North American Journal of Economics and Finance

18 (2007) 59–75

Attitude towards risk, uncertainty,

and fixed investment
Hong Bo a , Elmer Sterken b,∗
a Department of Financial & Management Studies, SOAS, University of London, London, United Kingdom
b Department of Economics, University of Groningen, P.O. Box 800, 9700 AV Groningen, The Netherlands
Received 4 October 2005; received in revised form 17 August 2006; accepted 8 September 2006
Available online 18 October 2006

Abstract
We explore the relevance of the risk attitude of managers to the investment-uncertainty relation. Higher
moments of the distribution of net profits are used to measure the risk premium of the firm, from which
we derive a proxy for the risk aversion of managers. Using an unbalanced panel of Dutch listed firms, we
find that in general a low degree of risk aversion coincides with a positive impact of demand uncertainty
on investment. More specifically, we find that risk-averse firms respond to demand uncertainty by cutting
investment, while the investment undertaken by risk-taking firms responds to demand uncertainty positively.
© 2006 Elsevier Inc. All rights reserved.

JEL classification: E2; G3

Keywords: Risk attitude; Uncertainty; Fixed investment

1. Introduction

The effect of uncertainty on investment is a prominent topic in investment studies. The literature
has identified several channels through which uncertainty can affect investment: (1) the risk
attitude of decision-makers (Nakamura, 1999; Nickell, 1978; Zeira, 1990); (2) the shape of the
marginal product of capital (Abel, 1983; Caballero, 1991; Hartman, 1972); (3) substitutability
of production factors (Hartman, 1976; Leahy & Whited, 1996); (4) irreversibility and the option
value of waiting (Dixit & Pindyck, 1994); and (5) financial constraints (Minton & Schrand, 1999).
Much effort has focused on testing the impact of channels (2)–(5) on investment. However, the
results are not conclusive.

∗ Corresponding author. Tel.: +31 50 3633723; fax: +31 50 3637337.

E-mail address: e.sterken@rug.nl (E. Sterken).

1062-9408/$ – see front matter © 2006 Elsevier Inc. All rights reserved.
doi:10.1016/j.najef.2006.09.001
60 H. Bo, E. Sterken / North American Journal of Economics and Finance 18 (2007) 59–75

First, it is not clear whether uncertainty discourages or encourages investment. Different

theoretical approaches propose different conclusions. Second, although most studies find the
negative effect of uncertainty on investment, it is not clear through which channel uncertainty
affects investment.1 From an empirical point of view, it is hard to identify each channel sep-
arately. Almost all empirical research on the effect of uncertainty on investment ignores a
fundamental factor in the objective function of the firm: the risk attitude of managers. Most
empirical studies assume that managers of the firm are risk neutral when they make investment
decisions.
In reality, however, managers are more likely to be risk-averse. When agency costs between
managers and shareholders are absent, managers represent shareholders and the objective func-
tion of managers is to maximize shareholders’ value. Nickell (1978) argues that only under the
assumptions of perfect capital markets and certainty, can one ignore the preferences of the firm’s
shareholders in considering the firm’s decisions. Therefore, under uncertainty shareholders’ con-
sumption behavior and their preference structure become important. In this case, the objective
function of the firm should be derived from the intertemporal optimization problem of the owners
(consumers) of the firm. When agency costs of management are high, the firm’s objective func-
tion is to maximize the utility of managers. Although protected by limited liability laws in many
countries, managers are trying to avoid bankruptcy since they do not want to lose their private
benefits of control. Although it facilitates the analysis, the assumption of risk-neutrality is not
realistic. This suggests that we should treat firms as agents who are maximizing the expected
discounted utility of profits rather than maximizing discounted profits.
The consequences of ignoring non-neutral risk attitudes of managers in investment studies can
be severe. First, attitude towards risk, reflected in the utility function of the investment decision
makers, is a prominent determinant of the investment-uncertainty relation. Second, neglecting
attitude towards risk messes up the impact of other important factors predicting the effect of
uncertainty on investment. For example, many authors, after assuming risk neutrality and obtaining
a negative sign of uncertainty in investment equations, conclude that irreversibility induces the
negative effect of uncertainty on investment.2 However, since a higher degree of risk aversion is
likely to generate a negative impact of uncertainty on investment, one could argue that it might
be the risk aversion attitudes of managers of the firm rather than irreversibility that induces the
negative effect. Defining separate empirical indicators of risk aversion and irreversibility would
be a requisite to identify both channels. Gollier (2000) finds that the degree of risk aversion of
agents increases with financial constraints. According to Gollier capital market imperfections may
discourage investment not only because of limited access to external finance but more directly
because they discourage individuals’ willingness to bear risk. Therefore, if the impact of risk
attitude could be isolated and controlled, the estimated impact of other factors, like irreversibility,
could be purified and made more informative.
The debates and mixed evidence in the literature on the investment-uncertainty relationship
suggest that we should go back to the risk attitude of managers to investigate firm investment
behavior under uncertainty. Caballero (1991) argues that “the relationship between changes in
price uncertainty and capital investment under risk neutrality is not robust. . .it is very likely that

1 For a review of the empirical research on the investment-uncertainty relationship, see Chapter 6, Lensink, Bo, and

Sterken (2001).
2 Among 20 empirical studies surveyed by Lensink et al. (2001), 17 papers find a negative effect of uncertainty on

investment, 11 of these 17 papers explain the negative effect of uncertainty by explicitly referring to the irreversibility
hypothesis.
 H. Bo, E. Sterken / North American Journal of Economics and Finance 18 (2007) 59–75 61

it will be necessary to turn back to risk aversion, incomplete markets and lack of diversification
to obtain a sturdier negative relationship between investment and uncertainty.”
Although theories predict that the risk attitude of managers influences the optimal output and
investment decisions of the firm under uncertainty (see the related literature in Section 2 of this
paper), the evidence in the literature is scant. One important obstacle is that empirical proxies for
the risk attitude of managers are not directly available. The contribution of this paper is to derive
an empirical proxy for the risk attitude of managers based on observed net returns data, which
enables us to test how risk attitude changes the effect of uncertainty on investment. To the best of
our knowledge, this paper is the first attempt to empirically analyze the impact of risk attitudes
of managers on the investment-uncertainty relationship.
Two issues remain largely unresolved. First, as in all other studies on the investment-uncertainty
relationship, it is hard to quantify the exact impact of any type of uncertainty on investment. The-
oretical and empirical studies focus on the sign of the relation and not on the magnitude of the
impact. Second, it is hard to establish whether the risk attitude assumption complements or substi-
tutes for other determinants of the investment-uncertainty relationship, because all explanations
interact. We show that abandoning the assumption of risk neutrality gives another channel for
the impact of uncertainty on investment. It remains an empirical quest, though, whether or not
assuming risk neutrality interferes with other channels. We do not solve this issue, but merely
explore the new role of the managerial risk attitude and leave the assessment of its relative impact
to future studies.
The rest of the paper is organized as follows. Section 2 briefly reviews the related literature
from which we draw an analytical model justifying why risk attitude should be important for
the investment decision of the firm under demand uncertainty. In Section 3, we describe the
construction of the empirical proxy for the risk attitude of managers. Here we first follow Fisher
and Hall (1969) to construct the risk premium of the firm. Then we derive the measure of risk
aversion for the managers of the firm based on the relation between the risk premium and the
measure of risk aversion predicted by utility theory (e.g., Arrow, 1971). Section 4 describes the
data and the measurement of demand uncertainty. Section 5 presents the results obtained by the
system Generalized Method of Moments (GMM) estimator. Section 6 summarizes and concludes
the paper.

