Journal of Corporate Finance: Hyun Joong Im, Ya Kang, Janghoon Shon T

Journal of Corporate Finance 64 (2020) 101642
Contents lists available at ScienceDirect
Journal of Corporate Finance

journal homepage: www.elsevier.com/locate/jcorpfin
How does uncertainty influence target capital structure?☆

T
Hyun Joong Ima,⁎, Ya Kangb, Janghoon Shonc
a
HSBC Business School, Peking University, University Town, Nanshan District, Shenzhen 518055, China
b
NUS Business School, National University of Singapore, BIZ 2 Building #B1-3, 1 Business Link, 117592, Singapore
c
HKUST Business School, Hong Kong University of Science and Technology, Clear Water Bay, Kowloon, Hong Kong
ARTICLE INFO ABSTRACT
JEL classification: This study investigates how uncertainty affects firms’ target capital structure using a panel data
G31 set of U.S. public manufacturers between 2003 and 2018 and finds that high-uncertainty firms
G32 have 10.1 (8.1) percentage points lower mean book (market) targets than low-uncertainty firms.
G33 This study also shows that the uncertainty effect on leverage targets is greater than the impact of
D22
firm size, market-to-book ratio, assets tangibility, R&D intensity, and industry median leverage,
D81
making uncertainty the most critical among all time-varying determinants of leverage targets.
Further, this study finds that heightened uncertainty decreases debt tax shields, increases po-
Keywords:
Uncertainty tential financial distress costs, and exacerbates debtholder–shareholder conflicts, thereby leading
Target leverage to a lower optimal or target leverage ratio.
Capital structure
Asset volatility
1. Introduction
A growing body of literature has shown that uncertainty has material effects on decision-making—particularly investment and
financing decisions. Emphasising the role of market frictions such as capital irreversibility and financial constraints and the im-
portance of real option to wait and see, early studies have centred around the effect of uncertainty on corporate investments.1 Given
that uncertainty is unprecedentedly high around the globe, quite a few impressive studies regarding the implications of uncertainty
on corporate financing decisions in different contexts exist Gungoraydinoglu et al., 2017; Çolak et al., 2018; Li and Qiu, 2018). For
example, Gungoraydinoglu et al. (2017), using international data, show that the speed of leverage adjustment is lower when political
or policy uncertainty is higher. In addition, Çolak et al. (2018) show that political uncertainty raises the securities’ placement costs
☆
We are grateful for the valuable comments from the editor Bart Lambrecht and two anonymous referees as well as Steve Bond, Soku Byoun,
Hyunsoo Doh, Fangjian Fu, Rachita Gullapalli, Vidhan Goyal, Chang Yong Ha, Zhangkai Huang, Tim Jenkinson, Fuxiu Jiang, Jun-Koo Kang,
Kenneth Kim, Sukjoong Kim, Jose Martinez, Colin Mayer, Alan Morrison, Thomas Noe, Stefano Rossi, Oren Sussman, Kelvin Tan, Yuhai Xuan and
Alexander Vadilyev; to the conference participants at the 2018 China International Conference in Finance, the 2018 PKU–NUS Annual International
Conference on Quantitative Finance and Economics, the 2017 AsFA Conference, the 2016 FMA Asia-Pacific Meeting and the 2016 FMA Annual
Meeting; and to the seminar participants at Sungkyunkwan University and Yonsei University. Hyun Joong Im acknowledges Steve Bond’s sug-
gestions on this topic and insightful lectures on dynamic panel data analyses and GMM estimation methods provided during his doctoral studies at
the University of Oxford.
⁎
Corresponding author.
E-mail addresses: [email protected] (H.J. Im), [email protected] (Y. Kang), [email protected] (J. Shon).
1
For example, Bernanke (1983) and Bloom et al. (2007) show that uncertainty increases the value of real options to the firm, inducing it to wait
for additional information before investing.
https://doi.org/10.1016/j.jcorpfin.2020.101642
Received 4 December 2018; Received in revised form 28 February 2020; Accepted 10 May 2020
Available online 20 May 2020
0929-1199/ © 2020 Elsevier B.V. All rights reserved.
H.J. Im, et al. Journal of Corporate Finance 64 (2020) 101642
for financial intermediaries, who pass on these costs to the issuing firms in the form of higher underwriter spreads, thereby increasing
the issuance costs for new equity and debt capital, and in turn leading to lower leverage ratios.
However, the impact of business uncertainty, i.e. asset volatility, on capital structure decisions has been little explored. Standard
capital structure theories such as Leland (1994) and Lambrecht and Myers (2017) provide interesting testable predictions about the
relationship between asset volatility and optimal capital structure, and Choi (2013) and Choi and Richardson (2016) provide a way to
empirically estimate asset volatility. In addition, the dynamic panel regression estimators such as the difference generalised method
of moments (GMM) (Arellano and Bond, 1991) or the system GMM (Arellano and Bover, 1995; Blundell and Bond, 1998) allow us to
estimate the optimal leverage using a partial adjustment framework, the relative scarcity of study on this subject is quite puzzling.
Although actual leverage and optimal leverage tend to be highly positively correlated, target leverage estimated using a partial
adjustment framework is closer to the optimal leverage in theoretical literature such as Leland (1994) and Lambrecht and Myers
(2017).2 In this study, we examine the impact of asset volatility on optimal capital structure and investigate the economic me-
chanisms through which asset volatility influences optimal capital structure.
