, E) interest rate and credit limit of the loan contract
rf (= 0) risk-free rate
P value of collateral physical asset
α1 stage 1 price of tradable asset
ξ = (ξ1 , ξ2 ) multiplicative demand intercept in product markets
Σ covariance matrix of ξ
ρ coefficient of correlation in ξ
σ standard deviation of ξ1 and ξ2 )
ΓT optimal stage 2 operating profits
ΠT optimal stage 2 equity value
πT optimal expected (stage 1) equity value
ΛF RM expected (stage 0) equity value of better technology with FRM
−F RM
Λ expected (stage 0) equity value of better technology without FRM
∗
Π optimal expected (stage 0) equity value
∆T Value of financial risk management with technology T

Table 6: Summary of Notation

Proof of Proposition 1: We start by formulating the stage 2 optimization problem. Let

³ ´
ΓT KT , eT , B̃, ξ̃ denote the optimal stage 2 operating profit as a function of the state vector
(KT , eT , B̃, ξ̃). Since we assume production is costless, this profit is equal to the maximum sales
revenue that can be obtained with the existing capacity.
In stage 1, the firm will have observed the budget realization B̃ and borrowed eT to invest
in capacity level KT . The remaining cash holdings of B̃ + eT − cT 10 KT − FT , non-negative by
construction, will have been invested into a cash account with return rf (= 0).
Two outcomes are possible in stage 2: If the firm’s final cash position (operating profits and
cash account holdings) is sufficient to cover the face value of the loan, i.e. ΓT (KT , eT , B̃, ξ̃) + (B̃ +
eT − cT 10 KT − FT ) ≥ eT (1 + a), then the firm does not default; otherwise, it does. If the firm does

38
not default, it repays the face value of its loan and liquidates the non-pledged technology and the
physical assets, generating γT FT and P , respectively. If the firm defaults, the cash on hand and
the ownership of the collateralized physical asset are transferred to the bank. The firm receives the
salvage value of the technology γT FT and the cash R(KT , eT , B̃, ξ̃) remaining after the face value
of the loan is deducted from its seized assets. We write

R(KT , eT , B̃, ξ̃) = P + ΓT (KT , eT , B̃, ξ̃) + (B̃ + eT − cT 10 KT − FT ) − eT (1 + a), (6)

where we invoke the assumptions that any additional fees in the default state (e.g. bankruptcy fee)
are borne by the creditor as out-of-pocket expenditures, and that the loan is fully-collateralized by
the physical asset.
Since the shareholders are risk neutral and the risk-free rate is 0, the stage 2 equity value can be
written as the sum of the individual components cash flows, regardless of when they are realized:

 ΓT (KT , eT , B̃, ξ̃) + (B̃ + eT − cT 10 KT − FT )
³ ´  if no default
ΠT KT , eT , B̃, ξ̃ = −eT (1 + a) + γT FT + P (7)



γT FT + R(KT , eT , B̃, ξ̃) if default

Inspecting (7) reveals that the equity value can simply be written as
³ ´
˜ + (B̃ + eT − cT 10 KT − FT ) + γT FT − eT (1 + a) + P
ΠT KT , eT , B̃, ξ̃ = ΓT (KT , eT , B̃, ξ) (8)

regardless of whether the firm defaults or not. Obtaining this unique functional form is essential in
preserving tractability and in deriving closed-form expressions for the firm’s capacity, technology
and financial risk management decisions for a subset of parameter levels.
The production decision only affects the operating profit Γ in (8), so optimizing the stage 2
equity value is equivalent to the following optimization problem:
0 1
max Q0 p(Q; ξ̃) = max ξ̃ Q1+ b . (9)
Q∈ΘT Q∈ΘT
³ ´
. .
Here, p(Q; ξ̃)0 = p(q1 ; ξ˜1 ), p(q2 ; ξ˜2 ) , ΘF = {Q : Q ≥ 0, 10 Q ≤ KF } and ΘD = {Q : Q ≥ 0, Q ≤
KD } are the feasibility sets for production quantity levels for each technology T .
. 0 1
Let f (Q) = ξ̃ Q1+ b and Q∗T denote the optimal production vector that solves (9) for technology
T ∈ {F, D}. It is easy to establish that f (Q) is strictly concave in Q0 = (q1 , q2 ). Since the
constraints are linear, KKT conditions are necessary and sufficient for optimality and Q∗T is unique.
1/b
Since ∂f = (1 + 1/b)ξ˜1 q
∂qi i > 0, and with b ∈ (∞, −1), limq →0+ ∂f → ∞, the non-negativity
i ∂qi
constraints will be non-binding and the capacity constraint will be binding at optimality. With the
dedicated technology, this yields Q∗D = KD and
³ ´ 0 1
ΓD KD , eD , B̃, ξ̃ = f (Q∗D ) = ξ̃ KD 1+ b .

39
 ¯ ¯
∂f ¯ ∂f ¯
With the flexible technology, according to the KKT conditions, Q∗F solves ∂q1 ¯q 1 = ∂q2 ¯K −q 1 .
F F F
KF −b
After some algebra, we obtain Q∗F = ξ̃ and
ξ̃1−b +ξ̃2−b

³ ´ 1+ 1b h i ³ ´− 1 1+ 1
KF
ξ˜1−b + ξ˜2−b = ξ˜1−b + ξ˜2−b
b
ΓF KF , eF , B̃, ξ̃ = f (Q∗F ) = ³ ´1+ 1 KF b .
ξ˜1−b + ξ˜2−b
b

³ ´− 1
. .
Defining NF = ξ˜1−b + ξ˜2−b
b
and ND = ξ̃ and substituting ΓT in (8) yields the expression for
the optimal equity value ΠT :
³ ´ 1
ΠT KT , eT , B̃, ξ̃ = N0T KT 1+ b + (B̃ + eT − cT 10 KT − FT ) + γT FT − eT (1 + a) + P (10)

Proof of Proposition 2: We start by formulating the stage 1 optimization problem. The optimal
expected (stage 1) equity value of the firm, πT (B̃), is given as follows:
 n o
 max ΨT (B̃), B̃ − (1 − γT )FT + P if B̃ + E > FT
πT (B̃) = (11)
 B̃ − (1 − γ )F + P if B̃ + E ≤ FT
T T

where
h ³ ´i
ΨT (B̃) = max B̃ + eT − (cT 10 KT + FT ) − (B + eT − cT 10 KT − FT ) + E ΠT KT , eT , B̃, ξ̃
KT , eT

s.t. eT ≥ cT 10 KT + FT − B̃

eT ≤ ET (12)

KT ≥ 0, eT ≥ 0.

We start with explaining the formulation of the optimization problem (12). The firm has available
budget B̃ and borrows eT from the creditor. Out of this sum B̃ + eT , the firm invests cT 10 KT + FT
in capacity and places the remainder (B + eT − cT 10 KT − FT ) into the cash account. The return
from the cash account and the operating profits from the capacity investment are included in the
h ³ ´i
expected value of the equity in stage 2, E ΠT KT , eT , B̃, ξ̃ . Using (8), the objective function
can be rewritten as B̃ + P − (1 − γT )FT + ΓT (KT , eT , B̃, ξ̃) − cT 10 KT − aeT . Here, the first three
terms are equal to the equity value of the firm if the firm does nothing (does not borrow and does
not invest). Note that since the firm has already committed to technology T , the fixed cost FT
is incurred even if KT = 0. The last three terms are the net profit derived from borrowing and
investing in capacity.
The first constraint ensures that the amount of external borrowing is greater than the difference
between the cost of the investment and the available budget, otherwise the investment is not feasible.

40
The second constraint states that the external borrowing is less than the credit limit (E) of the
firm.
Equation (11) states the firm will either choose a positive capacity level in stage 1 or do nothing
(not borrow and not invest in capacity). The former will be the case when the optimal capacity
investment level obtained in (12) is positive, and this solution dominates doing nothing; with
πT (B̃) = ΨT (B̃). In the latter case, the equity value of the firm is B̃ + P − (1 − γT )FT , with
K∗T (B̃) = 0 and e∗T (B̃) = 0. This is the optimal solution if (i) the budget plus the credit limit
is insufficient (or only sufficient) to cover the fixed cost of investment (B̃ + E ≤ FT ), so the
firm liquidates the physical asset and salvages the technology; or if (ii) the budget plus credit
limit is sufficient to cover the fixed cost, but the firm optimally chooses not to invest in capacity
(B̃ + P − (1 − γT )FT > ΨT (B̃) when B̃ + E ≥ FT ). Note that if KT = 0 in the optimal solution
of (12), the formulation in (12) forces the firm to (suboptimally) borrow E − B̃, but the optimal
objective function value is then dominated by B̃ + P − (1 − γT )FT , the value of doing nothing, so
the joint formulation in (5) and (6) yields the correct optimal solution.
Since a > 0, the firm optimally does not borrow if it does not invest in capacity (eT = 0 if
KT = 0) and only borrows exactly enough to cover the capacity investment when this investment
³ ´+
level is positive (eT = cT 10 KT + FT − B̃ if KT > 0). Substituting ΠT from (8) and ΓT from
Proposition 1 in (12), we obtain the equivalent formulation
³ ´+ 1
ΨT (B̃) = max B̃ − cT 10 KT − (1 − γT )FT − a cT 10 KT + FT − B̃ + E [NT ] 0 KT 1+ b + P
KT

s.t. cT 10 KT + FT − B̃ ≤ E (13)

KT ≥ 0.

Let g(KT ) denote the objective function in (13) and KpT (B̃) be the optimal solution of (13). The
³ ´+
corresponding optimal borrowing epT (B̃) is equal to cT 10 KpT (B̃) + FT − B̃ . For B̃ > FT , the
B̃−FT
function g(KT ) has a kink and is not differentiable at 10 KT = cT . We rewrite (13) as a
combination of two sub-problems i = 0, 1 with

 max Ψi (B̃) if B̃ > FT
i T
ΨT (B̃) = (14)
 Ψ1 (B̃) if B̃ ≤ FT
T

such that
³ ´ 1
ΨiT (B̃) = max B̃ − cT 10 KT − (1 − γT )FT − ai cT 10 KT + FT − B̃ + E [NT ] 0 KT 1+ b + P
KT

s.t. ZLi ≤ cT 10 KT + FT − B̃ ≤ ZUi (15)

KT ≥ 0,

41
where a0 = 0, a1 = a and ZL0 = −∞, ZL1 = 0, ZU0 = 0, ZU1 = E. Subproblem 0 (1) is the restriction
of the problem to the no borrowing (borrowing) regions. Let g i (KT ) denote the objective function
and KpTi (B̃) be the optimal solution of sub-problem i. We have

 g 0 (K ) if c 10 K + F ≤ B̃
T T T T
g(KT ) =
 g (KT ) if cT 1 KT + FT > B̃.
1 0

The remainder of the proof has the following structure:

1. We show that g i (KT ) is strictly concave and solve each sub-problem i for KpTi (B̃).

2. We show that g(KT ) is strictly concave. It follows that


 arg max Ψi (B̃) if B̃ > FT
i T
KpT (B̃) = KpTi (B̃) where i =
 1 if B̃ ≤ FT

We derive ΨT (B̃) by using KpTi (B̃).

3. We compare ΨT (B̃) with B̃ − (1 − γT )FT + P , the value of not investing in capacity, and
derive K∗T (B̃) and e∗T (B̃).