2. A theoretical model of risk attitude and the investment-uncertainty relationship

2.1. Related literature

Risk attitude has been theoretically identified as one important factor that affects firm behavior.
Sandmo (1971, p. 65) explicitly criticizes the assumption that the firm is risk-neutral. He proves
that risk aversion reduces the optimal production decision of a competitive firm under output price
uncertainty. Leland (1972) extends the Sandmo (1971) perfectly competitive model to imperfectly
competitive quantity- and price-setting firms. He shows that risk aversion leads to lower optimal
output for the quantity-setting firm, and a lower quantity and price when the firm sets both before
the uncertain demand is revealed, although the impact of risk aversion on a price-setting firm is
ambiguous. Hartman (1976) theoretically proves that the effect of output-price uncertainty on cap-
ital inputs depends on the substitutability between capital and labor. Moreover, he claims that risk
aversion influences the way the firm adjusts capital input in response to output-price uncertainty,
although the direction of the impact of risk aversion is ambiguous. Nickell (1978, p. 75) provides
a model in which he compares the output decision of a competitive firm under demand uncertainty
62 H. Bo, E. Sterken / North American Journal of Economics and Finance 18 (2007) 59–75

between risk neutrality and risk aversion. He concludes that when demand levels are uncertain,
risk-averse behavior is bound to lower capacity levels and the optimal capacity level is a declining
function of the degree of risk aversion. Zeira (1990) analyzes the effect of wage rate uncertainty
on the capital share using a general equilibrium framework. The degree of relative risk aversion is
found to change the impact of uncertainty on investment. Nakamura (1999) explicitly models the
risk-attitude parameter in the optimal investment rule. He derives an optimal investment rule that
is a function of output-price uncertainty, the degree of relative risk aversion, and an elasticity of
output to labor in the Cobb-Douglas production function. The effect of uncertainty on investment
is shown to be positive, zero, and negative, depending on the trade-off between the degree
of relative risk aversion and the elasticity of output to labor in the Cobb-Douglas production
function.

2.2. The model

We derive a model of risk attitude and firm investment under demand uncertainty. The model is
based on the framework of Sandmo (1970, 1971). We adapt the framework to the fixed investment
decision of the firm. The aim of deriving the model is to show why and how risk attitude affects
the impact of demand uncertainty on investment.
We start with a standard investment problem with no uncertainty. At time t the firm chooses the
amount of investment. Assuming a one-year time-to-build lag of investment, the year t investment,
It , will be used in production in the year (t + 1). Assuming that labor input is completely flexible,
the firm adjusts its capital inputs in order to maximize the expected utility of profit in the year
(t + 1). The profit function in the year (t + 1) is:

πt+1 (It ) = F (Kt+1 , Lt+1 ) − wt+1 Lt+1 − G(It , Kt ) − It , (1)

where πt+1 is net operating profits for the period (t + 1), F(Kt+1 , Lt+1 ) is the revenue function, Kt+1 ,
Lt+1 are the beginning-of-period capital stock and the labor input, It is the gross investment of the
firm at time t, wt+1 is the nominal wage rate in the period (t + 1), and G(It , Kt ) is the internal convex
cost function of adjusting the capital stock. Notice that in the profit function (1) we normalize
the prices of output and capital goods to the unit for simplicity, although relaxing this assumption
does not affect the conclusion of the model. Since we model the short-run investment decision,
it is assumed that there is no depreciation of the capital stock from year t to the year (t + 1); this
implies that Kt+1 = Kt + It , i.e., the capital stock at the beginning of the year (t + 1) is the sum of
the capital stock at the beginning of year t and investment implemented in year t. Therefore, the
profit function (1) is reduced to:

πt+1 (It ) = F ((Kt + It ), Lt+1 ) − wt+1 Lt+1 − G(It ) − It . (2)

Taking into account the risk attitude of the managers of the firm, the objective function of the
firm is to maximize the expected utility of the year (t + 1) profit:

Max E[U(πt+1 )] = Max E{U[F ((Kt + It ), Lt+1 ) − wt+1 Lt+1 − G(It ) − It ]}. (3)

The first-order condition for investment is:

∂E[U(πt+1 )]
= E[U  (πt+1 )(FIt − GIt − 1)] = 0. (4)
∂It
 H. Bo, E. Sterken / North American Journal of Economics and Finance 18 (2007) 59–75 63

The first-order condition (4) implies that at optimality the marginal revenue of investment
equals the marginal cost of investment, i.e., FIt = (GIt + 1). The second-order condition is:

∂E[U(πt+1 )]2
= E[U  (πt+1 )(FIt − GIt − 1)2 + U  (πt+1 )(FIt It − GIt It )] = H. (5)
∂It2
We assume H < 0 to ensure the existence of the maximized profit, which implies that we assume
a concave marginal product of capital (FIt It < 0) and a convex adjustment cost function of capital
(GIt It > 0). Notice that H < 0 does not necessarily imply U (πt+1 ) < 0 (risk aversion), because
in the optimal investment solution, FIt = (GIt + 1), which allows U (πt+1 ) to have any sign.
Therefore, the model applies to decision-makers who may be risk-averse, risk-neutral, or risk-
taking. The first- and the second-order conditions (4) and (5) ensure the optimal investment policy
when there is no uncertainty. Below we introduce uncertainty into the model. We first characterize
uncertainty, then we investigate how uncertainty affects the optimal investment policy.
Assuming that the firm is facing demand uncertainty only, changes in the demand condition
are reflected by changes in the revenue function. Following Sandmo (1970, 1971), we add two
shift parameters, (γ, θ), to the revenue function F(K, L), i.e., the revenue function under demand
uncertainty is (γF + θ). Further we assume that the stochastic demand can be characterized as a
mean reversion process (e.g., Sandmo, 1970), which requires:

E[γF + θ] = 0, (6)

Totally differentiating (6) gives: (dθ/dγ) = −E[F] = −μ, where μ is the mean of the revenue
distribution. With demand uncertainty characterized by (6), the profit function (2) now becomes:

πt+1 (It , γ) = [γF ((Kt + It ), Lt+1 ) + θ] − wt+1 Lt+1 − G(It ) − It . (7)

Now we can check the impact of demand uncertainty on the optimal investment policy. By
totally differentiating the first-order condition for investment (4), and after implementing the
stochastic shift γF + θ, with respect to the stochastic parameter γ:
  