To capture operating uncertainty or business uncertainty independent of capital structure, we use the asset return volatility
measure proposed by Choi (2013) and Choi and Richardson (2016) as a main measure of uncertainty. Modigliani and Miller (1958)
suggest that the value of real assets can be represented by the value of financial assets so that the nature of a firm’s free cash flow
stream is unaffected by the way in which it is carved up. Thus, the return on a firm’s assets can be considered as a weighted average of
the return on each of the firm’s financial assets, where the weights are determined by the relative market value of each financial
asset.3 Choi and Richardson (2016) argue that the approach of viewing a firm’s assets as a portfolio of the firm’s individual securities
provides a distinct advantage over other approaches in the literature in that it does not rely on a specific model of capital structure.
This measure can attenuate endogeneity bias because asset returns, unlike stock returns, are highly likely to be determined by the
fundamental characteristics of businesses rather than the leverage level.4
To estimate optimal capital structure, we employ the partial adjustment model of capital structure which is often used to study the
speed of leverage adjustment. Dynamic trade-off models, such as that of Fischer et al. (1989), maintain that market imperfections
such as taxes and bankruptcy costs generate a link between capital structure and firm value; however, most of the time, firms allow
their leverage to diverge from their optimal leverage and only take actions to offset deviations from their optimal leverage if it gets
too far out of line. According to the survey by Graham and Harvey (2001), 81% of firms consider a target debt ratio or range when
making their capital structure decisions. The speed at which firms adjust towards target leverage ratios depends on the costs and
benefits of adjusting leverage. With zero adjustment costs, the dynamic trade-off theory implies that firms should stick to their
optimal leverage at all times. However, if adjustment costs are very high, firms are more likely to be reluctant to adjust towards their
optimal leverage. Flannery and Rangan (2006) propose a partial adjustment model in which firms partially or incompletely adjust
towards their optimal/target leverage ratios, which depend on firm characteristics.
Uncertainty influences optimal leverage ratios through various channels: debt tax shields, potential financial distress costs, the
agency costs of debt and the agency benefits of debt. Although the effects of uncertainty on target leverage ratios through potential
financial distress costs and shareholder–debtholder agency conflicts are expected to be negative, the impacts through debt tax shields
and agency benefits of debt can be either positive or negative. Furthermore, recent theory papers based on dynamic agency models
such as Lambrecht and Myers (2017) show that the negative effect of uncertainty on optimal leverage can be more pronounced if
managers are risk averse.5 Thus, whether uncertainty will increase or decrease optimal leverage ratios is an empirical question. In this
paper, we address this question by first identifying the directional effect of heightened uncertainty on leverage targets and then
considering through which mechanisms uncertainty affects optimal/target leverage ratios. The results of this study provide evidence
that uncertainty increases potential financial distress costs and exacerbates shareholder–debtholder conflicts (e.g. underinvestment
and risk-shifting problems), thereby leading to a lower optimal leverage ratio.
Endogeneity is the obvious problem that we must address in identifying the causal effect of uncertainty on optimal/target
leverage ratios. Although we use asset volatility as an uncertainty measure to address the reverse causality and measurement error
2
Few prior studies have been conducted on the relationship between stock return or cash flow volatility and actual leverage; however, the
empirical findings are mixed. For example, Kim and Sorensen (1986) report a positive relationship, whereas Bradley et al. (1984), Friend and Lang
(1988) and Keefe and Yaghoubi (2016) report a negative relationship between stock return or cash flow volatility and actual leverage. Frank and
Goyal (2009) also investigate which of the 25 explanatory variables used in prior studies (including stock return volatility) are reliably important to
a firm’s capital structure decision. They report that six factors, including firm size, market-to-book ratio, profitability and tangibility, reliably
influence a firm’s capital structure; however, they do not find a robust relationship between stock return variance and actual leverage.
3
Modigliani and Miller’s (1958) second proposition suggests that the corporate cost of capital is independent of the corporation’s capital structure.
In equilibrium, the return on assets should be equal to the corporate cost of capital so that according to Modigliani and Miller’s proposition, the
return on assets is independent of the firm’s capital structure. Hence, as the leverage ratio changes, the returns on debt and equity must alter to
ensure that the weighted average of the return on each of the firm’s financial assets is equal to the return on assets.
4
Many studies including Leahy and Whited (1996), Bloom et al. (2007), Bloom (2009), Kellogg (2014), Kim and Kung (2016), Bloom et al. (2018)
and Alfaro et al. (2018) use the standard deviation of daily stock returns as an uncertainty measure. Although this measure has some attractive
features, as discussed in Bloom et al. (2007), reverse causality remains a concern when studying the effect of stock return volatility on actual/
optimal capital structure. Stock returns, unlike asset returns, are influenced by both the leverage level and fundamental characteristics of businesses.
Thus, to mitigate the reverse causality concern, we use asset return volatility as our main uncertainty measure. Moreover, the measurement error
concern arising from ignoring debt volatility is mitigated by including the debt-side information.
5
Refer to Section 2 for a detailed discussion.
2
concerns that arise when using equity volatility, further endogeneity concerns remain. First, target leverage ratios may be influenced
by some important omitted factors (e.g. managerial risk aversion). Although we take into account the firm fixed effects in target
leverage to minimise the omitted variables bias and measurement errors arising from the omission of some important factors, there
may be some omitted unobservable, time-varying variables correlated with the included target determinants. Second, both asset
volatility and target/optimal capital structure are quite possibly driven by a third factor (e.g. the amount and nature of investment
opportunities and industry life cycles). Finally, the reverse causality problem may not completely disappear even when we use asset
volatility as an uncertainty measure. To address these endogeneity concerns, we use instrumental variables (IVs) suggested by Alfaro
et al. (2018). Our identification strategy exploits firms’ differential exposure to 10 aggregate sources of uncertainty shock, such as the
annual changes in the expected return volatilities of energy, currencies and 10-year treasury bonds. Using this approach, we confirm
our main findings.