1. Solution for KpTi (B̃)

1.a. Flexible Technology:
·³ ´− 1 ¸
. −b −b b
Let A = E [NF ] = E ξ1 + ξ2 . The first and second order conditions in (15) are

∂g i 1/b
= −cF − ai cF + (1 + 1/b) A KF ,
∂KF
∂ 2gi 1 (1/b−1)
= (1 + 1/b) A KF .
∂KF2 b

(1/b−1) (1/b−1)
∂ 2 KF ∂ 2 KF
Since b < −1, we have limKF →0+ ∂KF2 → ∞ and ∂KF2 > 0 ∀KF ≥ 0. With b < −1, it
∂ 2 gi
follows that 2
∂KF
< 0 for KF ≥ 0 and the function g i (KF ) is strictly concave for i = 0, 1. Since the
constraints in (15) are linear, first-order KKT conditions are necessary and sufficient for optimality
for each sub-problem i and KFpi (B̃) is unique.
From KKT conditions if i has a non-empty feasible region then the optimal solution is either the
µ ¶−b
∂g i pi A(1+ 1b )
solution of ∂KF = 0, KF (B̃) = c (1+ai ) , or is a boundary solution. Since B̃ > FF for i = 0
F

from (14) and B̃ > FF − E for i = 1 from (11), the non-negativity constraint is never binding
i +B̃−F
ZL
∂g i F ∂g i
in (15). Since limKF →0+ ∂KF → ∞, KF = 0 is never optimal. If cF > 0 and ∂KF < 0
i +B̃−F
ZL
at this point, then KFpi (B̃) = cF
F
, i.e., the optimal solution occurs at the lower bound of

42
 i +B̃−F
ZU i +B̃−F
ZU
∂g i
the financing constraint. If ∂KF > 0 at KF = cF
F
> 0, then KFpi (B̃) = cF
F
, i.e., the
optimal solution occurs at the upper bound of the financing constraint. To summarize, KFpi (B̃) for
i = 0, 1 is characterized by
 µ ¶−b

 K0 . A(1+ 1b )
p0 F = cF if cF K0F + FF − B̃ ≤ 0
KF (B̃) = ³ ´ (16)

 K . B̃−FF
F = cF if cF K0F + FF − B̃ > 0,
 ³ ´
 . B̃−FF

 KF = cF if cF K1F + FF − B̃ ≤ 0

 µ ¶−b
KpF1 (B̃) = . A(1+ 1b )
K1F = if 0 < cF K1F + FF − B̃ ≤ E

 cF (1+a)

 ³ ´
 K . E+B̃−FF
F = cF if cF K1F + FF − B̃ > E.

Here, K0F is the budget-unconstrained optimal capacity investment and K1F is the credit-unconstrained
optimal capacity investment.
1.b. Dedicated Technology:
We obtain
∂ 2gi 1 j (1/b−1)
j 2
= (1 + 1/b) ξ j (KD ) < 0,
∂(KD ) b
· ¸ 2 Y1
∂ 2gi ∂ 2gi ∂ 2gi j (1/b−1)
1 )2 ∂(K 2 )2 − 1 K2 = (1 + 1/b) ξ j (KD ) −0>0
∂(KD D ∂KD D b
j

for i = 0, 1 and j = 1, 2. Therefore, the Hessian matrix D2 g i (KD ) is negative definite for KD ≥ 0
and g i (KD ) is strictly concave. Since the constraints in (15) are linear, first-order KKT conditions
are necessary and sufficient for optimality in each sub-problem i and KpDi (B̃) is unique.
0
If KpDi (B̃) is an optimal solution to (15), then there exist λi = (λi1 , λi2 ) and µi = (µi1 , µi2 ) that
0

satisfy

cD 10 KpDi (B̃) + FD − B̃ ≤ ZUi , (17)

cD 10 KpDi (B̃) + FD − B̃ ≥ ZLi , (18)

KpDi (B̃) ≥ 0, (19)

−(1 + ai )cD + (1 + 1/b) ξ KpDi (B̃)1/b − cD (λi1 − λi2 ) + µi = 0, (20)

λi1 [ZUi − cD 10 KpDi (B̃) − FD + B̃] = 0, (21)

λi2 [−ZLi + cD 10 KpDi (B̃) + FD − B̃] = 0, (22)

µi KpDi (B̃) = 0 (23)

∂g i
with λi ≥ 0 and µi ≥ 0 for i = 0, 1. Observe that limK j →0+ j → ∞ for j = 1, 2, so it is never
D ∂KD
optimal to invest in only one of the resources. Since we will compare ΨD (B̃) with B̃−(1−γD )FD +P

43
(the value of not investing in either resource) in Step 3, we can focus on KpDi (B̃) > 0 here. This
implies µi = 0 for (23) to be satisfied.
Case 1: cD 10 KpDi (B̃) + FD − B̃ < ZUi and cD 10 KpDi (B̃) + FD − B̃ > ZLi
In this case λi = 0, and (20) yields
à ¡ ¢ !−b
. 1 + 1b −b
KpDi (B̃) = KiD = i
ξ .
cD (1 + a )

For (17), (18) and (19) to be satisfied, and the solution KpDi (B̃) = KiD to be valid, we need
ZLi < cD 10 KiD + FD − B̃ < ZUi . Here, K0D is the budget-unconstrained optimal capacity investment
and K1D is the credit-unconstrained optimal capacity investment.
Case 2: cD 10 KpDi (B̃) + FD − B̃ = ZUi
In this case (18) holds as a strict inequality, so λi2 = 0 for (22) to be satisfied. Rewriting the
i +B̃−F −c K 1
ZU
2 =
equality as KD F D D
, and combining this with (20) yields
cD
õ ¶Ã −b
! µ ¶Ã −b
!!
ZUi + B − FD ξ1 ZUi + B − FD ξ2
KpDi (B̃)0 = −b −b
, −b −b
. (24)
cD ξ +ξ cD ξ +ξ
1 2 1 2

The condition λi1 ≥ 0 should be satisfied at optimality. After some algebra, this condition implies
that (24) is optimal if B̃ ≤ cD 10 KiD + FD − ZUi .
Case 3: cD 10 KpDi (B̃) + FD − B̃ = ZLi
This case is only relevant for i = 1 since ZL0 = −∞. In this case, (17) holds as a strict inequality,
1 +B̃−F −c K 1
ZL
so λ11 = 0 for (21) to be satisfied. Rewriting the equality as KD
2 = F
cD
D D
, and combining
with (20) yields
õ ¶Ã −b
! µ ¶Ã −b
!!
ZL1 + B − FD ξ1 ZL1 + B − FD ξ2
KpD1 (B̃)0 = −b −b
, −b −b
. (25)
cD ξ +ξ cD ξ +ξ
1 2 1 2

The condition λ12 ≥ 0 should be satisfied at optimality. After some algebra, this condition implies
that (25) is optimal if B̃ ≥ cD 10 K1D + FD − ZL1 .
Combining cases 1, 2 and 3, KpDi (B̃) for i = 0, 1 is characterized by
 µ ¶−b

 (1+ 1b ) −b
 K0D = cD ξ if cD 10 K0D + FD − B̃ ≤ 0
p0
KD (B̃) = µ³ ´ µ −b ¶ ³ ´ µ −b ¶¶ (26)

 0 B−FD ξ1 B−FD ξ2
 KD = cD −b −b , cD −b −b if cD 10 K0D + FD − B̃ > 0,
ξ 1 +ξ 2 ξ 1 +ξ 2
 µ³ ´ µ −b
¶ ³ ´ µ −b
¶¶

 0
KD = B−FD ξ1
, B−FD ξ2
if cD 10 K1D + FD − B̃ ≤ 0

 cD −b −b cD −b −b

 µ ¶
ξ 1 +ξ 2 ξ 1 +ξ 2
 −b
KpD1 (B̃) = (1+ 1b ) −b
K1D = cD (1+a) ξ if 0 < cD 10 K1D + FD − B̃ ≤ E

 µ³


 0 ´ µ −b ¶ ³ ´ µ −b ¶¶

 KD = E+B−FD ξ1
, E+B−FD ξ2
if cD 10 K1D + FD − B̃ > E.
cD −b −b
ξ 1 +ξ 2 cD −b −b
ξ 1 +ξ 2

44
2. Solution for KpT (B̃) and ΨT (B̃):
To show that g(KT ) is strictly concave, we need to show that ∀ KIT , KII
T ≥ 0 and λ ∈ (0, 1),

g(λKIT + (1 − λ)KII I II
T ) − λg(KT ) − (1 − λ)g(KT ) > 0. (27)

Since g i (KT ) is strictly concave, we only need to focus on KIT , KII 0 I

T such that cT 1 KT +FT ≤ B̃ and
¡ ¢
cT 10 KII 0 I II
T + FT > B̃. We have two cases to consider. First, if cT 1 λKT + (1 − λ)KT + FT ≤ B̃
then after some algebra, the left-hand side of (27) becomes
¡ ¢1+ 1b 1+ 1 1+ 1b
E [NT ] 0 λKIT + (1 − λ)KII
T − λE [NT ] 0 KIT b − (1 − λ)E [NT ] 0 KII
T + (1 − λ)a(cT 10 KII
T + FT − B̃).

1
Since x1+ b is strictly concave for x ≥ 0 and cT 10 KII
T + FT − B̃ is positive by definition, the above
¡ ¢
equation is strictly greater than 0. Second, if cT 10 λKIT + (1 − λ)KIIT + FT > B̃ then after some
algebra, the left-hand side of (27) becomes
¡ ¢1+ 1b 1+ 1 1+ 1b
E [NT ] 0 λKIT + (1 − λ)KII
T − λE [NT ] 0 KIT b − (1 − λ)E [NT ] 0 KII
T − λaT (cT 10 KIT + FT − B̃).

1
Since x1+ b is strictly concave for x ≥ 0 and cT 10 KIT +FT − B̃ is negative by definition, the equation
above is strictly greater than 0. Since (27) is satisfied for both cases, g(KT ) is strictly concave. It
follows that

 arg max Ψi (B̃) if B̃ > FT
i T
KpT (B̃) = KpTi (B̃) where i =
 1 if B̃ ≤ FT

is the unique maximizer of g. Combining (16) and (26), the unique optimal solution to problem
(13) and the corresponding optimal amount of borrowing are given by


 K0T if cT 10 K0T + FT ≤ B̃



 K
p T if cT 10 K1T + FT ≤ B̃ < cT 10 K0T + FT
KT (B̃) = (28)

 K1T if cT 10 K1T + FT − E ≤ B̃ < cT 10 K1T + FT




KT if B̃ < cT 10 K1T + FT − E,
³ ´+
epT (B̃) = cT 10 KpT (B̃) + FT − B̃

45
where
à ¡ ¢ !−b à ¡ ¢ !−b 
ξ 1 1 + 1b ξ 2 1 + 1b
K0D 0 =  , 
cD cD
õ ¶Ã −b
! µ ¶Ã −b
!!
B − FD ξ1 B − FD ξ2
KD 0 = −b −b
, −b −b
cD ξ1 + ξ2 cD ξ1 + ξ2
à ¡ ¢ !−b à ¡ ¢ !−b 
ξ 1 1 + 1b ξ 2 1 + 1b
K1D 0 =  , 
cD (1 + a) cD (1 + a)
õ ¶Ã −b
! µ ¶Ã −b
!!
E + B − FD ξ1 E + B − FD ξ2
KD 0 = −b −b
, −b −b
cD ξ1 + ξ2 cD ξ1 + ξ2
à ¡ ¢ !−b
A 1 + 1b
KF0 =
cF
µ ¶
B − FF
KF =
cF
à ¡ ¢ !−b
A 1 + 1b
KF1 =
cF (1 + a)
µ ¶
E + B − FF
KF = .
cF

We substitute (28) in (13) and find


 cT 10 K0

 B̃ − (1 − γT )FT + −(b+1) T
+P if cT 10 K0T + FT ≤ B̃

 ³ ´1+ 1


 MT B̃−FT b
+ γT FT + P if cT 10 K1T + FT ≤ B̃ < cT 10 K0T + FT
cT
ΨT (B̃) = cT 10 K1
(29)
 T (1+a)

 (B̃ − FT )(1 + a) + −(b+1) + γT FT + P if cT 10 K1T + FT − E ≤ B̃ < cT 10 K1T + FT

 ³ ´1+ 1


 −E(1 + a) + MT E+B̃−FT b
+ γT FT + P if B̃ < cT 10 K1T + FT − E.
cT

·³ ´− 1 ¸ ³ ´ 1
−b −b b −b −b − b
where MF = E ξ1 + ξ2 and MD = ξ 1 + ξ 2 . It follows from (11) that ΨT (B̃) is
relevant (and is defined) only for B̃ > FT − E.
3. Solution for K∗T (B̃) and e∗T (B̃):
To complete the characterization of K∗T (B̃) and e∗T (B̃), we compare ΨT (B̃) with B̃ −(1−γT )FT +P
(the value of the not borrowing and not investing in capacity) for B̃ > FT − E and establish that
the two functions intersect at most once on B̃ ∈ (FT − E, ∞); and find K∗T (B̃) and e∗T (B̃).
.
For B̃ > FT − E, we define GT (B̃) = ΨT (B̃) − (B̃ − (1 − γT )FT + P ), the difference between
the equity values in (29) and not borrowing and not investing in capacity. It is easy to verify that,
lim + ΨT (B̃) = lim ˜k > FT − E therefore, ΨT (B̃) and, in turn, GT (B̃)
− ΨT (B̃) for ∀ B
B̃→B˜k B̃→B˜k

46
are continuous functions of B̃. We have


 0 if cT 10 K0T + FT ≤ B̃



 ³ ´1

 MT 1 B̃−FT b
∂GT (B̃) cT (1 + b ) cT −1 if cT 10 K1T + FT ≤ B̃ < cT 10 K0T + FT
= (30)
∂ B̃ 
 a if cT 10 K1T + FT − E ≤ B̃ < cT 10 K1T + FT



 ³ ´1
 MT
 1 E+B̃−FT b
cT (1 + b ) cT −1 if B̃ < cT 10 K1T + FT − E.