 ∂πt+1 ∂πt+1 ∂It
E U (πt+1 ) + (FIt − GIt − 1)
∂γ ∂It ∂γ
 
∂It ∂It
+ U  (πt+1 ) FIt It − GIt It = 0. (8)
∂γ ∂γ

Since (∂πt+1 /∂γ) = F + (dθ/dγ) = F − μ and (∂πt+1 /∂It ) = Ft − GIt − 1, from (8) we have:
∂It
E[U  (πt+1 )(Ft − GIt − 1)2 + U  (πt+1 )(FIt It − GIt It )]
∂γ
= −E[U  (πt+1 )(F − μ)(FIt − GIt − 1)]. (9)

Notice that the term E[. . .] on the left-hand side of (9) is the second-order condition (H);
therefore:
∂It 1
= − E[U  (πt+1 )(F − μ)(FIt − GIt − 1)]. (10)
∂γ H
64 H. Bo, E. Sterken / North American Journal of Economics and Finance 18 (2007) 59–75

Rewriting (10) in terms of the Arrow-Pratt measure of absolute risk aversion

Ra = −(U (πt+1 )/U (πt+1 )), we obtain:
∂It 1
= E[Ra U  (πt+1 )(F − μ)(FIt − GIt − 1)]. (11)
∂γ H
Under uncertainty, (F − μ) and (FIt − GIt − 1) change in the same direction, so that the product
of the two terms is positive. We also know that U (πt+1 ) > 0 and H < 0; therefore, the sign of the
impact of demand uncertainty on investment depends on the sign of the measure of absolute risk
aversion Ra :
∂It ∂It ∂It
> 0, = 0, and < 0 if Ra < 0 (risk-taking), Ra = 0 (risk-neutral),
∂γ ∂γ ∂γ
Ra > 0 (risk-averse).

The above model shows that the risk-attitude of the managers of the firm influences the impact
of demand uncertainty on investment. The model provides a theoretical motivation to include the
risk attitude parameter of managers in empirical investment equations. The next issue is how to
construct an empirical proxy for the risk attitude of the managers of the firm.

3. Empirical proxy for the risk attitude of managers

One can use survey data to obtain a direct indicator of the risk attitude of the managers of the
firm. For example, one may ask the managers to compare project A with certain returns (X) with
project B with risky expected returns (Y). The managers may be asked to provide information
on the returns they require if they choose to invest in project B. If the required expected returns
of choosing the risky project is Ŷ , then the difference (Ŷ − X) indicates the expected wealth the
managers of the firm are prepared to forgo in order to change a risky choice into a certainty.
Since the risk premium measures how much value the managers are willing to pay to convert an
uncertainty choice into a certainty, the measure of risk attitude can be inferred from information on
the risk premium. Although the use of survey data to derive risk attitude parameter for managers
is scant in the literature on firm behavior, deriving a subjective measure of risk attitudes of
consumers is a popular approach in research on consumers’ intertemporal decisions (e.g., Barsky,
Juster, Kimball, & Shapiro, 1997; Guiso & Paiella, 2000; and Guiso, Jappelli, & Pistaferri, 2002).
Due to the lack of survey data, we follow a statistical approach to construct a proxy for the
risk attitude of managers. We follow Fisher and Hall (1969)3 who suggest observed net returns
data to estimate the risk premium for firms. As these authors point out, if the firm is following
the optimal decision rule, then the risk premium of the firm can be measured by the moments
of the distribution of net returns. After the risk premium of the firm is estimated, we can derive
the measure of risk aversion for the managers of the firm based on the relation between the risk
premium and the measure of risk aversion predicted by utility theory (e.g., Arrow, 1971).
Before deriving our empirical proxy for the risk attitude of managers, we briefly present the
economic theory underlying this approach. The firm is maximizing the utility function U(π + W),
where π is earnings (a random variable) and W is net worth. Expanding U(π + W) in a Taylor

3 Also see Antle (1989).

 H. Bo, E. Sterken / North American Journal of Economics and Finance 18 (2007) 59–75 65

series around the point (π + W) = E(π + W) gives:

U  
U(π + W) = U(π + W) + U  (π + W)(π − π ) + (π + W)(π − π )
2
2!
U  
(π + W)(π − π ) + · · ·.
3
+ (12)
3!

Taking expected values and holding W and π constant,

U   U  
E[U(π + W)] = U(π + W) + σπ2 (π + W) + σπ3 (π + W) + · · ·. (13)
2! 3!
After rearranging terms, the difference between expected utility and utility of expected earnings
is:
U   U  
U(π + W) − E[U(π + W)] = −σπ2 (π + W) − σπ3 (π + W) + · · ·. (14)
2! 3!
The above expression is the risk premium. It shows that the second, third and higher moments
affect the magnitude of the risk premium. Since U < 0 for a concave utility function (a risk-averse
agent), the risk premium is positively associated with larger variances of returns. However, since
U  could be positive, zero, or negative, the impact of the third moment on the risk premium is
ambiguous. Higher moments add little information about the characteristics of the distribution
and are often ignored (Fisher & Hall, 1969).
Expression (14) contains the relation between the risk premium and the measure of risk aver-
sion. The left-hand side of (14) is the risk premium, on the right-hand side the second derivative
of the utility function U reveals the risk attitude of the agent. Arrow (1971) illustrates the relation
between the risk premium and the measure of risk aversion:
 
1
risk premium = σ 2 Ra + terms of higher order, (15)
2
where σ 2 is the variance of a stochastic variable (e.g., net returns). Ra is the measure of absolute
risk aversion. Therefore, if we know the risk premium of the firm and the volatility of net returns,
it is possible to derive the measure of risk aversion for the managers of the firm.
According to Fisher and Hall (1969), if the firm is following the optimal decision rule, then the
risk premium can be measured by moments of the distribution of net returns. We use historical
data on net profits to construct the risk premium. We scale net profit by total assets, in order to
eliminate size effects (also see Fisher & Hall, 1969). The distribution of the profit rate per firm,
i.e., the variance and the skewness are computed over the past 5 annual observations.
Since the second and the third moments of the distribution of the profit rate are able to reflect
the risk premium of the firm, we are especially interested in how much of realized net profits can
be explained by the second and the third moments. To see that, we estimate the following equation
for the profit rate:
PTR ∗
t = PT + α1 SDt + α2 SKEWt , (16)
where PTR t is realized net profits (scaled by total assets) at time t. SDt is the standard deviation
of the profit rate at time t. SKEWt is the third moment of the profit rate (skewness). We compute
both the standard deviation and the skewness of the profit rate at time t by using the last 5 annual
observations of the profit rate. PT* is a constant, which reflects all influences on realized profits
66 H. Bo, E. Sterken / North American Journal of Economics and Finance 18 (2007) 59–75

not encompassed by the standard deviation and skewness: the ‘risk-adjusted’ rate of return. The
risk premium is the difference between realized profits and ‘risk-adjusted’ profits. Therefore, by
estimating Eq. (16), we obtain ‘risk-adjusted’ profits (PT* ) for each firm. By assuming that the
firm’s ‘risk-adjusted’ profits (PT* ) are constant over the whole sample period, we construct a
proxy for the risk premium as RPt = (PTR ∗
t − PT ).
Utility theory (see Eqs. (14) and (15)) suggests that we can derive a risk attitude parameter for
managers by looking at how much the risk premium is attributed to the second moment of the
distribution of profit. Therefore we estimate the following equation firm by firm:

RPt = RC ∗ SDt + ω ∗ SKEWt , (17)

where the estimated coefficient for the standard deviation of profits (RC) reveals the measure
of absolute risk-aversion of the managers of the firm. Table 1 presents the results of estimating
the risk aversion coefficient (RC) (Eq. (17)) for all the firms in the sample. Among the initial
94 firms (see data description below), 68 firms are valid samples. For these firms, the second
moment of the profit rate can significantly explain realized net profits, which indicates that risk
attitude is important in determining the magnitude of the risk premium for these firms. Table 1
shows that some sample firms have a negatively significant risk coefficient (RC < 0), some others
have a positively significant risk coefficient (RC > 0). According to utility theory, the positive
risk coefficient (RC > 0) indicates that the managers of the firm are risk-averse. If the estimated
risk coefficient is negative (RC < 0), then it suggests that the managers are risk-taking. Finding a
non-significant parameter estimate RC indicates risk neutrality.
By estimating Eq. (17), we obtain a parameter that measures the degree of risk aversion of the
managers of the firm, which we call the ‘risk coefficient’ (RC). Since the risk premium is proxied
by the component of realized profits that is explained by the second and third moments of the
distribution of profits, the estimated risk coefficient indicates how much of the risk premium can
be attributed to the second moment of the distribution of profits. From utility theory (e.g., Arrow,
1971), a higher risk coefficient corresponds to a higher degree of risk aversion. We will use the
estimated risk coefficient (RC) as the empirical proxy for the risk attitude of the managers of the
firm in the remainder of this paper.

4. Data and construction of the uncertainty measure

4.1. Data description

We start with a stratified unbalanced panel of 94 Dutch non-financial firms listed at the Ams-
terdam Stock Exchange (AEX) during the period 1985–2000. We restrict the sample to 68 firms,
whose risk attitude matters significantly in explaining realized profits (the estimated risk coef-
ficient RC is statistically significant in estimating Eq. (17)). This implies that we exclude the
26 firms with insignificant estimates of the risk-attitude parameter RC, thereby ruling out risk-
neutral firms. In order to construct the data on changes in tangible fixed assets (net investment)
and the annual growth rate of sales, the first-year observation is lost for each firm. To construct
the measure of demand uncertainty (see the explanations below), the first three observations are
lost for each firm. This implies that the longest time series in the empirical investment model is
1988–2000.
The 68 valid sample firms are from 9 main industries, including construction, metals/machinery,
chemical, textile, food, transportation, retail/wholesale, and business service/information. Table 2
 H. Bo, E. Sterken / North American Journal of Economics and Finance 18 (2007) 59–75 67

Table 1
Estimating the risk attitude coefficient: RPt = RC*SDt + ω*SKEWt
Firm RC t-Statistic R2 Firm RC t-Statistic R2

Aaberts −0.0566 −0.1548 0.0778 Internatio Muller −0.2198 −0.5735 0.6206

Ahold 0.9639 3.2257 0.0468 KBB 1.8485 4.0979 0.2927
Ahrend 0.5008 1.1058 0.0380 Klene 0.0706 0.2859 0.3005
Akzo −0.1804 −0.4365 0.0209 KLM −1.3798 −3.1487 0.4285
Alanheri −1.9578 −3.9281 0.2997 Krasnapolsky 2.7755 8.5862 0.5344
Atag −2.8018 −29.0308 0.9894 Lander −0.1557 −0.2586 0.3507
Athlon −0.1161 −0.2408 0.0107 Macintosh −0.0640 −0.2064 0.0003
Ballast 0.7081 4.2716 0.5141 Management share −0.8175 −2.1361 0.4046
Batenburg −0.3169 −0.6730 0.3053 Nagron −0.8940 −4.7155 0.4289
Beers −0.2390 −0.6951 0.0948 NBM 1.1892 3.6310 0.5483
Blydenstein 0.9108 6.1365 0.5752 Nedap 1.7594 4.6331 0.2393
Boer −2.2658 −5.2906 0.5787 Nedcon −1.7634 −6.1256 0.3651
Bolswessanen 2.0132 4.8794 0.3239 Nedlloyd 0.2053 1.4352 0.6269
Boskalis −0.7348 −3.7682 0.4614 Nedschroef 0.1745 0.2798 0.0482
Burgman −1.5155 −1.7604 0.4322 Neways 0.1194 0.5239 0.0826
Cate Ten 0.05231 0.1095 0.1341 NKF 0.0021 1.7798 0.3185
Cindu 2.5855 31.4980 0.9942 Norit −1.6555 −5.5189 0.3066
Content 0.4086 2.7804 0.4846 Oce 0.2516 0.6631 0.2538
CSM 2.0278 5.1722 0.5879 Ommeren 2.9059 7.0579 0.6551
Delft Instruments −1.2452 −3.1873 0.2970 Otra −1.7749 −4.6891 0.6502
Dico −3.4981 −19.7045 0.9219 PC Groep −2.2783 −15.0152 0.8809
Dorp −1.5795 −4.8288 0.4743 Pakhoed 0.3179 0.7483 0.1255
Draka 0.9919 2.8386 0.1484 Philips 2.1073 7.5946 0.7281
Drie Electronics −1.2516 −4.7605 0.3337 Polygram −2.9116 −9.3338 0.9198
DSM −0.6375 −1.9612 0.0649 Polynorm −0.1371 −0.3435 0.0905
Econosto −2.0913 −7.1415 0.5086 Porceleyne −2.2163 −5.7416 0.3483
Eriks 0.6854 1.3234 0.2730 Randstad 1.3179 4.3754 0.1366
Flexovit −1.9903 5.5535 0.5388 Reesink −1.4018 −8.7633 0.7553
Free Record −0.6297 −2.3549 0.1812 Rood Testhouse −0.3597 −0.0342 0.0415
Fugro 0.2208 0.4203 0.0086 Roto Smeets −1.2746 −4.4554 0.4107
Gamma −2.7925 −14.8293 0.8531 Samas −0.6368 −1.5631 0.0419
Gelderse −1.8928 −4.7454 0.5587 Schuitema 1.7391 5.1784 0.6350
Getronics −1.5517 −7.6314 0.7853 Schuttersveld −0.2617 −0.8578 0.0195
Geveke −0.5579 −1.6367 0.1967 Simac −1.1087 −2.6722 0.4489
Grolsch 1.0983 3.0002 0.1742 Sligro 0.0326 0.7577 0.7878
Grontmy −0.7143 −2.2777 0.3439 Smit International 0.3913 2.5746 0.5104
Gti −1.7930 −9.0192 0.8046 Sphinx −3.1353 −15.9950 0.8576
Hagemeyer 1.1679 3.5355 0.4413 Stork −2.9009 −11.6011 0.5592
Heineken 3.9838 13.9204 0.8102 Telegraaf 1.6243 5.8847 0.6541
Helvoet −1.9057 −11.3242 0.7899 Ubbink −1.2932 −2.0006 0.2618
Hes −0.6466 −1.2488 0.1438 Unilever −1.0842 −3.1222 0.5012
Heymans −0.6797 −7.8462 0.8159 VNU 0.4461 1.7261 0.4379
Hoeks −1.0979 −3.4393 0.1351 Volker −0.3189 −1.7043 0.4808
Holland Colours 0.2880 2.7729 0.3721 Wegener −0.4843 −1.7552 0.1963
Hollandse Beton −2.5689 −15.0609 0.9541 Weweler −1.3042 −5.0735 0.3271
Hoogovens −1.4230 −3.9302 0.2096 Wolters −0.4465 −1.3446 0.0813
IHC 0.2690 4.4174 0.4449