This study makes a significant contribution to the dynamic capital structure literature. First, this paper proposes firm-level asset
volatility as a new crucial determinant of optimal capital structure. The asset volatility measure is in line with a volatility term (σ) in
most dynamic capital structure studies such as Leland (1994) and Lambrecht and Myers (2017). Most of the empirical dynamic
capital structure studies do not find that asset volatility plays a crucial role in explaining capital structure (Frank and Goyal, 2009;
Leary and Roberts, 2005; Antoniou et al., 2008). Thus, to the best of our knowledge, this study is the first in the literature to propose
firm-level asset volatility as a vital determinant of target/optimal leverage. We also show that the uncertainty effect on leverage
targets is greater than the impact of firm size, market-to-book, assets tangibility, R&D intensity and industry median leverage, making
uncertainty the most crucial determinant among all time-varying determinants of leverage targets.
Second, we explore various economic mechanisms through which asset volatility influences optimal capital structure. We employ
a partial adjustment model of capital structure to precisely estimate optimal capital structure. This allows us to evaluate the im-
portance of uncertainty in accounting for the variation in optimal capital structure. Using theory-based proxies for uncertainty and
optimal leverage ratios, we show that uncertainty significantly lowers optimal leverage ratios by decreasing debt tax shields, in-
creasing potential financial distress costs and exacerbating shareholder–debtholder conflicts (e.g. underinvestment and risk-shifting
problems). In addition, we show that firms with more risk averse managers have lower optimal leverage ratios, consistent with a
dynamic agency model (e.g. Lambrecht and Myers, 2017) predicting that the negative effect of uncertainty on optimal leverage can be
more pronounced if managers are risk averse.
The remainder of the paper is organised as follows. In Section 2, we present the theoretical predictions about the effects of
uncertainty on optimal/target capital structure. Section 3 describes the sample selection procedures, measurement of variables,
descriptive statistics and research design, and Section 4 presents empirical findings. Section 5 concludes the study.
2. Predictions from theory
Quite a few standard theories on optimal capital structure (e.g. Leland, 1994; Lambrecht and Myers, 2017) predict that optimal
leverage is significantly negatively affected by asset volatility, i.e. the volatility of rate of return on assets.6 In addition, recent theory
papers based on dynamic agency models such as Lambrecht and Myers (2017) predict that the negative effect of uncertainty on
optimal leverage can be more pronounced if managers are risk averse. Lambrecht and Myers (2017) show that the firm has an optimal
target leverage ratio, similar to the leverage target predicted by the trade-off theory of capital structure. Their model predicts a
significantly lower leverage level because of the manager’s desire to smooth rents and that higher managerial risk aversion leads to a
lower leverage ratio.7 Keefe and Yaghoubi (2016) employ the Black–Scholes model (Black and Scholes, 1973) to illustrate the
relationship between cash flow volatility and the cost of debt.8
In addition, the dynamic trade-off literature such as Fischer et al. (1989) and other related influential papers in capital structure
Harris and Raviv, 1991; Fama and French, 2002; Frank and Goyal, 2009) show that leverage targets are driven by several forces such
as tax shields, financial distress costs and agency costs and benefits related to debt. We extend Fama and French’s (2002) discussion of
the static/dynamic trade-off theory to derive some predictions regarding the potential channels through which uncertainty influences
target leverage ratios.9
First, the effects of uncertainty on optimal/target leverage ratios through debt tax shields can be negative or positive depending on
the magnitude of two conflicting effects. The negative impact is related to the impact of uncertainty on the volatility of earnings. A
high-uncertainty firm is more likely to have more volatile earnings than a low-uncertainty firm. As a result, a high-uncertainty firm is
6
Figures 7 and 13 in Leland’s (1994) paper show the significant negative effect of asset volatility of optimal leverage. In his paper, σ is the
standard deviation of the instantaneous rate of return on V, where V is the value of the unlevered firm assumed to be unaffected by the capital
structure of the firm. Thus, an appropriate proxy for σ is the asset volatility used as a main uncertainty measure in this study.
7
Figure 2 in their paper clearly shows that higher asset volatility leads to a lower optimal leverage and that higher relative risk aversion lowers
optimal leverage. In Section 4.4, we further discuss the moderating role of managerial risk aversion in the relationship between asset volatility and
optimal leverage and provide some tests related to the predictions.
8
In their model, σ is the standard deviation of the return on the asset. Thus, their model also provides support for using asset volatility as a main
uncertainty measure.
9
Different predictions are derived from the pecking order theory (Myers, 1984; Myers and Majluf, 1984) and market timing theory (Baker and
Wurgler, 2002). As Graham and Harvey (2001) report that 81% of firms consider a target debt ratio or range when making their capital structure
decisions, some firms’ capital structure policies may not be consistent with what the trade-off theory predicts. In Section 4.4, we discuss the
predictions from the two different theories and try to address some related concerns.
3
expected to have a higher chance of having no taxable income, and consequently, its expected future taxable income will be lower
and its expected payoff from interest tax shields will be lower. Thus, the effects of uncertainty on optimal leverage ratios through debt
tax shields can be negative. This reasoning is well in line with Blouin et al. (2010) who argue that the classic literature like Graham
(1996), Graham (2000) and Graham and Kim (2009) underestimates the future income volatility and thus overestimates the future
taxable income and the value of tax of debt.