For cT 10 K1T + FT ≤ B̃ < cT 10 K0T + FT ,

à !1
MT 1 B̃ − FT b
MT 1 ¡ ¢1
(1 + ) −1> (1 + ) 10 K0T b − 1 = 0, (31)
cT b cT cT b

and for B̃ < cT 10 K1T + FT − E,

à !1
MT 1 E + B̃ − FT b
MT 1 ¡ ¢1
(1 + ) −1> (1 + ) 10 K1T b − 1 = a. (32)
cT b cT cT b

∂ ∂
It follows that limB̃→B˜ + G (B̃)
∂ B̃ T
= limB̃→B˜ − G (B̃)
∂ B̃ T
on the domain of GT (.). Therefore
k k
∂
GT (B̃) is differentiable for B̃ > FT − E and G (B̃)
∂ B̃ T
≥ 0 with equality holding only for B̃ ≥
cT 10 K0T + FT . For cT 10 K0T + FT ≤ B̃,

cT 10 K0T cT 10 K0T
GT (B̃) = B̃ − (1 − γT )FT + + P − (B̃ − (1 − γT )FT + P ) = > 0. (33)
−(b + 1) −(b + 1)

We showed that GT (B̃) strictly increases for B̃ ∈ (FT − E, cT 10 K0T + FT ) and is positive for
bT denote the budget level at which the two equity value curves
B̃ ∈ [cT 10 K0 + FT , ∞). Let B
T
bT ) = 0. For FT ≥ E, we have lim
intersect, i.e. GT (B B̃→(FT −E)+ GT (B̃) = −aE < 0. Since GT (B̃)
strictly increases in B̃, it follows that for FT ≥ E, there exists a unique BbT > FT − E such that
GT (BbT ) = 0. For FT < E, the domain of GT (B̃) is [0, ∞). For notational convenience, we let
bT =.
B 0 if the two curves do not intersect on this domain (GT (B̃) > 0 for B̃ ≥ 0). Since GT (B̃)
bT , if it exists on [0, ∞), is unique. For B̃ ≤ B
strictly increases in B̃, it follows that for FT < E, B bT
we have K∗T (B̃) = 0 and e∗T (B̃) = 0. Combining this with (28) gives the desired result.
Proof of Corollary 1: The expected (stage 1) equity value of the firm with a given budget level
B̃ follows directly from Proposition 2:

 cT 10 K0

 B̃ − (1 − γT )FT + −(b+1) T
+P if B̃ ∈ Ω0T

 ³ ´1+ 1



 MT B̃−F T b
+ γT FT + P if B̃ ∈ Ω1T

 cT
c 10 K1
T (1+a)
πT (B̃) = (B̃ − FT )(1 + a) + T −(b+1) + γT FT + P if B̃ ∈ Ω2T (34)

 ³ ´1+ 1


 −E(1 + a) + MT E+B̃−FT

b
+ γT FT + P if B̃ ∈ Ω3T

 cT


 B̃ − (1 − γ )F + P if B̃ ∈ Ω4T
T T

47
 ·³ ´− 1 ¸ ³ ´ 1
−b −b − b
ξ1−b ξ2−b
b
where MF = E + and MD = ξ 1 + ξ 2 .
We calculate


 (1, 0) if B̃ ∈ Ω0T

 µ ¶

 ³ ´1

 MT B̃−FT b 1 MT 1
−1
à !  cT (1 + 1/b) cT , b c(1+1/b) (1 + 1/b)(B̃ − F T ) b if B̃ ∈ Ω1T

 T
∂πT (B̃) ∂ 2 πT (B̃)
, = (1 + a, 0) if B̃ ∈ Ω2T
∂ B̃ ∂ B̃ 2 
 µ ³ ´1 ¶

 1
 MT E+B̃−FT b 1 MT −1
if B̃ ∈ Ω3T


 cT (1 + 1/b) cT , b (1+1/b) (1 + 1/b)(E + B̃ − FT ) b
cT



 (1, 0) if B̃ ∈ Ω4T

at the points where πT (B̃) is differentiable. It is easy to verify that limB̃→B˜ + ∂∂B̃ πT (B̃) =
k

limB̃→B˜ − ∂∂B̃ πT (B̃) for B˜k ∈ Ω0123

T , and πT ( B̃) is differentiable everywhere in its domain except at
k

BbT . Since πT (B̃) is a continuous function of B̃ it follows that πT (B̃) is strictly increasing in B̃.
∂2
We have π (B̃)
∂ B̃ 2 T
≤ 0 for each ΩiT and πT (B̃) is piecewise concave. From (31) we obtain
∂
π (B̃)
∂ B̃ T
> 1 for B̃ ∈ Ω1T and from (32) we have ∂∂B̃ πT (B̃) > 1 + a for B̃ ∈ Ω3T . Since πT (B̃) is
bT it follows that πT (B̃) is concave in B̃ for B̃ ≥ B
only kinked at B bT , but not globally concave.
Proof of Proposition 3: The optimal risk management level HT∗ is given by

HT∗ = argmaxHT E [πT (BF RM (α1 , HT ))] (35)

ω0F RM
s.t. − ≤ HT ≤ ω1F RM
α1

Since ξ and α1 are independent,

Eξ,α1 [πT (BF RM (α1 , HT ))] = Eα1 [Eξ [πT (BF RM (α1 , HT ))]] = Eα1 [πT (BF RM (α1 , HT ))] .

Therefore we can write the expectation in (35) over α1 . Let rα1 (.) and Rα1 (.) denote the density and
distribution function of α1 , respectively. Since BF RM (α1 , HT ) = ω0F RM + α1 (ω1F RM − HT ) + α1 HT ,
for each HT the unique distribution function of BF RM (HT ) is
à !
B̃ − ω0F RM − α1 HT
RBF RM (HT ) (B̃) = Rα1 B̃ ≥ ω0F RM + α1 HT . (36)
ω1F RM − HT

It follows that HT determines the range and the probability distribution of the available budget in
stage 1. Since we do not impose any specific assumption on the type of the distribution of α1 , we will
use general structural properties of the optimization problem (35) to solve for HT∗ . In particular,
we will focus on the functional form of πT (B̃) since the expected (stage 0) value of the equity is
the expectation of this function with respect to the budget random variable. We first provide the
following lemma that we will use throughout the proof. The proof is relegated to Appendix C.

48
 bT = 0 iff FT ≤ F T , and B
Lemma 1 There exist unique fixed cost threshold F T such that B bT > 0
iff FT > F T .

We now conclude the proof by analyzing each case in Proposition 3.

Case (i), FT ≤ F T :
bT = 0. Since B
It follows from Lemma 1 that B bT = 0, from Corollary 1 we have that πT (B̃) is
concave for B̃ ≥ 0. From Jensen’s inequality,
¡ ¢
E [πT (BF RM (α1 , HT ))] ≤ πT (E[BF RM (α1 , HT )]) = πT ω0F RM + α1 ω1F RM )
¡ ¡ ¢¢
= πT BF RM α1 , ω1F RM (37)
h F RM i
ω
for HT ∈ − 0α1 , ω1F RM . This implies that HT∗ = ω1F RM .
Case (ii), FT > F T :
cT > 0 and we cannot guarantee the concavity of πT for the whole range
From Lemma 1, we have B
of B̃. Therefore Jensen’s inequality is not sufficient to find HT∗ . In this case, HT∗ is either a solution
∂ ω0F RM
to the first order condition ∂HT E[πT ] = 0, or occurs at a boundary, i.e. HT∗ ∈ {− F RM }.
α1 , ω1
To write the first-order condition, we utilize the following lemma proven in Appendix C:

Lemma 2 For any argument κT of πT , the expectation and the derivative operators can be inter-
h i
changed, i.e. ∂κ∂T E[πT ] = E ∂κ∂T πT .

Let
. cT 10 K0
T +FT −ω0
F RM −H α . cT 10 K1
T +FT −ω0
F RM −H α
αT0 = F
ω1 RM −HT
T 1
, αT1 = F
ω1 RM −HT
T 1
,
. 0 1
cT 1 KT +FT −E−ω0 F RM −HT α1 . BbT −ω F RM −HT α1
αT2 = ω1F RM −HT
, αTB = 0
ω1F RM −HT
.
∂
From Lemma 2 (letting κT = HT ), we can write the first-order condition ∂HT E[πT ] by using the
expression for πT (B̃) in (34) of Corollary 1 and the equivalence in (36). The integration ranges
correspond to the regions ΩiT in (34) of Corollary 1.
· ¸ Z ∞
∂πT
E = (α1 − x) rα1 (x) dx (38)
∂HT max(α0T ,0)
Z max(α0 ,0) µ F RM ¶1
T MT 1 ω0 + x(ω1F RM − HT ) + α1 HT − FT b
+ (1 + ) (α1 − x) rα1 (x) dx
max(α1T ,0) cT b cT
Z max(α1 ,0)
T
+ (α1 − x)(1 + a) rα1 (x) dx
max(α2T ,0,αB
T)
Z max(α2T ,0,αB µ ¶ 1b
T) MT 1 ω0F RM + x(ω1F RM − HT ) + α1 HT + E − FT
+ (1 + ) (α1 − x) rα1 (x) dx
max(0,αB
T)
cT b cT
Z max(0,αB
T)
+ (α1 − x) rα1 (x) dx
0

49
Both the limits of integration and the integrants in (38) are functions of HT . Since we do not
impose any distributional assumptions on α1 it is not always possible to find a closed-form solution
for HT∗ .
bT , αB ≥ α1 . Therefore, for
We have αT0 > αT1 > αT2 by definition. For ω0F RM + α1 ω1F RM ≤ B T
F
ω0 RM F
+ α1 ω1 RM b
≤ BT , we either have αT > αT > αt > αT ≥ α1 > 0 or αT0 > αT1 > αTB ≥ α1 >
0 1 2 B

0 > αT2 . Similar to (31) and (32) we establish

Z α0T µ ¶ 1b
MT 1 ω0F RM + x(ω1F RM − HT ) + α1 HT − FT
(1 + ) (α1 − x) rα1 (x) dx
α1T cT b cT
Z α0T
< (α1 − x) rα1 (x) dx,
α1T

Z max(α2T ,αB µ ¶1
T) MT 1 ω0F RM + x(ω1F RM − HT ) + α1 HT − FT b
(1 + ) (α1 − x) rα1 (x) dx
αB
T
cT b cT
Z max(α2 ,αB )
T T
< (α1 − x)(1 + a) rα1 (x) dx.
αB
T

It follows that
Z ∞ Z α1T
∂πT
< (α1 − x) rα1 (x) dx + a (α1 − x) rα1 (x) dx. (39)
∂HT 0 αB
T

∂πT ω0F RM
The first term is equal to 0 and the second term is negative, therefore ∂HT < 0 and HT∗ = − α1 .
This concludes the proof for part (1) of this case.
h i¯
bT , then H ∗ either satisfies E ∂πT ¯¯
If ω0F RM + α1 ω1F RM > B = 0 or occurs at a boundary
T ∂HT HT∗
ω0F RM F RM }
{− α1 , ω 1 depending on the distributions of α1 and ξ. From Jensen’s inequality, ω1F RM
BbT −ω F RM
dominates HT ≥ α1
0
because by (36) and Corollary 1, πT (BF RM (α1 , HT )) is concave over
bT −ω F RM
nn b −ω F RM
o © ªo
B ∗ ∈ B
its domain for HT ≥ 0
α1 . It follows that HT HT < T α10 ∪ ω1F RM .
Proof of Proposition 4: We first prove the existence of cF (cD , H∗ ). Notice from (35) that the
optimal financial risk management level HT∗ depends on cT 4 . For each financial risk management
level HT , the expected (stage 0) equity value E [πT (cT , BF RM (α1 , HT ))] is a continuous function
of cT . It follows that the expected (stage 0) equity value at the optimal risk management level
E [πT (cT , BF RM (α1 , HT∗ (cT )))] is also a continuous function of cT (because it is the upper envelope
∗ (c )))] is also finite. It is
of continuous functions). For a finite cD > 0, E [πD (cD , BF RM (α1 , HD D

easy to prove that

lim E [πF (cF , BF RM (α1 , HF∗ (cF )))] = ω0F RM + α1 ω1F RM − (1 − γF )FF + P,
cF →∞
lim E [πF (cF , BF RM (α1 , HF∗ (cF )))] → ∞.
cF →0
4
Since from Proposition 3 we cannot guarantee the uniqueness of HT∗ , HT∗ (cT ) is a correspondence.