Notes: Data sources: REACH, 94 Dutch listed nonfinancial firms in 1985–2000.

68 H. Bo, E. Sterken / North American Journal of Economics and Finance 18 (2007) 59–75

Table 2
Summary statistics
Mean Median Std. Dev. Skewness Kurtosis Obs.

I/K 0.0781 0.0685 0.0742 −0.8384 28.1526 739

SG 0.1071 0.0738 0.2217 4.1818 35.4453 739
CF/K 0.1073 0.1098 0.0569 −2.4243 20.7597 739
PT/K 0.0538 0.0569 0.0547 −2.6481 23.7436 739
UMS 0.1797 0.1312 0.2043 4.3936 29.5739 739
RP −0.0211 −0.0116 0.0626 −3.0681 26.2908 739
RC −0.4899 −0.8940 1.6992 0.5044 2.5139 739
RCdum *UMS 0.0898 0.0264 0.1728 5.8414 53.9620 739
(1 − RCdum )*UMS 0.0898 0.0000 0.1675 3.9547 24.1826 739

Notes: (1) Data sources: REACH, 68 Dutch listed nonfinancial firms in 1985–2000. (2) Explanations of variables: I/K,
gross investment scaled by total assets; SG, annual growth rate of sales; CF/K, cash flow scaled by total assets; PT/K, net
profit scaled by total assets; UMS, measure of demand uncertainty; RP, risk premium; RC, risk coefficient; RCdum , risk
coefficient dummy, which is defined as one if RC is above the sample median, otherwise it is zero.

reports the summary statistics for the relevant variables. We also report the summary statistics
for the constructed risk premium and the estimated risk coefficients. As shown in Table 2, for
both the risk premium and the risk coefficient the mean and the median are negative. These
statistics suggest that the sample firms on average show risk-taking behavior. Table 3 presents the
correlation matrix for the variables used in the empirical analyses.

4.2. The measure of demand uncertainty

The investment model derived in Section 2 embeds a structural link between attitude towards
risk and the effect of demand uncertainty on investment. Stochastic demand is found to be an
important source of uncertainty faced by the firm (e.g., Guiso & Parigi, 1999), because it captures
the overall external uncertainty surrounding the firm. To construct the measure of demand uncer-
tainty, we first specify a forecasting equation for sales. In the forecasting equation, we assume
that detrended sales follow an AR(1) process:

St = c0 + c1 Trend + c2 St−1 + ξt , (18)

Table 3
Correlation matrix of the variables used in the estimations
I/K SG CF/K UMS RCdum *UMS (1 − RCdum )*UMS

I/K 1.0000
SG 0.3024 1.0000
CF/K 0.0879 0.1712 1.0000
UMS −0.0174 0.0527 0.0522 1.0000
RCdum *UMS −0.0147 −0.0251 0.0339 0.6168 1.0000
(1 − RCdum )*UMS −0.0061 0.0902 0.0287 0.5834 −0.2795 1.0000

Notes: (1) Data sources: REACH, 68 Dutch listed nonfinancial firms in 1985–2000. (2) Explanations of variables: I/K,
gross investment scaled by total assets; SG, annual growth rate of sales; CF/K, cash flow scaled by total assets; PT/K, net
profit scaled by total assets; UMS, measure of demand uncertainty; RP, risk premium; RC, risk coefficient; RCdum , risk
coefficient dummy, which is defined as one if RC is above the sample median, otherwise it is zero.
 H. Bo, E. Sterken / North American Journal of Economics and Finance 18 (2007) 59–75 69

where c0 is a constant, c1 and c2 are parameters, and ξ is an error term. We estimate Eq. (18) firm by
firm. The estimation of the forecasting equation is based on the original data set (1985–2000). The
residuals are saved for the period of 1986–2000. We then compute the rolling standard deviations
of the residuals as the proxy for demand uncertainty.
This method of constructing the uncertainty measure and the similar forecasting equation have
been adopted in the literature (e.g., Bo & Sterken, 2002; and Ghosal & Loungani, 2000). The
measure of uncertainty constructed in this way emphasizes the unpredictable part of the stochastic
process that governs demand. The rolling standard deviation of the residuals is based on all the
past information of the residuals. More specifically, the rolling standard deviation of the residuals
for the year 1988 is the standard deviation of the residuals computed over 1986, 1987, and 1988.
For the year 1989, the rolling standard deviation of the residuals is the standard deviation of the
residuals computed over 1986, 1987, 1988, and 1989, and so on. In estimating the investment
equations, the standard deviation of the residuals is scaled by total assets to eliminate size effects.
The scaled standard deviation is used as the measure of demand uncertainty and it is denoted by
UMSit .

5. Attitude towards risk and the investment-uncertainty relation

In this section we document the role of the risk attitude of managers in establishing the sign
of the investment-uncertainty relation. We estimate a simple accelerator (sales growth) model
including demand uncertainty. We first interact sales uncertainty with a dummy variable of the
risk attitude of managers. To check for the robustness of the results, we further split the sample
by the proxy for the risk attitude of managers based on whether the estimated risk coefficient is
positive or negative.