The potential positive impact is derived from the impact of uncertainty on non-debt tax shields (DeAngelo and Masulis, 1980). A
high-uncertainty firm is expected to benefit less from non-debt tax shields because higher uncertainty leads to lower R&D ex-
penditures and lower future depreciation expenses, which are driven by the reduction in capital expenditures, as is evidenced by
Bloom et al. (2007) and Gulen and Ion (2015). In this case, it is possible that the firm benefits more from debt tax shields if the
amount of debt is given, in which case the effects of uncertainty on optimal leverage ratios through debt tax shields can be positive.
However, the reduction in debt tax shields owing to increased earnings volatility is likely to be the first-order effect, whereas the
increase in debt tax shields arising from the reduction in nondebt tax shields is most likely the second-order effect.
Second, the effects of uncertainty on optimal/target leverage ratios through potential financial distress costs are expected to be
negative. A high-uncertainty firm tends to have higher expected bankruptcy costs because it is expected to have a higher probability of
bankruptcy and meet higher direct and indirect bankruptcy costs given bankruptcy. A high-uncertainty firm is expected to have more
volatile earnings. Consequently, the probability of bankruptcy increases. This is well supported by Merton’s (1974) distance to default
(DD) model.10 When uncertainty is higher, indirect costs are expected to be higher, wherein suppliers may withdraw trade credits,
customers may turn to competitors and some key employees may leave firms.11 Thus, ceteris paribus, uncertainty is positively
associated with bankruptcy costs. Consequently, it has a negative effect on optimal leverage ratios.
Third, the effects of uncertainty on optimal/target leverage through debtholder–shareholder agency problems are predicted to be
negative. Agency problems such as assets substitution and under-investment arise when shareholders’ interests are not aligned with
debtholders’ interests (Fama and Miller, 1972; Jensen and Meckling, 1976; Myers, 1977). A high-uncertainty firm, compared with a
low-uncertainty firm, is expected to face more severe under-investment and asset substitution problems because high uncertainty will
make both assets in place and investment projects riskier. As a result, its debt will become riskier. Therefore, a high-uncertainty firm
has a stronger incentive to control shareholder–debtholder conflicts and will have a lower optimal leverage ratio.
Finally, the effects of uncertainty on optimal/target leverage through the agency benefits of debt can be either negative or positive.
Jensen (1986) shows that agency costs increase with free cash flows.12 However, debt may reduce the free cash flow agency problem
by ensuring that managers are disciplined, make efficient investment decisions and do not pursue private benefits, as this increases
bankruptcy risk (Jensen, 1986; Stulz, 1990). The direction of the effect of uncertainty on optimal leverage through agency benefits
arising from the disciplining role of debt depends on the composition of a firm’s free cash flow, i.e. its earnings from the assets in
place vs. the size of its profitable investments, given that a firm’s free cash flow is defined as its earnings from assets in place minus
the size of its profitable investments (Jensen, 1986).13
In summary, although the effects of uncertainty on optimal/target leverage ratios through potential financial distress costs and
shareholder–debtholder agency conflicts are expected to be negative, the effects through debt tax shields and agency benefits of debt
can be either negative or positive. Therefore, whether uncertainty will increase or decrease optimal/target leverage ratios is an open
question. Furthermore, the magnitude of the effect of uncertainty depends on managerial risk aversion. Thus, this study aims to
understand the magnitude of uncertainty’s effect on a firm’s optimal/target capital structure and the underlying mechanisms through
which uncertainty lowers optimal capital structure.
3. Empirical framework
3.1. Sample selection
Our empirical analysis uses the CRSP/Compustat Merged (CCM) database for annual financial statements data, the Center for Research in
Security Prices (CRSP) database for monthly stock return data, the Trade Reporting and Compliance Engine (TRACE) database for monthly
bond pricing data, the Loan Syndications and Trading Association (LSTA) Mark-to-Market Pricing database for monthly loan sale pricing data,
10
Although Bharath and Shumway (2008) find that Merton’s DD model does not provide sufficient statistics for the probability of default, they
conclude that its functional form is still useful for forecasting defaults.
11
The indirect costs of financial distress—identified as the reduction in valuable capital expenditures, losses of critical customers, losses of
important suppliers, etc.—are known to be much larger than the direct costs of financial distress (Andrade and Kaplan, 1998).
12
In the agency models of Jensen and Meckling (1976), Easterbrook (1984), Jensen (1986) and Stulz (1990), the interests of managers are not
aligned with those of shareholders, and managers tend to waste free cash flows on perquisites such as corporate jets, plush offices and building
empires as well as on bad investments.
13
We extended Fama and French’s (2002) framework to come up with the predictions. If a firm has many profitable assets in place, the prof-
itability of its assets in place will be lower when it is faced with high uncertainty. For the firm, high uncertainty will lead to less free cash flows and
therefore lower shareholder–manager agency costs. Thus, high uncertainty will lower the value of debt as a disciplining device to the firm, implying
that the effect of uncertainty on the optimal leverage ratio will be negative. However, if a firm has many profitable investment opportunities, a high-
uncertainty firm will have a higher value of the real option to wait (Bloom et al., 2007; Gulen and Ion, 2015), leading to more free cash flows and
therefore higher shareholder–manager agency costs. Thus, it is possible that a high-uncertainty firm could receive more benefits from the dis-
ciplining role of debt.
4
the Compustat database for credit rating data and the Bureau of Economic Analysis (BEA) database for GDP deflator data from 2002 to 2018.