50
 ∗ (c )))] > ω F RM + α ω F RM −
Since the equity value is continuous in cF , if E [πD (cD , BF RM (α1 , HD D 0 1 1

(1 − γF )FF + P , then there exists a cF such that the equity values with both technologies coincide.
∗ (c )))] ≤ ω F RM +α ω F RM −(1−γ )F +P then the threshold does not
If E [πD (cD , BF RM (α1 , HD D 0 1 1 F F

exist and the flexible technology is always preferred over the dedicated technology. This concludes
the proof for existence of cF (cD , H∗ ). The existence of cF (cD , 0) can be proven in the same manner
by substituting BF RM (.) with B−F RM (.) and HT∗ (cT ) with 0.
To prove the uniqueness of cF (cD , H∗ ) and cF (cD , 0) we first provide the following lemma and
relegate the proof to Appendix C:

Lemma 3 In the optimal set of financial risk management levels, for a fixed level of H, the expected
(stage 0) value of the equity with technology T strictly decreases in the unit capacity investment
cost ( ∂c∂T E[πT (cT , BF RM (α1 , H))] < 0).

From Lemma 3 it follows that the expected (stage 0) equity value with flexible technology is
strictly decreasing in cF for any (relevant) financial risk management level HF . This implies the
uniqueness of cF (cD , H∗ ). The uniqueness of cF (cD , 0) follows from Lemma 3 using the identity
B−F RM (α1 ) = BF RM (α1 , H) for H = 0 and FF RM = 0. For the comparative statics results with
respect to demand variability and correlation we first provide the following
·³ two lemmas and relegate
´− 1 ¸
their proofs to Appendix C. Recall from Corollary 1 that MF (ξ) = E ξ1−b + ξ2−b
b
.

0 0
Lemma 4 MF (ξ) ≤ MF (ξ ) for ξ that is obtained from ξ with an increase in σ in one of the
following ways:
0
i) ξ is obtained by an increase in σ where ξ has a symmetric bivariate lognormal distribution,
0 0 0 0
ii) ξ and ξ have independent marginal distributions, equal means (ξ = ξ ), and ξi ºv ξi (ξi is
stochastically more variable than ξi ) for i = 1, 2 or the variability ordering holds for only one
of the marginals and the other marginal is identical,
0
iii) ξ is random (σ 6= 0) while ξ is deterministic (σ = 0).

0 0
Lemma 5 MF (ξ) ≥ MF (ξ ) for ξ that is obtained from ξ with an increase in ρ in one of the
following ways:
0
i) ξ is obtained by an increase in ρ where ξ has a symmetric bivariate lognormal distribution,
0 0
ii) ξ dominates ξ according to the concordance ordering (ξ ºc ξ),
0
iii) ξ is perfectly positively correlated (ρ = 1) and ξ is less than perfectly positively correlated
(ρ < 1).

51
In Lemma 4 and Lemma 5, case i imposes distributional assumptions on ξ to analyze the effect
of σ and ρ, respectively. Case ii of each lemma analyzes different stochastic orderings to capture
the effect of product market conditions. Variability ordering is often used in the literature to
0
analyze the effect of increasing variability. Concordance ordering ξ ºc ξ, as stated in Corbett and
0 0
Rajaram (2005, p. 13), essentially means that (ξ1 , ξ2 ) move together more closely than (ξ1 , ξ2 ).
Case iii focuses on limiting cases.
To establish the comparative statics results, we provide the following lemma and relegate the
proof to Appendix C:

Lemma 6 In the optimal set of financial risk management levels, for a fixed level of H, the expected
(stage 0) value of the equity with technology T

i) strictly decreases in the fixed cost of technology and strictly increases in the salvage rate
( ∂F∂T E[πT (FT , BF RM (α1 , H))] < 0 and ∂
∂γT E[πT (γT , BF RM (α1 , H))] > 0),

∂
ii) decreases in unit financing cost ( ∂a E[πT (a, BF RM (α1 , H))] ≤ 0), and the equality only holds
for H such that ω0F RM + α1 H ≥ cT 10 K1T + FT ,

∂
iii) increases in credit limit ( ∂E E[πT (E, BF RM (α1 , H))] ≥ 0), and the equality only holds for H
such that ω0F RM + α1 H ≥ cT 10 K1T + FT − E,

∂
iv) increases in demand variability ( ∂σ E[πT (E, BF RM (α1 , H))] ≥ 0),

∂
v) decreases in demand correlation ( ∂ρ E[πT (E, BF RM (α1 , H))] ≤ 0).

Since the expected (stage 0) equity value is a continuous function of parameters a, E, FT , γT , ρ, σ for
a given financial risk management level H, the expected (stage 0) equity value at the optimal risk
management level (which also depends on these parameters) is also continuous in these parameters.
Therefore the monotonic relations stated in Lemma 6 are also satisfied in the weak sense (not
strict inequality) at the optimal financial risk management level without assuming differentiability
(because the expected (stage 0) equity value might not be differentiable at the points where the
optimal financial risk management level changes). The comparative static results for cF (cD , H∗ )
follow from Lemma 6. The comparative static results for cF (cD , 0) also follow from Lemma 6 using
the identity B−F RM (α1 ) = BF RM (α1 , H) for H = 0 and FF RM = 0.
With symmetric fixed costs and salvage rates, we establish the functional form of cSF (cD ) with
the following Lemma and relegate the proof to Appendix C:

Lemma 7 When the fixed costs and the salvage rates of the two technologies are symmetric, at cF =
cSF (cD ) expected (stage 1) equity values, expected (stage 0) equity values at an arbitrary financial

52
risk management level H and the optimal financial risk management actions are the same for both
¯
¯
technologies, i.e. πF (cF , B̃)¯ S
= πD (cD , B̃) for B̃ ≥ 0, E[πF (cSF (cD ), BF RM (α1 , H))] =
cF =cF (cD )
E[πD (cD , BF RM (α1 , H))] and HF∗ (cSF (cD )) = HD
∗ (c ).
D

It follows from Lemma 7 that cSF (cD ) is the unique threshold with financial risk management in
the symmetric case (cF (cD , H∗ ) = cSF (cD )). Using the identity B−F RM (α1 ) = BF RM (α1 , H) for
H = 0 and FF RM = 0, it follows from Lemma 7 that cSF (cD ) is also the unique threshold without
financial risk management in the symmetric case (cF (cD , 0) = cSF (cD )). We now prove the relation
cSF (cD ) ≥ cD . It is sufficient to show
·³ ´− 1 ¸
E−b ξ1−b + ξ2−b
b
≥ E−b [ξ1 ] + E−b [ξ2 ].

From Hardy et al. (1988, p.133,146) if d ∈ (0, 1) and X and Y are non-negative random variables
then the following is true:
h i
E 1/d (X + Y )d ≥ E1/d [X d ] + E1/d [Y d ] (40)

where the equality only holds when X and Y are effectively proportional, i.e. X = λY . In the
expression for cSF (cD ) we have d = − 1b ∈ (0, 1) and ξ > 0 therefore we can use this inequality.
Replacing X with ξ1−b and Y with ξ2−b gives the desired result. Notice that cSF (cD ) = cD only if
ξ1 = kξ2 for k > 0. This is only possible if either ξ is deterministic or it is perfectly positively
correlated and has a proportional bivariate distribution.
Proof of Corollary 2: If the capital markets are perfect we have E = P ≥ cT 10 K0T + FT and
a = 0 (as we discussed in Assumption 6). Since we have Ω1234
T = ∅, it follows from Proposition 2 that
the firm invests in the budget-unconstrained capacity investment level for any budget realization,
K∗T (B̃) = K0T , and borrows to finance this capacity level, e∗T (B̃) = [cT 10 K0T + FT − B]+ . We obtain

cT 10 K0T
E [πT (B−F RM (α1 ))] = E [πT (BF RM (α1 , H))]|FF RM =0 = ω0 + α1 ω1 − (1 − γT )FT + + P,
−(b + 1)
T
and it follows from Proposition 12 that F F RM = 0 for T ∈ {D, F }. If the product markets are
perfect (Σ = 0), then with symmetric fixed costs and salvage rates, it follows from Proposition 4
that cF (cD , H∗ ) = cF (cD , 0) = cD .
Proof of Corollary 3: The proof of the first argument follows from Proposition 3. For the second
argument, we provide a numerical example where the firm optimally fully speculates with flexible
technology and fully hedges with dedicated technology. We focus on the case with FF RM = 0 such
that financial risk management is costless. The horizontal line in Figure 1 denotes the value of not
investing any technology; hence the firm optimally chooses flexible technology with full speculation
in this example.

53
Figure 1: Optimal Speculation is triggered by flexible technology investment: Dedicated technology
∗ = ω = 4) is dominated by flexible technology with full speculation (H ∗ =
with full hedging (HD 1 F
−w
α1 = −0.61.
0

Proof of Proposition 5: With a hedging constraint, the range of forward contracts is [0, ω1F RM ]
∂πT
in (35). Substituting FF RM = 0 in (38) of Proposition 3, similar to (39), we obtain ∂HT < 0. It
follows that HT∗ = 0.
Proof of Corollary 4: It follows from Proposition 4 that for symmetric fixed costs and salvage
rates of technologies and for FF RM = 0, the optimal risk management portfolio is flexible (dedi-
cated) technology with financial risk management if cF < cSF (cD ) (cF > cSF (cD )). From the proof of
ω0
Proposition 12, for β = ω0 +α0 ω1 we can have a sufficiently large feasible FF RM such that engaging
in financial risk management is not profitable. In this case, the optimal risk management portfolio
is flexible (dedicated) technology with financial risk management if cF < cSF (cD ) (cF > cSF (cD )).
Proof of Proposition 6: The invariance of cF (cD , H∗ ) and cF (cD , 0) to the unit financing cost,
the fixed cost of both technologies and the internal endowment of the firm follows from the definition
of cSF (cD ) in Proposition 4. For F = FD < FF = F + δ with δ > 0, we obtain cF (cD , H∗ ) < cSF (cD )
and cF (cD , 0) < cSF (cD ) from Proposition 4. We first provide the proof of the results with respect
to technology fixed costs. Comparative statics with respect to the internal endowment follow from
a similar argument. We define
.
S −F RM (cF ) = E[πF (cF , F + δ, B−F RM (α1 ))] − E[πD (cD , F, B−F RM (α1 ))] where S −F RM (cF (cD , 0)) = 0.(41)
³ ´−1 ¯¯
∂
From the implicit function theorem we have ∂F cF (cD , 0) = − ∂F S ∂ −F RM ∂ −F RM ¯
∂cF S ¯ .
cF (cD ,0)
From Lemma 2, we can interchange derivative and expectation operators, and using Lemma 3 with
B−F RM (α1 ) = BF RM (α1 , H) for H = 0 and FF RM = 0, we obtain
¯ · ¸¯
∂S −F RM ¯¯ ∂πF (B−F RM (α1 )) ¯¯
=E < 0.
∂cF ¯cF (cD ,0) ∂cF ¯
cF (cD ,0)

54
Similarly we have
¯ · · ¸ · ¸¸¯
∂S −F RM ¯¯ ∂πF (B−F RM (α1 )) ∂πD (B−F RM (α1 ))) ¯¯
¯ = E −E ¯ .
∂F cF (cD ,0) ∂F ∂F cF (cD ,0)
¯
Since cF (cD , 0) < cSF and δ > 0 it follows that cF KiF ¯c + F + δ > cD 10 KiD + F for i = 0, 1.
F (cD ,0)
This implies that Ω0F ⊂ Ω0D and Ω2F ⊃ Ω2D . We obtain
¯ Z
∂S −F RM ¯¯
¯ = T
(−1 + 1) dRB−F RM (B̃) (42)
∂F cF (cD ,0) Ω0F Ω0D
¡ ¢Ã ! 1b 
Z
MF 1 + 1b B̃ − F − δ
+ − + 1 dRB−F RM (B̃)
T 0 cF (cD , 0) cF (cD , 0)
Ω1F ΩD