5.1. Risk attitude dummy

Based on the estimated risk coefficient (RC), a dummy variable is constructed: we define the
risk coefficient to be equal to unity if the value of the estimated risk coefficient is above the
sample median and to zero otherwise. By interacting the risk coefficient dummy with the measure
of demand uncertainty, we can check whether the effect of demand uncertainty differs due to
the differences in the degree of risk aversion of managers. More specifically, we estimate the
following investment equation:
 
I
= fi + ft + β1 SGit + β2 (UMS ∗ RCdum )it + β3 (UMS ∗ (1 − RCdum ))it + eit ,
K it
(19)

where Iit is gross investment of firm i at time t, Kit are the beginning-of-period total assets of firm
i at time t, fi and ft are fixed effect and time effect, respectively, SG is the annual growth rate of
sales, representing the accelerator effect, UMSit is the measure of demand uncertainty, RCdum
denotes the risk coefficient dummy, and eit is an error term.
The investment model is estimated by the system GMM estimator (Arellano & Bond, 1998).
The system GMM estimation consists of a set of first-differenced equations and a set of lev-
els equations. Moment conditions for equations in first differences are combined with moment
conditions for equations in levels in order to compute the optimal weighting matrix providing
consistent estimators. Time dummies are added in all models to correct for business cycle effects.
70 H. Bo, E. Sterken / North American Journal of Economics and Finance 18 (2007) 59–75

We also take into account the industry effect by adding industry dummies. In the system GMM
estimation, the instruments for the first-difference equations are the lagged levels of the right-hand
side variables.
In most of the estimations of this paper, we use the level observations lagged from t − 2 to
t − 5 periods of the right-hand side variables as the instruments for the first-difference equations.4
This decision is based on the failure of the Sargan test of over-identifying restrictions to reject the
null hypothesis of instrument validity. In addition, we use the lagged first differences of the right-
hand side variables as instruments for the levels equations. Since inference based on asymptotic
standard errors for the one-step estimators is more reliable than that for the two-step estimators,
we only report the results for the one-step estimators with robust test statistics (see Arellano &
Bond, 1991; and Blundell & Bond, 1998).
Table 4 reports the GMM estimation results. Column 1 of Table 4 presents the estimated parame-
ters and diagnostic checks of the benchmark model (19). The most important result is the difference
between the estimated coefficients β2 and β3 . We observe that the estimated coefficient for demand
uncertainty for high-risk coefficient firms (UMS*RCdum ) is not significant. In contrast, the esti-
mated coefficient for demand uncertainty for low-risk coefficient firms (UMS*(1 − RCdum )) is
highly significant with a positive sign. Since a higher risk coefficient corresponds to a higher
degree of risk aversion, the insignificant coefficient of (UMS*RCdum ) suggests that managers
with a higher degree of risk aversion are more cautious when responding to demand uncertainty:
they simply do not increase the capital stock in responding to demand uncertainty loosely speak-
ing. However, managers with a lower degree of risk aversion (UMS*(1 − RCdum )) increase fixed
investment expenditures in responding to high demand uncertainty.
In column 2 of Table 4 we control for investment persistence in order to check the robustness of
the results. Specifically, we control the lagged one-period investment rate and lagged one-period
sales in the estimations. In column 3 of Table 4 we further control for cash flow in the estimations,
because cash flow is mostly found to be important in explaining investment. The results shown in
columns 2 and 3 are quite consistent with those in column 1 of Table 4. The models in Table 4 meet
all necessary tests. m1 and m2 are tests for first- and second-order serial correlation in the first-
differenced residuals, asymptotically distributed as normal distribution under the null of no serial
correlation. If the differenced residuals display significant negative first-order serial correlation
(m1 is significantly negative) and no second-order serial correlation (m2 is not significant), there
is no serial correlation in the residuals. The results presented in Table 4 show that the tests of serial
correlation in the first-differenced residuals (m1 and m2 ) indicate that there is no serial correlation
in the residuals. The Sargan (k) tests for the overidentifying restrictions (asymptotically distributed
as χ2 (df = k) under the null of instrument validity). The Sargan test statistic supports the validity
of the instruments used in the estimations.

5.2. Risk-averse versus risk-taking managers

To further check the robustness of the results, in this section we use the estimated risk coefficient
(RC) to split the sample into two sub-groups, based on whether the estimated risk coefficient (RC)
is positive or negative. Remember that firms used in estimating the empirical investment equation

4 In the estimations shown in columns 1–3, the instruments for the first-difference equations vary a bit because the

sub-sample of risk-averse firms is relatively small, for which use of the instruments uniformly lagged form t − 2 to t − 5
periods of the right-hand side variables sometimes results in unsatisfactory model performance. For example, in some
cases serial correlation remains in the residuals.
 H. Bo, E. Sterken / North American Journal of Economics and Finance 18 (2007) 59–75
Table 4
System GMM estimation results: risk attitude dummy (I/K)it = fi + ft + β1 SGit + β2 (UMS*RCdum )it + β3 (UMS*(1 − RCdum ))it + eit
1 2 3

(I/K)t−1 0.0325 (0.6324) 0.0182 (0.3657)

SGt 0.0596 (2.5625) 0.0653 (3.2359) 0.0744 (4.1488)
SGt−1 0.0073 (0.3603) 0.0071 (0.3540)
(CF/K)t 0.0398 (0.8606)
UMSt *RCdum 0.0156 (0.6782) 0.0197 (0.6485) 0.0184 (0.6043)
UMSt *(1 − RCdum ) 0.0872 (2.4354) 0.0832 (2.4659) 0.0754 (2.2346)
m1 −2.919 −4.484 −4.360
m2 −1.139 −0.573 −0.769
Sargan (k) 50.029 (144) 43.327 (191) 44.585 (239)
Instruments (dif.) SGt−2. . .t−5 , (UMS*RCdum )t−2. . .t−5 , (I/K)t−2. . .t−5 , SGt−2. . .t−5 , (I/K)t−2. . .t−5 , SGt−2. . .t−5 , (CF/K)t−2. . .t−5 ,
(UMS*(1 − RCdum ))t−2. . .t−5 (UMS*RCdum )t−2. . .t−5 , (UMS*RCdum )t−2. . .t−5 , (UMS*(1 − RCdum ))t−2. . .t−5
(UMS*(1 − RCdum ))t−2. . .t−5
Instruments (levels) SGt−1 , (UMS*RCdum )t−1 , (I/K)t−1 , SGt−1 , (UMS*RCdum )t−1 , (I/K)t−1 , SGt−1 , (CF/K)t−1 , (UMS*RCdum )t−1 ,
(UMS*(1 − RCdum ))t−1 (UMS*(1 − RCdum ))t−1 (UMS*(1 − RCdum ))t−1

Notes: (1) Data source: REACH. (2) The one-step GMM estimators are reported. (3) Heteroskedasticity consistent asymptotic t-statistics are in parentheses. (4) m1 and m2
are tests for first- and second-order serial correlation in the first-differenced residuals, asymptotically distributed as normal distribution under the null of no serial correlation.
(5) Sargan (k): test of the overidentifying restrictions, asymptotically distributed as Chi-square (k) under the null of instrument validity. (6) Time effects are controlled in all
estimations by adding time dummies. Industry dummies are also controlled. (7) Explanations of variables: see notes to Table 2.