Our sample period starts from 2003 because bond return data is available in the TRACE database only from 2002.14
To construct our final sample, we go through the following sample selection and data cleaning procedures. The dataset covers all
manufacturing firms with the two-digit North American Industry Classification System (NAICS) sector codes of 31, 32 or 33.15 We
require that each firm has at least 10 years of uninterrupted observations. We exclude firms with missing or negative total assets or
negative book equity and the firms whose stock is not traded on the three major stock exchanges in the U.S. (i.e. NYSE, NASDAQ, and
AMEX). We retain the firm-year observation if variables other than total assets and book equity are missing. The final sample is an
unbalanced panel dataset comprising 14,546 firm-year observations corresponding to 1,455 firms.
3.2. Measurement of variables
3.2.1. Measuring uncertainty

As the main uncertainty proxy, we use asset volatility (i.e. volatility of the return on assets) proposed by Choi (2013) and Choi and
Richardson (2016). Specifically, asset volatility (σi,t) is defined as the sample standard deviation of monthly return on assets (RA) in a
given year, where the monthly return on assets is defined as follows.
E B L
RA = RE + RB + RL ,
E+B+L E+B+L E+B+L (1)
E B L
where R , R and R denote the monthly returns on equity, bonds and bank loans, respectively, and E, B and L are market values of
equity, bonds and loans, respectively.16 That is, asset returns are defined as weighted averages of returns on a firm’s various sources
of financing, including equity, loans and bonds (Choi, 2013; Choi and Richardson, 2016). The weights given to different claims are
determined by their market values.
The monthly total returns, i.e. returns taking dividends into consideration, and market value of equity are from the CRSP database. The
monthly bond returns are obtained from the TRACE database and are computed as the value-weighted averages of returns on individual
bonds issued by a given firm. Individual bond returns are calculated by considering the transaction prices, coupon rates and accrued interests
at the end of each month.17 After the initiation of a syndicated loan in the primary market, a bank decides whether to keep it on its balance
sheet or sell it off in the secondary market. LSTA collects more than 80% of all U.S. trading volume of the secondary loan market on a daily
basis according to Drucker and Puri (2008), allowing us to calibrate the loan returns. However, the following two issues make the calibration
complex: (1) of all syndicated loans, only around 20% are traded in the secondary market (Drucker and Puri, 2008); and (2) our loan sale data
covers the period of 2003–2013 whereas our sample period is up to 2018. Treating both loans and bonds as contingent claims on the same
underlying assets, in the spirit of Choi (2013), we estimate loan returns using bond returns at monthly frequency for firm-month observations
without loan sale data. The idea is to approximate the hypothetical transaction prices assuming that those loans are traded. A detailed account
of the estimation of loan returns can be found in the Appendix.
Asset volatility is an appealing uncertainty measure for the following reasons. First, it is a forward-looking uncertainty measure that
implicitly incorporates future business prospects and changes in business environments. Insofar as asset returns reflect the prospect of firms’
future performances and business environments reasonably well, we expect the impact of different sources of business uncertainty to be
adequately incorporated into the asset returns. Therefore, asset return volatility correctly weighs the relative effect of different sources of
business uncertainty on the firm value.18 Earnings volatility and cash flows volatility are also reasonable proxies for business uncertainty, but
they are not as good as asset volatility because those measures are not forward-looking measures and the low frequency of earnings and cash
flows, i.e. yearly or quarterly, makes exploiting within-firm temporal variations difficult.
Second, unlike equity return volatility that is directly affected by financing decisions, asset return volatility is not directly affected
by leverage.19 Because this study attempts to examine the causal effect of uncertainty on target/optimal capital structure, using an
uncertainty measure that is not directly influenced by capital structure is important. If we use equity return volatility as the
14
We use the TRACE database for monthly bond pricing data because the TRACE database incorporates all over-the-counter transactions on
corporate bonds and has been widely used in prior research (e.g. Edwards et al., 2007). In addition, the TRACE database has an advantage in that it
provides transaction prices rather than dealer quotes as in other databases such as the Bridge EJV (now Thomson Reuters Pricing Service) database.
15
By focusing on the manufacturing sector, one can rule out unobserved heterogeneity across sectors or industries that cannot be captured by
industry fixed effects. In addition, as reported in Flannery and Rangan’s (2006) study, it was not possible to find a dynamic panel regression model
that satisfies the Sargan–Hansen test of over-identifying restrictions when we use the sample covering all the sectors. If the validity of the in-
struments is not satisfied, one cannot make a reliable inference based on the target leverage ratio estimated using the dynamic panel regression
model.
16
As the market value of loans is not readily available, we use the book value of debt minus the market value of bonds to approximate the loan
value. Choi (2013) reports that loans and bonds constitute about 94% of the book value of debt, suggesting that it is reasonable to approximate the
loan value by subtracting market value of bonds from book debt.
17
Dick-Nielsen (2009, 2014) provides guidance on how to appropriately clean the TRACE bond pricing data. The method involves taking into
account the change in data structure in 2012, removing all corrections and cancellations related to the data and deleting agency–customer
transactions without commissions to avoid any duplication.
18
Note that realised (or implied) stock return volatility shares this attribute (Bloom et al., 2007). However, the realised (or implied) stock return
volatility is a noisier proxy for a firm’s overall business uncertainty.
19
A number of studies (e.g. Christie, 1982; Bhandari, 1988; George and Hwang, 2010) suggest that leverage ratios proxying distress risk affect
stock returns and variances.