Z µ ¶" ³ ´ 1b
#
1 MF MD ³ ´ 1b
+ T 1
1+ − 1 B̃ − F − δ + 1+ 1 B̃ − F dRB−F RM (B̃)
Ω1F ΩD b (cF (cD , 0))1+ b cD b
Z
+ T
(−(1 + a) + 1) dRB−F RM (B̃)
Ω2F Ω0D
Z " #
1 MD ³ ´ 1b
+ T
−(1 + a) + (1 + ) 1+ 1 B̃ − F dRB−F RM (B̃)
Ω2F Ω1D b c b
D
Z
+ T
(−(1 + a) + (1 + a)) dRB−F RM (B̃).
Ω2F Ω2D

∂ −F RM ¯
¯ T T
From (31), we have ∂F S cF (cD ,0)
< 0 for B̃ ∈ Ω1F Ω0D and B̃ ∈ Ω2F Ω1D . From Lemma 7,
¯
we have MD
1+ 1
= MF
1+ 1
. Since cF (cD , 0) < cSF (cD ) and δ > ∂
0, we obtain ∂F S −F RM ¯c (c ,0) < 0
cD b (cS (cD )) b F D

T
F ¯ ¯
¯
for B̃ ∈ Ω1F 1 ∂
ΩD . In conclusion, we have ∂cF S −F RM
¯ < 0 and ∂
∂F S −F RM ¯ cF (cD ,0)
≤ 0. It
cF (cD ,0)
∂
follows from the implicit function theorem that ∂F cF (cD , 0) ≤ 0 where the equality holds only for
¯
ω0 > cF K0F ¯c (c ,0) + F + δ.
F D

To prove the result for cF (cD , H∗ ), we define S F RM (cF ), the counterpart of (41) by replacing
B−F RM (α1 ) with BF RM (α1 , HT∗ ). We have HF∗ (cF ) = HD ∗ = ω F RM for c
1
∗
F = cF (cD , H ). We
¯ ¯
¯ ¯
establish ∂c∂F S F RM ¯ ∗
< 0 using ∂c∂F HF∗ ¯ = 0. The rest of the proof follows in a
cF (cD ,H ) ¯ cF (cD ,H∗ )
similar manner using the facts that ∂F ∂
HT∗ ¯c (c ,H∗ ) = 0 and that with full-hedging ω0F RM +α1 ω1F RM
F D

is realized in only one of the regions in (42). In conclusion, it follows from the implicit function
∂ T
theorem that ∂F cF (cD , H∗ ) ≤ 0 where the equality holds only for ω0F RM + α1 ω1F RM ∈ Ω0F Ω0D or
T
ω0F RM + α1 ω1F RM ∈ Ω2F Ω2D .
To prove the results with respect to the unit financing cost for cF (cD , 0), we follow the same

55
steps by replacing F with a in S −F RM (cF ). We obtain
¯ Z ³ ´
∂S −F RM ¯¯ 1
¯
¯
¯ = B̃ − (cF KF + FF ) cF (cD ,0)
dRB−F RM (B̃) (43)
∂a cF (cD ,0) Ω2F \Ω2D
Z ³ ´
¯
+ c D 10 1
K D + F D − (cF K 1
F + FF )¯ dRB−F RM (B̃)
T c (c ,0)F D
Ω2F Ω2D
Z
+ −E dRB−F RM (B̃)
Ω3F \Ω23
Z D
³ ´
¯
+ T
−B̃ + (cF K1F + FF )¯c −E dRB−F RM (B̃).
F (cD ,0)
Ω3F Ω2D

The first term and the last integrands are negative by the definition of the regions. From above
¯
(comparative static with respect to fixed cost) we have cD 10 K1D + FD < cF K1F ¯c (c ,0) + FF .
∂ −F RM ¯
¯ ∂
F D

This implies ∂a S cF (cD ,0)

≤ 0. We conclude ∂a cF (cD , 0) ≥ 0 where the equality holds for
¯
ω0 > cF K1 ¯ + FF .
F cF (cD ,0)
∂
The result for cF (cD , H∗ ) can be proven in a similar fashion. It follows that ∗
∂a cF (cD , H ) ≥ 0
S¡ 3 T 3 ¢
where the equality holds if ω0F RM + α1 ω1F RM ∈ Ω01
F ΩF ΩD .
Proof of Proposition 7: We only prove the results for small firms. Results related to large firms
follow from similar arguments. We define
£ ¡ ¢¤
ϕ . ∂∆T ∂E πT BF RM (α1 , ω1F RM ) ∂E [πT (B−F RM (α1 ))]
Υ = = − (44)
∂ϕ ∂ϕ ∂ϕ

as the derivative of the value of full hedging with respect to the argument ϕ. For small firms, we
£ ¡ ¢¤ c 10 K1
T (1+a)
have E πT BF RM (α1 , ω1F RM ) = (ω0F RM + α1 ω1F RM − FT )(1 + a) + T −(b+1) + γT FT + P . We
analyze each comparative static result separately.
Fixed cost of technology. We obtain

Z Ã Z !1
FT 1 MT B̃ − FT b
Υ = −(1 + a) − −1 dRB−F RM (B̃) − (1 + ) dRB−F RM (B̃)
Ω0T Ω1T b cT cT
Z
− −(1 + a) dRB−F RM (B̃).
Ω2T

³ ´1
It is easy to show that (1 + 1b ) M B̃−FT b
cT
T
cT < 1 + a for B̃ ∈ Ω1T . It follows that ΥFT < 0.
λω0
Initial endowment. After parameterizing the initial endowment, we obtain β λ = λω0 +α 0 λω1
=
ω0 F RM F RM . 1−β
ω0 +α0 ω1 = β. We have (ω0 , ω1 ) = (λω0 −βFF RM , λω1 − α0 FF RM ) and for small firms, it fol-
∂E[πT (BF RM (α1 ,ω1F RM ))]
lows that ∂λ = (ω0 + α1 ω1 )(1 + a). After parameterizing the initial endowment,

56
 cT 10 K0T +FT cT 10 K1T +FT
. −ω0 . −ω0
we define αT0 = λ
ω1 , αT1 = λ
ω1 . We obtain
Z ∞
λ
Υ = (ω0 + α1 ω1 )(1 + a) − (ω0 + xω1 ) rα1 (x) dx
max(α0T ,0)
Z max(α0T ,0) µ ¶1
MT (1 + 1/b) λ(ω0 + xω1 ) − FT b
− (ω0 + xω1 ) rα1 (x) dx
max(α1T ,0) cT cT
Z max(α1T ,0)
− (ω0 + xω1 )(1 + a) rα1 (x) dx
0

Notice that negative terms above are the expected value of the following function

 if α1 ≥ αT0
 ω0 + α1 ω1

 ³ ´1
f (α1 ) = MT (1+1/b) λ(ω0 +α1 ω1 )−FT b
 cT cT (ω0 + α1 ω1 ) if αT0 > α1 ≥ αT1


 (ω + α ω )(1 + a) if α1 < αT1
0 1 1

with respect to the asset price distribution α1 . It is easy to prove that (ω0 +α1 ω1 )(1+a) ≥ f (α1 ) for
α1 ≥ 0 with strict inequality for some α1 . It follows that E[(ω0 +α1 ω1 )(1+a)] = (ω0 +α1 ω1 )(1+a) >
E[f (α1 )] and we obtain Υλ > 0.
To analyze the effect of cash holdings (ω0 ) on the value of financial risk management, we only
parameterize the cash holdings as (λ0 ω0 , ω1 ) and set β = 0 such that FF RM is only deducted from
FF RM 0
the value of asset holdings ω1 . It follows that ω0F RM = λ0 ω0 and ω1F RM = ω1 − α0 . Υλ > 0
follows from the similar lines with Υλ > 0.
Demand variability and correlation. We only provide the proof for demand variability. The
proof for demand correlation is along the similar lines. It is sufficient to focus on flexible technology
because dedicated technology is not affected from changes in σ and ρ. We obtain
à !−b−1 Z µ ¶
σ ∂MF cF MF (1 + 1b ) ∂MF 1 −b−1
Υ = − cF MF (1 + ) dRB−F RM (B̃)
∂σ 1+a Ω0F ∂σ b
Z Ã !1+ 1 Z Ã !−b−1
∂MF B̃ − FF b
∂MF cF MF (1 + 1b )
− dRB−F RM (B̃) − dRB−F RM (B̃).
Ω1F ∂σ cF Ω2F ∂σ 1+a
³ ´1+ 1 ³ ´
B̃−FF b cF MF (1+ 1b ) −b−1 ∂
It is easy to show cF > 1+a for B̃ ∈ Ω1F . From Lemma 4, we have ∂σ MF ≥0
and it follows that Υσ ≤ 0.
Unit financing cost. We obtain
Z
a (1 − β)α1
Υ = ω0 + α1 ω1 − (β + )FF RM − cT 10 K1T − FT − (B̃ − cT 10 K1T − FT ) dRB−F RM (B̃).
α0 Ω2T

It follows that for ω0 > cT 10 K1T + FT , when the non-hedged firm does not borrow at all, we have
Υa < 0. We focus on the case where the firm borrows at some budget states without financial risk

57
management (ω0 < cT 10 K1T + FT ).
R
For FF RM = 0, we have Υa = Ω2 (cT 10 K1T + FT − B̃) dRB−F RM (B̃) − (cT 10 K1T + FT − B) where
T

B = ω0 + α1 ω1 . Notice that the first term is the expected value of the function

 c 10 K1 + F − B if B̃ ≤ c 10 K1 + F
T T T T T T
f (B̃) =
 0 0 1
if B̃ > cT 1 KT + FT

with respect to the budget distribution. Since f (B̃) is a convex function, Υa ≥ 0 for FF RM = 0
follows from Jensen’s inequality.
For FF RM > 0, we have
Z
(1 − β)α1
Υa = (cT 10 K1T + FT − B̃) dRB−F RM (B̃) − (cT 10 K1T + FT + (β + )FF RM − B).
2
ΩT α0

We observe that the first term is strictly less than cT 10 K1T + FT − ω0 . For FF RM ≥ FF0 RM =
α0 ω1 (1−β)α1
, we obtain cT 10 K1T + FT + (β + )FF RM − B ≥ cT 10 K1T + FT − ω0 and it follows
(1−β)+ αβ α0
1
α0 ω1
that Υa < 0. Notice that FF RM ≤ (1−β) is the feasiblity condition; hence such FF RM exists.
∂ a (1−β)α1
We calculate ∂FF RM Υ = −(β + α0 ) < 0. Since Υa strictly decreases in FF RM , Υa ≥ 0 for
FF RM = 0 and Υa < 0 for FF0 RM , we conclude that there exists a unique FbF RM such that Υa < 0
for FF RM > FbF RM and Υa ≥ 0 for FF RM ≤ FbF RM .
Proof of Proposition 8: We focus on the case where it is profitable for the firm to engage
in financial risk management. To prove the proposition, we use the ordering between cF (cD , H∗ )
and cF (cD , 0). If cF (cD , 0) < cF (cD , H∗ ) (cF (cD , 0) > cF (cD , H∗ )) then flexible technology and
financial risk management are complements (substitutes) because engaging in financial risk man-
agement enables the firm to invest in flexible (dedicated) technology at some technology cost levels
where dedicated (flexible) technology was more profitable without financial risk management. From
Proposition 4, we obtain cF (cD , H∗ ) < cSF (cD ) and cF (cD , 0) < cSF (cD ). From Assumption 8, we
∗ (c ) = H ∗ (c (c , H∗ )) = ω F RM . From Lemma 3, it follows that c (c , H∗ ) ≷ c (c , 0)
have HD D F F D 1 F D F D

if and only if
£ ¤ £ ¤
E πD (BF RM (α1 , ω1F RM )) ≶ E πF (cF (cD , 0), BF RM (α1 , ω1F RM ))) . (45)

Recall that 4T (cT , FT ) is the value of financial risk management with technology T ∈ {D, F } at
given cost parameters (cT , FT ) as defined in (5). Inequality (45) holds if and only if 4F (cF (cD , 0), FF ) ≷
4D (cD , FD ). We will use the relation between 4F (cF (cD , 0), FF ) and 4D (cD , FD ) to prove the
proposition. We provide the following lemma and relegate the proof to Appendix C.

Lemma 8 For FT < F T and E > cT 10 K1T + FT ,

∂
(i) If ω0F RM + α1 ω1F RM ∈ Ω0T (∈ Ω2T ) then ∂FT 4T ≥ 0 ( ∂F∂T 4T < 0);
(ii) If ω0F RM + α1 ω1F RM ∈ Ω0T (∈ Ω2T ) then ∂c∂T 4T ≤ 0 ( ∂c∂T 4T > 0).