71
72 H. Bo, E. Sterken / North American Journal of Economics and Finance 18 (2007) 59–75

Table 5
System GMM estimation results: whole sample (I/K)it = fi + ft + θ 1 SGit + θ 2 UMSit + εit
1 2 3

(I/K)t−1 0.0286 (0.5612) 0.0082 (0.1691)

SGt 0.0739 (2.9686) 0.0728 (3.3847) 0.0787 (4.1085)
SGt−1 0.0042 (0.2009) 0.0032 (0.1567)
(CF/K)t 0.0565 (1.0437)
UMSt 0.0451 (2.0544) 0.0439 (1.9554) 0.0343 (1.4679)
m1 −2.908 −4.438 −4.331
m2 −1.238 −0.678 −0.964
Sargan (k) 40.881(96) 41.202(143) 42.054(191)
Instruments (dif.) SGt−2. . .t−5 , UMSt−2. . .t−5 (I/K)t−2. . .t−5 , SGt−2. . .t−5 , (I/K)t−2. . .t−5 , SGt−2. . .t−5 ,
UMSt−2. . .t−5 (CF/K)t−2. . .t−5 , UMSt−2. . .t−5
Instruments (levels) SGt−1 , (UMS)t−1 (I/K)t−1 , SGt−1 , (I/K)t−1 , SGt−1 ,
(UMS)t−1 (CF/K)t−1 , (UMS)t−1

Notes: (1) Data source: REACH. (2) The one-step GMM estimators are reported. (3) Heteroskedasticity consistent asymp-
totic t-statistics are in parentheses. (4) m1 and m2 are tests for first- and second-order serial correlation in the first-differenced
residuals, asymptotically distributed as normal distribution under the null of no serial correlation. (5) Sargan (k): test of
the overidentifying restrictions, asymptotically distributed as Chi-square (k) under the null of instrument validity. (6) Time
effects are controlled in all estimations by adding time dummies. Industry dummies are also controlled. (7) Explanations
of variables: see notes to Table 2.

are firms for whom the estimated risk coefficient is significant in explaining the risk premium
(Eq. (17)). For these firms, the estimated risk coefficient is either positive or negative. According
to utility theory, the positive risk coefficient (RC > 0) indicates that the managers of the firm are
risk-averse. If the estimated risk coefficient is negative (RC < 0), then it suggests that the managers
are risk-taking. It is logical to assume that the effect of demand uncertainty on investment is highly
likely to differ between risk-averse and risk-taking firms.
Therefore, we split the sample into two groups of firms, one with positive risk coefficients and
the other with negative risk coefficients. Among 68 valid sample firms, 44 firms have negative risk
coefficients and 24 firms have positive risk coefficients.5 For both groups of firms we estimate
the following investment equation:
 
I
= fi + ft + θ1 SGit + θ2 UMSit + εit . (20)
K it
The aim of estimating the investment model (20) is to test whether the estimated effect of
demand uncertainty, θ 2 , differs between risk-averse and risk-taking firms. Before estimating Eq.
(20) for sub-samples of firms, we present in Table 5 the results when all sample firms are pooled
together in order to better understand the differences between the two sub-sample firms.
Column 1 of Table 5 displays the results of estimating the investment Eq. (20) for the whole
valid sample. In column 2 we control investment persistence and in column 3 we further control
for the cash-flow effect. The results in Table 5 suggest that when all firms are pooled together
the estimated effect of demand uncertainty on investment is positive overall and it is significant
in two out of three estimated equations. The estimated positive effect of demand uncertainty on

5 Note that we exclude the risk-neutral cases: firms with an insignificant estimate of RC. On the one hand, inclusion of

risk-neutral firms in our estimation sample makes the argument more pronounced, but on the other leads to a slight loss
of generality.
 H. Bo, E. Sterken / North American Journal of Economics and Finance 18 (2007) 59–75
Table 6
System GMM estimation results: risk-averse vs. risk-taking firms, (I/K)it = fi + ft + θ 1 SGit + θ 2 UMSit + εit
Risk-averse firms (RC > 0) Risk-taking firms (RC < 0)

1 2 3 4 5 6

(I/K)t−1 −0.0330 (−0.2869) −0.0476 (−0.4574) −0.0675 (1.7011) 0.0466 (1.2256)

SGt 0.0669 (2.2635) 0.0599 (1.8804) 0.0705 (2.1217) 0.0859 (2.9998) 0.0787 (2.8741) 0.0655 (2.3909)
SGt−1 0.0014 (0.0498) 0.0078 (0.2736) 0.0211 (0.7299) 0.0216 (0.7584)
(CF/K)t −0.1996 (−0.9531) 0.0638 (1.2859)
UMSt −0.1809 (−1.8976) −0.1643 (−1.8916) −0.1286 (−1.8783) 0.0387 (2.0652) 0.0557 (2.6255) 0.0515 (2.2498)
m1 −3.257 −3.247 −3.203 −1.989 −3.171 −2.983
m2 −1.050 −1.083 −1.346 −1.257 −0.276 −0.663
Sargan (k) 0.506 (109) 0.000 (131) 167.505 (140) 17.290 (96) 17.708 (143) 18.223 (191)
Instruments (dif.) SGt−2. . .t−5 , (I/K)t−2. . .t−3 , (I/K)t−2. . .t−3 , SGt−2. . .t−5 , (I/K)t−2. . .t−5 , (I/K)t−2. . .t−5 ,
UMSt−2. . .t−7 SGt−2. . .t−4 , SGt−2. . .t−3 , UMSt−2. . .t−5 SGt−2. . .t−5 , SGt−2. . .t−5 ,
UMSt−2. . .t−7 (CF/K)t−2. . .t−3 , UMSt−2. . .t−5 (CF/K)t−2. . .t−5 ,
UMSt−2. . .t−5 UMSt−2. . .t−5
Instruments (levels) SGt−1 , (I/K)t−1 , SGt−1 , (I/K)t−1 , SGt−1 , SGt−1 , (I/K)t−1 , SGt−1 , (I/K)t−1 , SGt−1 ,
(UMS)t−1 (UMS)t−1 (CF/K)t−1 , (UMS)t−1 (UMS)t−1 (CF/K)t−1 ,
(UMS)t−1 (UMS)t−1

Notes: (1) Data source: REACH. (2) The one-step GMM estimators are reported. (3) Heteroskedasticity consistent asymptotic t-statistics are in parentheses. (4) m1 and m2
are tests for first- and second-order serial correlation in the first-differenced residuals, asymptotically distributed as normal distribution under the null of no serial correlation.
(5) Sargan (k): test of the overidentifying restrictions, asymptotically distributed as Chi-square (k) under the null of instrument validity. (6) Time effects are controlled in all
estimations. Industry dummies are also controlled. (7) Explanations of variables: see notes to Table 2.