5
uncertainty measure, reverse causality prevents us from identifying the causal effect of uncertainty on target leverage. Although
several ways to obtain asset returns (asset betas) from equity returns (equity betas) have been proposed (e.g. Hamada, 1972; Miles
and Ezzell, 1985), Choi’s (2013) and Choi and Richardson’s (2016) approach is more straightforward, given that the authors directly
calculate asset returns instead of estimating asset returns using assumptions about the capital structure policy.20
Finally, asset volatility varies across firms and over time, allowing us to evaluate if the target leverage ratios of high-uncertainty firms
are significantly different from those of low-uncertainty firms. Given the stylised facts that there is substantial unobserved heterogeneity
across firms’ leverage ratios (Lemmon et al., 2008) and that some determinants of firms’ leverage ratios, such as financial distress costs, are
influenced by macroeconomic conditions (Chen, 2010), an uncertainty measure with both cross-sectional and temporal variations is more
suitable than an uncertainty measure with only cross-sectional variation (e.g. annual earnings volatility) or an uncertainty measure with
only temporal variation (e.g. macroeconomic uncertainty measures).21
3.2.2. Measuring control variables

Following the dynamic capital structure literature, such as Fama and French (2002), Flannery and Rangan (2006), Faulkender
et al. (2012) and Elsas et al. (2014), we control for a vector of firm and industry characteristics that may affect a firm’s target/optimal
capital structure. All variables are computed for firm i over its fiscal year t. In the dynamic panel data regressions used to estimate
target leverage, the control variables include the firm size measured by the natural logarithm of book total assets denominated in
year-2000 dollars; market-to-book ratio defined as a ratio of the sum of the book value of debt and the market value of equity to the
book value of total assets; profitability measured by the ratio of earnings before interests and taxes (EBIT) to total assets; assets
tangibility measured by the ratio of net property, plant and equipment (PP&E) to total assets; depreciation and amortisation mea-
sured by the ratio of depreciation and amortisation expenses to total assets; R&D intensity measured by research and development (R
&D) expenses as a proportion of total assets; a zero R&D firm indicator defined as a dummy variable for zero R&D expenses; and
industry median book (or market) leverage ratios based on Fama and French’s (1997) 48 industries. Detailed definitions of the
variables used in this study are provided in Panel A, Table 1.
3.3. Descriptive statistics
To minimise the effect of outliers, we winsorize all ratio variables at the top and bottom 1% of each variable’s distribution. Panel B in
Table 1 provides the summary statistics for the main variables used in this study. On average, a firm in the final sample has a book (market)
leverage ratio of 17.1% (15.2%) and a book (market) target leverage ratio of 18.5% (16.1%).22 The uncertainty measure, asset volatility (σi,t),
has a mean value of 0.099 and a median value of 0.088. Panel B in Table 2 also reports the summary statistics for the control variables. In the
sample, an average firm has book total assets of US$517 million, a market-to-book ratio of 1.68, EBIT scaled by total assets of 2.0%, PP&E
scaled by total assets of 20.0%, depreciation expenses scaled by total assets of 3.8% and R&D expenses scaled by total assets of 6.6%.
Before investigating the effect of uncertainty on target leverage ratios, we examine whether actual leverage ratios are associated with
uncertainty. Assuming that uncertainty is an important factor in determining target leverage ratios and that firms actually manage their
leverage towards target leverage, actual leverage ratios should be correlated with targets. Thus, as a preliminary analysis, we sort firms based
on uncertainty and test whether uncertainty is correlated with actual leverage ratios.23 Panel A of Table 2 reports the summary statistics for
actual leverage ratios for high- and low-uncertainty firms. A firm-year observation is categorised into either high- or low-uncertainty group
depending on whether its asset volatility is above the sample median or equal to or below the sample median. The table shows that high-
uncertainty firms tend to have substantially lower book and market leverage ratios than low-uncertainty firms. Panel A shows that the mean
(median) difference in actual book leverage ratios between high-uncertainty and low-uncertainty firms is −10.9% (−16.3%), whereas the
mean (median) difference in actual market leverage ratios is −9.3% (−12.5%).24 Using Student’s t-test and Wilcoxon’s rank-sum test, we
show that the mean and median differences are statistically significant at the 1% level. The differences are also economically significant,
especially given that we define high-uncertainty and low-uncertainty firms based on the median of our uncertainty measure.
In addition, we examine whether the target leverage ratios estimated following the procedure used in Faulkender et al. (2012) are
associated with uncertainty. In particular, we follow the procedure described in Section 3.4; however, we do not include asset
20
Hamada (1972) calculates a delevering formula with the Modigliani–Miller assumptions, under which the level of corporate debt is exogenously
fixed. In contrast, Miles and Ezzell (1985) derive a delevering formula with the assumption that leverage ratios remain constant.
21
We further investigate the relationship between asset volatility and macroeconomic uncertainty in Section 4.4.
22
Note that the mean market and book leverage ratios are lower than those reported in other papers (e.g. Flannery and Rangan (2006): 24.9% and
27.8% for book and market leverage ratios, respectively; Frank and Goyal (2009): 29% and 28% for book and market leverage ratios, respectively;
and Faulkender et al. (2012): 25.3% and 27.6% for book and market leverage ratios, respectively). The main reason for the differences is that this
study focuses on the manufacturing sector. We find that manufacturing firms have lower leverage ratios compared with most non-manufacturing
firms in the same sample period. For example, mean book (market) leverage ratio of firms in mining, utilities and construction industries is 30.2%
(35.4%), which is much higher than that of manufacturing firms. When we consider all firms from the CRSP/Compustat Merged (CCM) database for
the same sample period, the mean book and market leverage ratios are 22.7% and 24.6%, respectively, which are very close to those reported in the
previous literature. In addition, there is a weakly declining trend in leverage ratios in the recent period. Note that our sample period is from 2003 to
2018 owing to the availability of some data required to calculate our main uncertainty measure, i.e. asset volatility.
23
We sincerely appreciate an anonymous referee’s suggestion on this analysis.
24
The difference in means (medians) is calculated as the mean (median) of high-uncertainty firms minus the mean (median) of low-uncertainty
firms.