58
For large firms (ω0F RM + α1 ω1F RM ∈ Ω0F ), we obtain from Lemma 8, cSF (cD ) > cF (cD , 0) and
FF ≥ FD that

4D (cD , FD ) = 4F (cSF (cD ), FD ) ≤ 4F (cF (cD , 0), FD ) ≤ 4F (cF (cD , 0), FF ).

From the proof of Lemma 8, the inequalities above are strict for sufficiently low ω0 . We conclude
that cF (cD , H∗ ) ≥ cF (cD , 0) and large firms tend to use flexible technology and financial risk
management as complements.
For small firms (ω0F RM + α1 ω1F RM ∈ Ω2F ), we obtain

4D (cD , FD ) = 4F (cSF (cD ), FD ) > 4F (cF (cD , 0), FD ) > 4F (cF (cD , 0), FF ).

We conclude that cF (cD , H∗ ) < cF (cD , 0) and small firms tend to substitute flexible technology
with financial risk management.
Proof of Proposition 9: We only prove the results for small firms. Results related to large firms
follow from similar arguments. Recall that in the proof of Proposition 6 we defined
£ ¡ ¢¤ £ ¡ ¢¤
S F RM = E πF BF RM (α1 , ω1F RM ) − E πD BF RM (α1 , ω1F RM )

S −F RM = E [πF (B−F RM (α1 ))] − E [πD (B−F RM (α1 ))]

as the value of operational risk management with and without financial risk management respec-
tively. The value of operational risk management is more robust to a change in ϕ ∈ {a, ρ, σ} with
financial risk management then without if
¯ ¯ ¯ ¯
¯ ∂ S F RM ¯ ¯ ∂ S −F RM ¯
¯ ¯<¯ ¯.
¯ ∂ϕ ¯ ¯ ∂ϕ ¯

To analyze the robustness of the value of operational risk management, we focus on the cases
where operational risk management has a value, i.e. flexible technology is preferred over dedicated
technology with and without financial risk management. Recall from the proof of Proposition 6
that we have cF (cD , H∗ ) < cSF (cD ) and cF (cD , 0) < cSF (cD ) in this setting. Therefore, for any
relevant unit investment cost pair (cF , cD ) we have cF < cSF (cD ). We now analyze each market
condition separately.
Robustness with respect to capital market condition (a). Since cF < cSF (cD ), it follows from
∂ F RM ∂ −F RM ∂ F RM ∂ F RM
(43) that ∂a S ≤ 0 and ∂a S ≤ 0. Therefore, it is sufficient to show ∂a S ≤ ∂a S
to prove the result of lower robustness. It follows from (44) that this condition is equivalent to
∂ ∂
∂a ∆F ≤ ∂a ∆D . We obtain
Z
∂∆F ∂∆D
− = ω0 + α1 ω1 − s(α1 )FF RM − cF K1F − FF − (B̃ − cF K1F − FF ) dRB−F RM (B̃)
∂a ∂a Ω2F
Z
0 1 2
− [ω0 + α1 ω1 − s(α1 )FF RM − cD 1 KD − FD ]χ(B ∈ ΩD ) + (B̃ − cD 10 K1D − FD ) dRB−F RM (B̃).
Ω2D

59
 (1−β)α1
where s(α1 ) = β + α0 , B = ω0 + α1 ω1 − s(α1 )FF RM and χ(.) is the indicator function.
We have the indicator function because a small firm (that always borrows with financial risk
management with flexible technology) need not to borrow with financial risk management with
∂ ∂
dedicated technology. We now show that ∂a ∆F ≤ ∂a ∆D by focusing on two cases.
Case i : (B ∈ Ω2D )We obtain
Z
∂∆F ∂∆D
− = (cF K1F + FF − B̃) dRB−F RM (B̃)
∂a ∂a Ω2F \Ω2D
Z
+ T
(cF K1F + FF − cD 10 K1D − FD ) dRB−F RM (B̃) − (cF K1F + FF − cD 10 K1D − FD ).
Ω2F Ω2D

∂ ∂
Since for B̃ ∈ Ω2F \Ω2D we have B̃ ≥ cD 10 K1D + FD , it follows that ∂a ∆F < ∂a ∆D .
Case ii : (B ∈ Ω01
D ) We obtain
Z
∂∆F ∂∆D
− = (cF K1F + FF − B̃) dRB−F RM (B̃)
∂a ∂a 2 2
ΩF \ΩD
Z
+ T
(cF K1F + FF − cD 10 K1D − FD ) dRB−F RM (B̃) − (cF K1F + FF − ω0 − α1 ω1 + s(α1 )FF RM ).
Ω2F Ω2D

∂ ∂
Since we have ω0 + α1 ω1 − s(α1 )FF RM > cD 10 K1D + FD , it follows that ∂a ∆F < ∂a ∆D . This
concludes the proof for the robustness result with respect to capital market condition.
Robustness with respect to product market conditions (ρ, σ). We only provide the proof
∂ F RM ∂ −F RM
for ρ. From Lemma 6, we have ∂ρ S ≤ 0 and ∂ρ S ≤ 0. Therefore, it is sufficient to
∂ F RM ∂ F RM
show ∂ρ S ≥ ∂ρ S to prove the result of higher robustness for small firms. It follows from
∂
(44) that this condition is equivalent to ∂a ∆F ≥ 0. The result follows from Proposition 7.
Proof of Proposition 10: To demonstrate the ambiguous effect of financial risk management
on expected (stage 0) capacity investment level, it is sufficient to provide examples for each case
of E[10 K∗T∗ (B−F RM (α1 ))] T E[10 K∗T∗ (BF RM (α1 , HT∗ ∗ ))]. We consider FF = FD = 0 which implies
from Proposition 3 that the firm optimally fully hedges with both technologies (Ω4T = ∅). Let
FF RM = 0 such that financial risk management is costless. Without loss of generality we consider
cF < cF which implies from Proposition 4 that T ∗ = F with or without financial risk management.
cT 10 K1
T (1−ab)
Let E be sufficiently large (E ≥ −(b+1)a is sufficient as follows from Lemma 9 in Appendix
B) such that the firm does not borrow up to the credit limit (Ω3F = ∅). With these parameter
restrictions, we obtain
Z Z Z
E[K∗F (B−F RM (α1 ))] = K0F dRB−F RM (B̃) + KF dRB−F RM (B̃) + K1F dRB−F RM (B̃),
Ω0F Ω1F Ω2T

 0 if ω0 + α1 ω1 ∈ Ω0F
 KF

E[K∗F (BF RM (α1 , ω1 ))] = KF if ω0 + α1 ω1 ∈ Ω1F


 1
KF if ω0 + α1 ω1 ∈ Ω2F .

60
We have K0F > K1F , and K0F > KF ≥ K1F for B̃ ∈ Ω1F with equality only holding for the lower
bound of the region Ω1F . For ω0 ∈ Ω0F (and hence ω0 + α1 ω1 ∈ Ω0F ), E[K∗F (B−F RM (α1 ))] =
E[K∗F (BF RM (α1 , ω1 ))]. For ω0 ∈ Ω2F and ω0 +α1 ω1 ∈ Ω0F , E[K∗F (B−F RM (α1 ))] < E[K∗F (BF RM (α1 , ω1 ))].
For ω0 + α1 ω1 ∈ Ω2F (and hence ω0 ∈ Ω2F ), E[K∗F (B−F RM (α1 ))] > E[K∗F (BF RM (α1 , ω1 ))].
If we relax our assumption on E, we obtain
Z Z
∗ 1
E[eF (B−F RM (α1 ))] = [cF KF − B̃] dRB−F RM (B̃) + E dRB−F RM (B̃),
Ω2F Ω3F


 0
 if ω0 + α1 ω1 ∈ Ω01
F
∗
E[eF (BF RM (α1 , ω1 ))] = cF K1F − ω0 − α1 ω1 if ω0 + α1 ω1 ∈ Ω2F



E if ω0 + α1 ω1 ∈ Ω3F .

It follows that for ω0 + α1 ω1 ∈ Ω3F we have E[e∗F (B−F RM (α1 ))] < E[e∗F (BF RM (α1 , ω1 ))] and for
ω0 + α1 ω1 ∈ Ω0F we have E[e∗F (B−F RM (α1 ))] > E[e∗F (BF RM (α1 , ω1 ))].
Proof of Corollary 5: The proof follows from Proposition 8.
Proof of Corollary 6: From Proposition 3, it follows that small firms, as we define in §7,
optimally fully speculates HT∗ = − αω01 . For ω0 = 0, the firm optimally does not engage in financial
risk management. The low value of integration follows from a continuity argument and the bounded
derivative of expected (stage 0) equity value with respect to ω0 . For large firms, financial risk
management does not have any value if ω0 ≥ cT 10 K0T + FT , i.e. the cash level is sufficient to
finance the budget-unconstrained optimal capacity investment level. Low value of financial risk
management at high cash levels follow from similar arguments with small firms. When the firm
uses financial risk management only for hedging purposes, it follows from Proposition 5 that small
firms optimally do not engage in financial risk management. Therefore, the value of integration is
zero for small firms. Large firms tend to use financial risk management for full-hedging purposes.
For ω0 < cT 10 K0T +FT , financial risk management has positive value; hence the value of integration
is higher for large firms than small firms. This concludes the proof.

Proposition 11 If the firm does not engage in financial risk management, there exists a unique
cT 10 K0
technology fixed cost threshold F −F
T
RM
< T
−(b+1)(1−γT ) for technology T ∈ {D, F } such that when
FT < F −F
T
RM
, investing in technology T without financial risk management is more profitable than
not investing in technology.
If the firm engages in financial risk management, only one of the following cases holds, depending
on the level of the fixed cost FF RM :
cT 10 K0
 
T− (1−β)α1
β+ FF RM
i) There exists a unique technology fixed cost threshold F FT RM ≤ −(b+1)
1−γT
α0
for
technology T ∈ {D, F } such that when FT < F FT RM , investing in technology T is more

61
 profitable than not investing in technology; this case occurs at sufficiently low levels of FF RM .

ii) Not investing in technology is more profitable for FT ≥ 0.

Proof of Proposition 11: We first prove the first part of the proposition. From Lemma 6
in the proof of Proposition 4 (using H = 0 and FF RM = 0), E [πT (B−F RM (α1 ), FT )] is strictly
.
decreasing in FT . We define LT (B̃) = πT (B̃) − (B̃ + P ), the difference between the equity values of
investing in technology T and not investing in technology at each state B̃. It is easy to verify that
. £ ¤
for FT0 = 0, πT (B̃) > B̃ + P for B̃ ≥ 0. It follows that E πT (FT0 , B−F RM (α1 )) > ω0 + α1 ω1 + P .
cT 10 K0 £ ¤
For FT1 > −(b+1)(1−γ T
T)
we have LT (B̃) < 0 for B̃ ≥ 0. It follows that E πT (FT1 , B−F RM (α1 )) <
ω0 + α1 ω1 + P . Since E [πT (FT , B−F RM (α1 ))] is strictly decreasing in FT , there exists a unique
cT 10 K0
F −F
T
RM
< −(b+1)(1−γT ) .
T

The second part of the proposition follows from a similar argument. We obtain E [πT (BF RM (α1 , HT∗ ))] ≤
³ ´
cT 10 K0
ω0 + α1 ω1 − β + (1−β)α
α0
1
FF RM − (1 − γT )FT + −(b+1) T
+ P , where the latter is the expected
(stage 0) equity value with budget-unconstrained optimal capacity investment. It follows that for
cT 10 K0
 
T− (1−β)α1
β+ FF RM
FT > FT1 = −(b+1)
1−γT
α0
, not investing in technology is more profitable. Two cases
may arise with respect to the level of FF RM . When FF RM is sufficiently low, for FT0 = 0 we have
£ ¤
E πT (FT0 , BF RM (α1 , HT∗ )) > ω0 +α1 ω1 +P . In this case (case i), a unique F FT RM < FT1 exists since
E [πT (FT , BF RM (α1 , HT∗ ))] is strictly decreasing in FT . For a sufficiently high level of FF RM and
appropriate allocation scheme β (that makes such a FF RM feasible), not investing in technology
is more profitable for FT = 0. In this case (case ii), F FT RM does not exist and not investing in
technology is more profitable for FT ≥ 0.