73
74 H. Bo, E. Sterken / North American Journal of Economics and Finance 18 (2007) 59–75

investment may be explained by the notion that more than half of the sample firms have a negative
risk coefficient (see Table 1). This suggests that the sample firms on average show risk-taking
behavior when adjusting investment in responding to demand uncertainty. It is more interesting to
further check whether the effect of demand uncertainty differs between risk-averse and risk-taking
firms.
We estimate the same investment equations shown in Table 5 for risk-averse and risk-taking
firms, respectively. Table 6 reports the estimation results. The most important result emerging
from Table 6 concerns the difference in the estimated effect of demand uncertainty on investment,
θ 2 , between risk-averse and risk-taking firms. In all models, the estimated coefficient for the
measure of demand uncertainty is significant with a negative sign for risk-averse firms, while it is
significant with a positive sign for risk-taking firms. Consistent with the predictions of theoretical
models (e.g., Nickell, 1978), our results show that risk aversion discourages firm investment under
demand uncertainty, while risk-taking firms increase investment expenditures with higher demand
uncertainty. The evidence clearly suggests that the risk attitude of managers is indeed important
in affecting how firm investment responds to demand uncertainty.

6. Summary and conclusions

We argue that the risk attitude of managers is a heavily neglected determinant of the sign of the
investment-uncertainty relation. The obvious reason for this omission is the lack of an empirical
proxy for the risk attitude of managers. We try to fill this gap in this paper and show that simply
assuming risk neutrality might blur econometric estimates of the sign of the investment-uncertainty
relationship.
Following the idea put forward by Fisher and Hall (1969), we compute the second and the third
moments of the distribution of net profits and use these distribution variables to explain realized
net profits.6 The part of realized net profits that can be explained by these distribution variables
is defined as the risk premium of the firm. Since there is a theoretical connection between risk
premium and the measure of risk aversion, we derive an empirical proxy for the risk attitude of
managers by looking at how much the risk premium of the firm can be explained by the second
moment of the distribution of net profits. The reason is that only the second moment is relevant
to measuring the degree of risk aversion, according to utility theory (e.g., Arrow, 1971).
We test whether the proxy for the risk attitude of managers has an impact on the effect of
demand uncertainty on fixed investment for an unbalanced panel of Dutch listed non-financial
firms in the period of 1985–2000. The results show that in general a low degree of risk aversion is
associated with a positive effect of demand uncertainty on fixed investment. More specifically, we
obtain evidence that risk-averse firms respond to demand uncertainty by cutting investment, while
the investment undertaken by risk-taking firms responds to demand uncertainty positively. These
results are in line with Nickell (1978), who concludes that when demand is uncertain, risk-averse
behavior is bound to lower capacity levels and the optimal capacity level is a declining function
of the degree of risk aversion.
In sum, the evidence of this paper suggests that the risk attitude of managers is indeed important
in influencing how firm investment responds to demand uncertainty, which implies that the risk
attitude of managers might be a heavily neglected driving determinant of the often reported
negative sign of the investment-uncertainty relation in the empirical literature. Our results suggest

6 Also see Antle (1989).

 H. Bo, E. Sterken / North American Journal of Economics and Finance 18 (2007) 59–75 75

that the assumption of risk neutrality does not apply to fixed-investment decision makers. This
finding puts in question the results by other empirical studies that assume risk neutrality in
estimating the sign of the investment-uncertainty relationship.7 An issue for future research is the
exact identification of the different channels through which different types of uncertainty affect
investment.

References

Abel, A. B. (1983). Optimal investment under uncertainty. American Economic Review, 73, 228–233.
Antle, J. M. (1989). Nonstructural risk attitude estimation. American Journal of Agricultural Economics, 71, 774–784.
Arellano, M., & Bond, S. (1991). Some tests of specification for panel data: Monte Carlo evidence and an application to
employment equations. Review of Economic Studies, 58, 277–297.
Arellano, M., & Bond, S. (1998). Dynamic panel data estimation using DPD98 for gauss: A guide for users. Institute for
Fiscal Studies (IFS) working paper no. 88/15.
Arrow, K. J. (1971). Essays in the theory of risk-bearing. North Holland: Markham Publishing Company.
Barsky, R. B., Juster, F. T., Kimball, M. S., & Shapiro, M. D. (1997). Preference parameters and behavioral heterogeneity:
An experimental approach in the health and retirement study. Quarterly Journal of Economics, 111, 537–579.
Blundell, R., & Bond, S. (1998). Initial conditions and moment restrictions in dynamic panel data models. Journal of
Econometrics, 87, 115–143.
Bo, H., & Sterken, E. (2002). Volatility of the interest rate, debt and firm investment: Dutch evidence. Journal of Corporate
Finance, 8, 179–193.
Caballero, R. J. (1991). Competition and the non-robustness of the investment-uncertainty relationship. American Eco-
nomic Review, 81, 279–288.
Dixit, A. K., & Pindyck, R. S. (1994). Investment under uncertainty. Princeton, NJ: Princeton University Press.
Fisher, I. N., & Hall, G. R. (1969). Risk and corporate rates of return. Quarterly Journal of Economics, 83, 79–92.
Ghosal, V., & Loungani, P. (2000). The differential impact of uncertainty on investment in small and large business.
Review of Economics and Statistics, 82, 338–349.
Gollier, C. (2000). What does the classical theory have to say about portfolio choice? In L. Guiso, M. Haliassos, & T.
Jappelli (Eds.), Household portfolios. Cambridge, MA: MIT Press.
Guiso, L., Jappelli, T., & Pistaferri, L. (2002). An empirical analysis of earnings and employment risk. Journal of Business
& Economic Statistics, 20, 241–253.
Guiso, L., & Paiella, M. (2000). Risk aversion, wealth and background risk. Working paper, University of Sassari, Italy.
Guiso, L., & Parigi, G. (1999). Investment and demand uncertainty. Quarterly Journal of Economics, 113, 185–227.
Hartman, R. (1972). The effects of price and cost uncertainty on investment. Journal of Economic Theory, 5, 258–266.
Hartman, R. (1976). Factor demand with output price uncertainty. American Economic Review, 66, 675–682.
Leahy, J. V., & Whited, T. M. (1996). The effect of uncertainty on investment: Some stylized facts. Journal of Money,
Credit, and Banking, 28, 64–83.
Leland, H. E. (1972). Theory of the firm facing uncertain demand. American Economic Review, 62, 278–291.
Lensink, R., Bo, H., & Sterken, E. (2001). Investment, capital market imperfections and uncertainty: Theory and empirical
results. Cheltenham, UK: Edward Elgar.
Minton, B. A., & Schrand, C. (1999). The impact of cash flow volatility on discretionary investment and the costs of debt
and equity financing. Journal of Financial Economics, 54, 423–460.
Nakamura, T. (1999). Risk-aversion and the uncertainty-investment relationship: A note. Journal of Economic Behavior
& Organization, 38, 357–363.
Nickell, S. J. (1978). The investment decision of firms. Cambridge, MA: Cambridge University Press.
Sandmo, A. (1970). The effect of uncertainty on saving decisions. Review of Economic Studies, 37, 353–360.
Sandmo, A. (1971). On the theory of the competitive firm under price uncertainty. American Economic Review, 61, 65–73.
Zeira, J. (1990). Cost uncertainty and the rate of investment. Journal of Economic Dynamics and Control, 14, 53–63.

7 See Lensink et al. (2001) for an extensive overview.