6
Table 1
Variable definitions and summary statistics.
Panel A. Variable definitions
Definition Calculation
Leverage-related variables
Book leverage Total debt ([dltt]+[dlc]) over book total assets ([at])
Market leverage Total debt ([dltt]+[dlc]) over market value of total assets ([dltt]+[dlc]+[cshpri]×[prcc_f])
Book target Target book leverage ratio estimated using system GMM results in Table 3
Market target Target market leverage ratio estimated using system GMM results in Table 4
Uncertainty-related variables
Asset volatility Standard deviation of a firm’s monthly asset returns for each fiscal year, where asset returns are calculated following Choi and
Richardson (2016)
Control variables
Firm size Natural logarithm of total assets denominated in year-2000 dollars
Market-to-book ratio Market total assets ([dlc]+[dltt]+[cshrpi]×[prcc_f]) to book total assets ([at])
Profitability Earnings before interests and taxes ([ib]+[xint]+[txt]) over total assets ([at])
Tangibility Total property, plant and equipment net of accumulated depreciation ([ppent]) over total assets ([at])
Depreciation Depreciation and amortisation expenses ([dp]) over total assets ([at])
R&D intensity R&D expenses ([xrd]) over total assets ([at]) (0 if missing)
Zero R&D firm indicator Dummy variable, which equals one if a firm does not report R&D expenses in year t, and zero otherwise.
Industry median book leverage Industry median book leverage, where the industry is defined following Fama and French (1997)
Industry median market leverage Industry median market leverage, where the industry is defined following Fama and French (1997)
Panel B. Summary statistics
Variables Obs. Mean S.D. P05 P25 Median P75 P95
Leverage-related variables
Book leverage 14,546 0.171 0.157 0.000 0.009 0.152 0.278 0.465
Market leverage 14,546 0.153 0.168 0.000 0.005 0.106 0.232 0.506
Book target 14,546 0.185 0.150 0.000 0.054 0.169 0.278 0.462
Market target 14,546 0.161 0.156 0.000 0.041 0.123 0.233 0.485
Uncertainty-related variables
Asset volatility 14,546 0.099 0.053 0.033 0.059 0.088 0.129 0.205
Control variables
Firm size 14,546 6.249 2.045 3.028 4.741 6.184 7.662 9.789
Market-to-book ratio 14,546 1.684 1.287 0.540 0.897 1.307 1.991 4.133
Profitability 14,546 0.020 0.201 −0.395 −0.005 0.069 0.121 0.221
Tangibility 14,546 0.200 0.150 0.022 0.088 0.163 0.277 0.511
Depreciation 14,546 0.038 0.021 0.010 0.024 0.034 0.047 0.077
R&D intensity 14,546 0.066 0.103 0.000 0.004 0.027 0.084 0.272
Zero R&D firm indicator 14,546 0.206 0.405 0.000 0.000 0.000 0.000 1.000
Industry median book leverage 14,546 0.146 0.083 0.031 0.080 0.146 0.205 0.283
Industry median market leverage 14,546 0.117 0.087 0.013 0.045 0.099 0.174 0.295
Panel A shows the definitions of the variables used in this study. The italicised codes in square brackets represent item codes in the CRSP/Compustat
Merged database. Panel B shows the summary statistics for the variables used in this study. The variables are constructed using a sample of U.S.
public firms in the manufacturing industry from 2003 to 2018. The sample comprises firms that have at least 10 years of uninterrupted observations.
All ratio variables are winsorized at the 1st and 99th percentiles.
volatility as a regressor to estimate target leverage ratios. Panel B of Table 2 shows that the mean (median) difference in book target
leverage ratios between high-uncertainty and low-uncertainty firms is −8.8% (−12.4%), whereas the mean (median) difference in
market target leverage ratios between high-uncertainty and low-uncertainty firms is −5.9% (−7.1%). Using Student’s t-test and
Wilcoxon’s rank-sum test, we show that the mean and median differences are statistically significant at the 1% level. The differences
are also economically significant, given that the mean book and market target leverage ratios are 18.5% and 18.7%, respectively.
3.4. Research design
To investigate the effect of uncertainty on long-run leverage targets, we extend Flannery and Rangan’s (2006) partial adjustment
framework as stated below:
Li , t Li, t 1 = (Li, t Li, t 1) + t + i, t , (2)
where Li,t is firm i’s current leverage, Li,t⋆
is firm i’s target leverage ratio, κt is an error component reflecting year fixed effects and υi,t is
a white-noise error term. Li,t − Li,t−1 measures the actual change in leverage, or leverage adjustment, and Li,t⋆ − Li,t−1 measures the
7
Table 2
Comparison of actual and target leverage ratios by the level of uncertainty.