Proposition 12 Only one of the following cases holds for technology T :

i) There exists a unique financial risk management fixed cost threshold F TF RM such that when
FF RM < F TF RM , it is more profitable to engage in financial risk management than not;

ii) For any feasible FF RM , engaging in financial risk management is more profitable than not.

Proof of Proposition 12: The proof follows from showing that E [πT (BF RM (α1 , HT∗ ))] strictly
decreases in FF RM . From Lemma 2, we can interchange the derivative and expectation operators

62
and using the Leibniz’ rule we obtain
· ¸ Z ∞
∂πT (BF RM (α1 , H)) 1−β
E = (−β − x) rα1 (x) dx (46)
∂FF RM max(α0T ,0) α0
Z max(α0 ,0) µ ¶1
T MT (1 + 1/b) U (x) b 1−β
+ (−β − x) rα1 (x) dx
max(α1T ,0) cT cT α0
Z max(α1 ,0)
T 1−β
+ (−β − x)(1 + a) rα1 (x) dx
max(α2T ,0,αBT)
α0
Z max(α2 ,0,αB ) µ ¶1
T T M (1 + 1/b) U (x) + E b 1−β
T
+ (−β − x) rα1 (x) dx
max(0,αBT)
cT cT α0
Z max(0,αB )
T 1−β
+ (−β − x) rα1 (x) dx
0 α0

for any feasible H, where U (x) = ω0F RM + x(ω1F RM − H) + α1 H − FT . Since all terms are neg-
ative, it follows that E [πT (BF RM (α1 , HT∗ ))] is strictly decreasing in FF RM . For FF RM = 0, we
have E [πT (BF RM (α1 , HT∗ ))] ≥ E [πT (B−F RM (α1 ))] from the optimality of HT∗ . The existence of
³ ´
F TF RM ≤ min ωβ0 , α1−β
0 ω1
depends on the allocation scheme β. If β is such that a sufficiently large
level of FF RM is feasible, then since E [πT (BF RM (α1 , HT∗ ))] is strictly decreasing in FF RM , there
exists a unique F TF RM (case i). Otherwise, since E [πT (BF RM (α1 , HT∗ ))] is preferred for FF RM = 0,
case ii holds.
ω0
We show that ∃ β such that case i holds. Let β = ω0 +α 0 ω1
. It follows that the condition FF RM ≤
³ ´
ω 0 α0 ω 1
min β , 1−β is equivalent to FF RM ≤ ω0 +α0 ω1 . We obtain limFF RM →ω0 +α0 ω1 BF RM (α1 , H) = 0;
therefore E [πT (BF RM (α1 , HT∗ ))]|FF RM →ω0 +α0 ω1 < E [πT (B−F RM (α1 ))]. It follows that a unique
F TF RM exists.

Proposition 13 For technology T ∈ {D, F } there exists a unique variable cost threshold cT (c−T , HT∗ , 0)
such that investing in technology T with financial risk management is more profitable than investing
in the other technology (−T ) without financial risk management.

Proof of Proposition 13: The proof follows as in Proposition 4, and is omitted.

63
B bT
Appendix B. Characterization of B

Recall from Proposition 2 that BbT is the budget threshold below which the firm does not borrow
or invest. From the proof of Proposition 2, for FT ≥ E, B bT > FT − E is the unique solution to
.
bT ) = 0 where GT (B̃) = ΨT (B̃) − (B̃ − (1 − γT )FT + P ), the difference between the equity
GT (B
values in (29) and not borrowing and not investing in capacity. For FT < E, B bT , if it exists on
bT =.
[0, ∞), is unique. For notational convenience, we let B 0 if the two curves do not intersect on
the domain of GT (.) for FT < E. From (29) for B̃ ≥ FT we obtain limKT →0+ ∇KT ΨT → ∞. It
follows that the firm always optimally invests in capacity if internal budget B̃ is sufficient to cover
the fixed cost of the technology. We conclude that FT − E < B bT < FT . Since ΨT (B̃) can take four
different forms we have four different cases to analyze.
Case 1: cT 10 K0T + FT ≤ B̃
bT .
From (33), GT (B̃) > 0 in this range, so it is not possible to have cT 10 K0T + FT ≤ B
Case 2: cT 10 K1T + FT ≤ B̃ < cT 10 K0T + FT
µ ¶− 1
(B̃ − FT ) cT b
GT (B̃) = MT + γT FT + P − (B̃ − (1 − γT )FT + P )
cT B̃ − FT
µ ¶− 1
(B̃ − FT ) 1 b −1
≥ MT 0 0 + FT − B̃ = (B̃ − FT ) > 0.
cT 1 KT b+1

bT < cT 10 K0 + FT .
Therefore, it is not possible to have cT 10 K1T + FT ≤ B T
Case 3: cT 10 K1T + FT − E ≤ B̃ < cT 10 K1T + FT

bT ) cT 10 K1T (1 + a)
bT − FT )(1 + a) + bT − (1 − γT )FT + P ) = 0
GT (B = (B + γT F T + P − ( B
−(b + 1)
0 1
bT = FT − cT 1 KT (1 + a) .
⇒ B
−(b + 1)a

bT to be feasible in Case 3, B
For B bT ≥ 0 and BbT ≥ cT 10 K1 + FT − E should hold. Therefore, if
T
0 1
c 1 KT (1+a) 0 1
c 1 KT (1−ab)
FT ≥ T and E ≥ T bT is feasible. Otherwise, it is not possible to have
then B
−(b+1)a −(b+1)a
cT 10 K1T bT <
+ FT − E ≤ B cT 10 K1T + FT .
Case 4: cT 10 K1T + FT − E > B̃
In this case, we can derive a sufficient condition for non-existence of intersection. We obtain
µ ¶− 1
(E + B̃ − FT ) cT b
GT (B̃) = −E(1 + a) + MT + γT FT + P − (B̃ − (1 − γT )FT + P )
cT E + B̃ − FT
µ ¶− 1
(E + B̃ − FT ) 1 b
≥ −E(1 + a) + MT + FT − B̃
cT 10 K1T
E(1 + a) 1 − ab
≥ + (B̃ − FT ).
−(b + 1) −(b + 1)

64
Therefore if FT < E(1+a) 0 1 b
1−ab , then GT (B̃) > 0 and it is not possible to have cT 1 KT + FT − E > BT .
Otherwise, BbT is a solution of a non-integer polynomial of degree b and it is not possible to find
b+1
closed-form expression in the whole range of parameters. The following lemma summarizes the
bT for a subset of parameter levels.
analysis and provides a closed-form expression for B
c 10 K1 (1−ab)
Lemma 9 Let E be such that E ≥ T −(b+1)a
T
.
1
cT KT (1+a) bT = 0 and Ω = ∅.
If FT ≤ −(b+1)a then B 34
T
0 1
cT 1 KT (1+a) cT 10 K1
b T (1+a)
If FT > −(b+1)a then BT = FT − −(b+1)a and Ω3T = ∅.

We also provide the following lemma which we will occasionally use in the comparative statics
analysis throughout the paper.

bT is increasing in cT , FT , a and decreasing in E.

Lemma 10 The budget threshold B

Proof We only provide the proof for the result related to a. The other results can be shown in a
similar fashion. Let B bT (ai ), i = 0, 1 define the threshold levels for an arbitrary a0 < a1 . We want
to show that B bT (a0 ) ≤ BbT (a1 ). Notice that not only the functional form of GT (B̃) in any region
but also the budget levels defining the regions in (29) depend on a. We obtain


 0 if cT 10 K0T + FT ≤ B̃



∂GT (B̃)  0 if cT 10 K1T + FT ≤ B̃ < cT 10 K0T + FT
=
∂a  B̃ − cT 10 K1 − FT
 if cT 10 K1T + FT − E ≤ B̃ < cT 10 K1T + FT

 T


−E if FT − E < B̃ < cT 10 K1T + FT − E.

∂
at the points where GT (B̃) is differentiable in a. It follows that ∂a GT (B̃) ≤ 0 for any B̃ where
the function is differentiable. Since GT (B̃) is a continuous function of B̃ for any a, we conclude
that GT (B̃) is decreasing in a. This implies GT (B̃, a0 ) ≥ GT (B̃, a1 ) for B̃ > FT − E. At this
point, two different cases may arise regarding the definition of B bT (a0 ). If B
bT (a0 ) is the solution of
bT (a0 ), a1 ) ≤ GT (B
GT (B̃, a0 ) = 0, then we have GT (B bT (a0 ), a0 ) = 0. Since GT (B̃) is increasing in
B̃ from (30), it follows that B bT (a1 ) ≥ B bT (a0 ). If B
bT (a0 ) = 0 because GT (B̃, a0 ) > 0 for B̃ ≥ 0,
then from (30) either we have B cT (a1 ) = 0, (GT (B̃, a1 ) > 0 for B̃ ≥ 0) or B
bT (a1 ) is a solution to
GT (B̃, a1 ) = 0. In either case, we have B bT (a1 ) ≥ B bT (a0 ).

65
C Appendix C. Proofs of Supporting Lemmas

Proof of Lemma 1: From Appendix B, we calculate



 0 if cT 10 K0T + FT ≤ B̃



 ³ ´1
 MT
∂GT (B̃)  − cT (1 + b )
1 B̃−FT b
cT +1 if cT 10 K1T + FT ≤ B̃ < cT 10 K0T + FT
=
∂FT 
 −a if cT 10 K1T + FT − E ≤ B̃ < cT 10 K1T + FT



 ³ ´ 1

 − MT (1 + 1 ) E+B̃−FT b + 1 if FT − E < B̃ < cT 10 K1 + FT − E.
cT b cT T

From (31), (32) and the continuity of GT (B̃), it follows that GT (B̃) strictly decreases in FT for
B̃ < cT 10 K0 + FT . Recall from Proposition 2 (or Appendix B) that either B bT is a solution to
T
bT = 0 (if GT (B̃) > 0 for B̃ ≥ 0).
GT (B̃) = 0 or B
We first prove the necessity of the second argument. Let F T be the fixed cost that satisfies
bT (F T ), F T ) = 0 with B
GT (B bT (F T ) = 0. In other words, F T is the fixed cost of technology T that
makes the two equity values intersect at B̃ = 0. From Appendix B, it follows that for FT = 0,
GT (B̃) > 0 for B̃ ≥ 0. For FT ≥ E, we have limB̃→(FT −E)+ GT (B̃) < 0, and two curves intersect
bT > FT − E. Since GT (B̃) is strictly decreasing in FT , such an F T < E always exists. Let
at B
FT0 ≤ F T be an arbitrary fixed cost. We have GT (B bT (F T ), F 0 ) < GT (B bT (F T ), F T ) = 0 since
T
bT < cT 10 K0 + FT (follows from Appendix B) and GT strictly decreases in FT . From (30) we have
B T
bT (F 0 ) > B
GT (B̃) is strictly increasing in B̃ so it follows that B bT (F T ) = 0.
T
We now prove the necessity of the first argument. Let FT1 < F T be an arbitrary fixed cost.
bT (F T ) = 0 and GT (B̃) strictly decreases in FT , we have GT (B̃, F 1 ) > 0 for B̃ ≥ 0. This
Since B T
implies that BbT (FT ) = 0 for FT < F T . The uniqueness of F T follows from the fact that GT (B̃) is
bT .
strictly decreasing in FT and the uniqueness of B
The proof for sufficiency follows easily using a contrapositive argument.
Proof of Lemma 2: The expectation and differentiation operators can be interchanged if the
function under expectation is integrable and satisfies the Lipschitz condition of order one (Glasser-
man 1994, p.245). The function πT (α̃1 ) satisfies the Lipschitz condition of order one if
0 00
|πT (α̃1 ) − πT (α̃1 )| 0 00
0 00 ≤ YπT ∀ (α̃1 , α̃1 ) > 0 for some YπT with E[YπT ] < ∞. (47)
|α̃1 − α̃1 |
¯ ¯ ³ ´³ ´
¯ T¯
Clearly, condition (47) is satisfied if ¯ ∂π
∂ α̃1 ¯ is bounded. Note that ∂
π
∂ α̃1 T = ∂
π
∂ B̃ T
∂
∂ α̃1 B̃ =
³ ´
∂
π (ω1 − HT ). From Corollary 1, we know that πT is differentiable in α̃1 everywhere except
∂ B̃ T
at αTB as defined in (38). If B̃ ∈ Ω1T we have

∂πT MT 1 ¡ ¢1
≤ (1 + ) 10 K1T b ≤ (1 + a),
∂ B̃ cT b

66
 cT ≥ 0 and ET > FT (from (11)) we have
and for B̃ ∈ Ω3T since B
µ ¶1
∂πT MT 1 E − FT b
≤ (1 + ) ≤ YT
∂ B̃ cT b cT
¯ ¯
¯ T¯
where 1 + aT < YT < ∞. It follows that ¯ ∂π B
∂ α̃1 ¯ ≤ YT (ω1 − HT ) < ∞ for α1 ≥ 0 except αT . Since
πT is continuous in α1 and the first derivative is bounded at the differentiable points of πT , the
non-differentiability at αTB does not violate (47). Since πT (α̃1 ) is integrable, the interchange of the
derivative and expectation is justified.
Proof of Lemma 3: From Lemma 2, we can interchange the derivative and the expectation
operators and using the Leibniz’ rule we obtain
Z
∂E[πT (BF RM (α1 , H))]
= −10 K0T dRBF RM (H) (B̃) (48)
∂cT 0
ΩT
Z Ã !1+ 1
b
1 MT B̃ − FT
+ −(1 + ) dRBF RM (H) (B̃)
Ω1T b cT cT
Z
+ −10 K1T (1 + a) dRBF RM (H) (B̃)
Ω2T
Z Ã !1+ 1
b
1 MT E + B̃ − FT
+ −(1 + ) dRBF RM (H) (B̃).
Ω3T b cT cT
∂ F RM and
It follows that ∂cT E[πT (BF RM (α1 , H))] ≤ 0 with equality holding only for H = ω1
ω0F RM + α1 ω1F RM ∈ Ω4T (i.e. Ω0123 T = ∅). From Proposition 3, we know that in this case
ω0F RM
∗
HT = − α1 , so we can ignore H = ω1 RM . In other words, in the relevant set of BF RM (α1 , H)
F

we have ∂c∂T E[πT (BF RM (α1 , H))] < 0.

Proof of Lemma 4:
Case i): The proof follows from Lemma 3 of Chod et al. (2006) by substituting τ = 1 and noting
that ρ and σ in that paper correspond to parameters of the underlying bivariate normal distribution
(ln ξ) of ξ. In our paper, ρ and σ are the parameters of ξ in the covariance matrix Σ.
0
Case ii): We only prove the more general case where both of the marginal distributions of ξ are
pairwise stochastically more variable than the marginal distributions of ξ. The proof for the case
0 0
where one of the marginals is identical for ξi and ξi is a special case of this proof. For ξi ≥ 0, ξi ≥ 0
0 0
and ξ i = ξ 0 i it follows from Ross (1983, p.271) that ξi ºv ξi if and only if E[h(ξi )] ≤ E[h(ξi )] for all
convex functions h(.). With independent marginal distributions of ξ we have
·³ ´− 1 ¸ Z ∞ Z ∞ ³ ´− 1 Z ∞
E ξ1−b + ξ2−b x−b −b
b b
= 1 + x2 f (x )f
1 1 2 2(x )dx1 dx2 = g(x1 ; x2 )f1 (x1 )dx1
0 0 0
R∞ ³ ´− 1
where fi (.) is the marginal distribution of ξi and g(k; x2 ) = k −b + x−b b
f2 (x2 )dx2 for k ≥ 0.
0 2
³ ´− 1
To conclude the proof, we need to show that g(k; x2 ) is convex in k and k −b + x−b
b
2 is convex

67
 0
in x2 . To prove both of the desired convexity results, it is sufficient to show that g (k, x2 ) is convex
in k. We obtain
∂2g
0 ³ ´− 1 −1 x−b
= (−b − 1) k −b + x−b
b −b−2 ³ 2
2 k ´ ≥0
∂k k + x−b
−b
2

for k ≥ 0 and x2 ≥ 0. This concludes the proof.

Case iii): Follows from (40) in the proof of Proposition 4.
Proof of Lemma 5:
Case i): The proof follows from Lemma 4 of Chod et al. (2006) by substituting τ = 1 and noting
that ρ and σ in that paper correspond to parameters of the underlying bivariate normal distribution
(ln ξ) of ξ. In our paper, ρ and σ are the parameters of ξ in the covariance matrix Σ.
0
Case ii): The proof of this case is adapted from Corbett and Rajaram (2005). If ξ ºc ξ, it
0 0
follows from Muller and Scarsini (2000, p.110) that ξ ºsm ξ (ξ dominates ξ in the sense of
supermodular order). From the definition of supermodular stochastic ordering, it is sufficient to
³ ´− 1
show that g(ξ1 , ξ2 ) = − ξ1−b + ξ2−b
b
is supermodular. From Muller and Scarcini (2003), it
∂2
follows that g is supermodular if and only if all mixed derivatives are non-negative, i.e. ∂ξ1 ∂ξ2 g ≥0
for ξ ≥ 0. We obtain
∂2g ³ ´− 1 −2
= (−b − 1) ξ1−b + ξ2−b
b
(ξ1 ξ2 )−b−2 ≥ 0.
∂ξ1 ∂ξ2
This concludes the proof.
Case iii): Follows from (40) in the proof of Proposition 4.
Proof of Lemma 6:
Case i): As in Lemma 3, we obtain
Z
∂E[πT (BF RM (α1 , H))]
= −(1 − γT ) dRBF RM (H) (B̃) (49)
∂FT Ω0T
 Ã !1 
Z b
1 M T B̃ − F T
+ − (1 + ) − γT  dRBF RM (H) (B̃)
ΩT1 b cT cT
Z
+ −(1 − γT + a) dRBF RM (H) (B̃)
Ω2T
 Ã !1 
Z b
1 MT E + B̃ − FT
+ − (1 + ) − γT  dRBF RM (H) (B̃)
Ω3T b cT cT
Z
+ −(1 − γT ) dRBF RM (H) (B̃).
Ω4T

Since γT < 1 by definition, it follows from (31) and (32) that the second and the fourth terms are
∂ ∂
negative. This implies that ∂FT E[πT (BF RM (α1 , H))] < 0. We have ∂γT E[πT (BF RM (α1 , H))] = FT

68
and it follows that the expected (stage 0) equity value is strictly increasing in the salvage rate for
FT > 0.
Case ii): We obtain
 ¡ ¢Ã !1 
Z
∂E[πT (BF RM (α1 , H))] MT 1 + 1b E + B̃ − FT b 
= −(1 + a) + dRBF RM (H) (B̃).
∂E Ω3T cT cT

∂ F RM + α H ≥
It follows that ∂E E[πT (BF RM (α1 , H))] ≥ 0 with equality holding for H such that ω0 1
0 1
cT 1 KT + FT − E; or H = ω1 F RM F
and ω0 RM F
+ α1 ω1 RM b
< BT . From Proposition 3 we know that
ω F RM
in the latter case HT∗ = − 0α1 and we can ignore this case in the relevant set of financial risk
management levels.
Case iii): We obtain
Z ³ ´ Z
∂E[πT (BF RM (α1 , H))] 0 1
= B̃ − cT 1 KT − FT dRBF RM (H) (B̃) − ET dRBF RM (H) (B̃).
∂a Ω2T Ω3T

∂ F RM + α H ≥ c 10 K1 + F ;
It follows that
∂a E[πT (BF RM (α1 , H))] ≤ 0 with equality holding for ω0 1 T T T

or H = ω1F RM and ω0 F RM + α1 ω1F RM b

< BT . From Proposition 3 we know that in the latter case
ω F RM
HT∗ = − 0α1 and we can ignore this case in the relevant set of financial risk management levels.
Case iv): The expected (stage 0) equity value with dedicated technology is independent of σ.
Therefore, we focus only on flexible technology. We obtain
Z µ ¶
∂E[πF (BF RM (α1 , H))] ∂MF 1 −b−1
= cF MF (1 + ) dRBF RM (H) (B̃) (50)
∂σ Ω0F ∂σ b
Z Ã !1+ 1
b
∂MF B̃ − FF
+ dRBF RM (H) (B̃)
Ω1F ∂σ cF
Z Ã !−b−1
∂MF cF MF (1 + 1b )
+ dRBF RM (H) (B̃)
Ω2F ∂σ 1+a
Z Ã !1+ 1
b
∂MF E + B̃ − FF
+ dRBF RM (H) (B̃).
Ω3F ∂σ cF

∂
From Lemma 4, we have ∂σ MF ≥ 0 with respect to our definitions of demand variability. It follows
∂
that ∂σ E[πT (BF RM (α1 , H))] ≥ 0.
Case v): The proof of the comparative static result with respect to ρ is similar to σ and is omitted.

¯
¯
Proof of Lemma 7: It is easy to verify that we have cF KjF ¯ = cD 10 KjD for j = 0, 1.
cS
F (cD )
bF = B
Since FF = FD and γF = γD from (29) we have ΨF (B̃) = ΨD (B̃) which implies B bD . It
follows that the regions in (1) overlap, i.e. ΩiF ≡ ΩiD for i = 0, .., 4. Since the budget distribution

69
BF RM (H) is independent of cost parameters, the expected (stage 0) equity values are the same at
the threshold level. Moreover, from (35), it follows that HF∗ (cSF (cD )) = HD
∗ (c ) because both of
D

them are solutions to the same optimization problem.

Proof of Lemma 8: Recall from the proof of Proposition 7 we have
£ ¡ F RM )
¢¤
. ∂∆ T ∂E π T BF RM (α 1 , ω1 ∂E [πT (B−F RM (α1 ))]
Υϕ = = −
∂ϕ ∂ϕ ∂ϕ

For ϕ = cT (ϕ = FT ), we calculate the derivative from Lemma 3 (Lemma 6) by letting Ω34

T = ∅
(because of our assumptions on FT and E).
In (48) of Lemma 6, for B̃ ∈ Ω1T we have
 Ã !1 
b
1 M T B̃ − FT
1 − γT < (1 + ) − γT  < 1 + a − γT .
b cT cT

For ω0F RM + α1 ω1F RM ∈ Ω0T , it follows that

∂E[πT (BF RM (α1 , ω1F RM ))] ∂E[πT (B−F RM (α1 ))]

= −(1 − γT ) ≥ ,
∂FT ∂FT

and we obtain ΥFT ≥ 0 where the equality holds for ω0 > cT 10 K0T + FT .
For ω0F RM + α1 ω1F RM ∈ Ω2T ,

∂E[πT (BF RM (α1 , ω1F RM ))] ∂E[πT (B−F RM (α1 ))]

= −(1 + a − γT ) < ,
∂FT ∂FT

and we obtain ΥFT < 0. This concludes the proof for part (i).
Similarly, in (49) of Lemma 6, for B̃ ∈ Ω1T we have
¯ à !1+ 1 ¯¯
¯
¯ 0 0¯ ¯ 1 MT B̃ − FT b
¯ ¯ 0 1 ¯
¯1 KT ¯ > ¯(1 + ) ¯ > ¯1 KT (1 + a)¯ .
¯ b cT cT ¯
¯ ¯

For ω0F RM + α1 ω1F RM ∈ Ω0T , it follows that

∂E[πT (BF RM (α1 , ω1F RM ))] ∂E[πT (B−F RM (α1 ))]

= −10 K0T ≤
∂cT ∂cT

and we obtain ΥcT ≤ 0 where the equality holds for ω0 > cT 10 K0T + FT .
For ω0F RM + α1 ω1F RM ∈ Ω2T ,

∂E[πT (BF RM (α1 , ω1F RM ))] ∂E[πT (B−F RM (α1 ))]

= −10 K1T (1 + a) >
∂cT ∂cT

and we obtain ΥcT > 0. This concludes the proof for part (ii).

70
References

Chod, J., N. Rudi, J.A. Van Mieghem. 2006. Mix, time and volume flexibility: Valuation and
corporate diversification. Working Paper, Kellogg School of Management, Northwestern Uni-
versity.

Corbett, C.J., K. Rajaram. 2005. A generalization of the inventory pooling effect to non-normal
dependent demand. Working Paper, Anderson Graduate School of Management, University of
California, Los Angeles.

Glasserman, P. 1994. Perturbation analysis of production networks,” D. Yao, eds. Stochastic

modelling and analysis of manufacturing systems, Sringer-Verlag, NY.

Hardy, G., J.E. Littlewood, G. Polya. 1988. Inequalities. Cambridge Press, NY.

Muller, A., M. Scarsini. 2000. Some remarks on the supermodular order. Journal of Multivariate
Analysis 73 107–119.

Muller, A., M. Scarsini. 2003. Archimedean Copulae and Positive Dependence. Working Paper,
ICER.

Ross, S.M. 1983. Stochastic Processes. Wiley Series in Probability and Mathematical Statistics.

71