Panel A. Comparison of actual leverage ratios by the level of uncertainty
Actual book leverage of all firms 14,546 0.171 0.157 0.000 0.009 0.152 0.278 0.465
High-uncertainty firms 7,273 0.117 0.142 0.000 0.000 0.061 0.197 0.405
Low-uncertainty firms 7,273 0.226 0.153 0.000 0.111 0.224 0.326 0.497
Difference (High–Low) −0.109⁎⁎⁎ −0.163⁎⁎⁎
t-stat/z-stat 44.57 44.67
Actual market leverage of all firms 14,546 0.153 0.168 0.000 0.005 0.106 0.232 0.506
Difference (High–Low) −0.093⁎⁎⁎ −0.125⁎⁎⁎
Panel B. Comparison of target leverage ratios estimated without uncertainty by the level of uncertainty
Book targets estimated without 14,546 0.185 0.151 0.000 0.052 0.168 0.278 0.463
asset volatility
Difference (High−Low) −0.088⁎⁎⁎ −0.124⁎⁎⁎
Market targets estimated without 14,546 0.157 0.154 0.000 0.040 0.120 0.227 0.480
asset volatility
This table reports the summary statistics for actual and target leverage ratios for high- and low-uncertainty firms in Panels A and B, respectively. Each panel
reports the summary statistics for actual or target leverage ratios in the full sample and those for high- and low-uncertainty firms. In Panel B, target leverage
ratios are estimated following the procedure used in Faulkender et al. (2012). In particular, we follow the procedure described in Section 3.4; however, we do
not include asset volatility as a regressor to estimate target leverage ratios. A firm-year observation is classified into a subsample of high-uncertainty firms (low-
uncertainty firms) if asset volatility is above median (is equal to or below median). The differences in mean and median leverage ratios between the two groups
are reported with the t-statistic of Student’s t test or z-statistic of Wilcoxon’s rank-sum test. Superscript ⁎⁎⁎ indicates statistical significance at the 1% level.
deviation from the target leverage ratio. The speed of adjustment parameter, λ, measures how fast a typical firm’s actual leverage
adjusts to its target leverage. The parameter is expected to lie between 0 and 1 with a higher λ indicating a faster speed of adjustment.
Each year, a typical firm closes a proportion λ of the gap between where it stands (Li,t−1) and where it hopes to be (Li,t⋆). As leverage
measures (Li,t), we consider both the book leverage ratio (BLi,t) and market leverage ratio (MLi,t).
To estimate target leverage ratios, we assume that a firm’s target leverage (Li,t⋆) is a linear function of various firm characteristics
(Xi,t−1) with firm fixed effects (ηi⋆) included.
Li, t = + Xi, t 1 + i (3)
Xi,t−1 includes a firm-level uncertainty measure (σi,t) described in Section 3 as well as a set of firm and industry characteristics
used in recent dynamic capital structure studies including Fama and French (2002), Flannery and Rangan (2006), Antoniou et al.
(2008), Faulkender et al. (2012) and Elsas et al. (2014). The variables include firm size, market-to-book ratio, profitability, assets
tangibility, depreciation and amortisation, R&D intensity, a zero R&D firm indicator, and industry median leverage ratios. Unlike the
existing literature, we explicitly assume that there is unobserved heterogeneity in target leverage.25 This study proposes a method to
estimate fixed effects in target leverage and evaluate the importance of the fixed effects in the variation of target leverage. Panel A in
Table 1 presents the definitions and summary statistics for the variables used in this study.
Substituting the target leverage equation into Eq. (2), we obtain the following model:
Li , t = + (1 ) Li , t 1 + X i, t 1 + t + i + i, t , (4)
where ληi⋆
and κt represent firm fixed effects and year fixed effects in actual leverage, respectively. Eq. (2) can be rewritten as the
following standard dynamic panel regression model, which will serve as our main econometric framework:
25
Most of the existing studies assume that there is unobserved heterogeneity in actual leverage and are silent about how they estimate the fixed
effects in target leverage. Among others, Im (2019) uses an approach very similar to our approach.
8
Li, t = b0 + b1 Li, t 1 + b2 Xi, t 1 + t + i + i, t , (5)
where b0 = λα, b1 = (1−λ), b2 = λβ, and ηi = ληi⋆. We include year dummies to control for year fixed effects (κt).26 The speed of adjustment
can be estimated as = 1 b1. After obtaining , it is straightforward to obtain , , i and target leverage estimates (Li, t ) using Eq. (3).27
We employ four econometric methodologies to ensure that the estimated effect of uncertainty on target leverage is not attributable to the
choice of estimation methods or instrument sets, although more weight will be given to system GMM and least squares dummy variables with
a bias correction (LSDVC) results. Considering the increasingly recognised important role in the estimation of dynamic panel data models in
corporate finance research, there is a need to resolve several estimation issues arising from fixed effects and lagged dependent variables. For
instance, the ordinary least squares (OLS) and within groups (WG) estimates of the coefficient of the lagged dependent variable tend to be
biased upwards and downwards, respectively. This situation is particularly true when the data have short panel length (Nickell, 1981; Bond,
2002). Therefore, the OLS or WG estimates of the coefficients of Xi,t−1 in Eqs. (4) and (5) are also likely to be biased.
Using simulated panel data, Flannery and Hankins (2013) show that the estimation performance of various econometric meth-
odologies substantially varies depending on data complications, such as fixed effects, the persistence of the dependent variable,
endogenous independent variables and error term autocorrelations. Flannery and Hankins (2013) find that the LSDVC estimator
proposed by Bruno (2005) performs the best in the absence of endogenous independent variables, whereas the system GMM estimator
(Arellano and Bover, 1995; Blundell and Bond, 1998) appears to be the best choice in the presence of endogenous independent
variables and even second-order serial correlation if the dataset includes shorter panels. Thus, we employ both system GMM and
LSDVC estimators as well as OLS and WG estimators.
4. Empirical results
4.1. The effect of uncertainty on long-run leverage targets
To examine whether uncertainty increases or decreases a typical firm’s target leverage ratio, we first estimate the dynamic panel
regression model specified in Eq. (4). Tables 3 and 4 present the estimation results for book and market leverage ratios as the
dependent variable, respectively. The four columns in each panel report the estimation results based on OLS, WG, LSDVC and System
GMM estimators. We include firm fixed effects to control for unobserved time-invariant firm-specific characteristics in all estimation
methods save OLS whereas we incorporate year fixed effects to account for temporal variations in all four specifications. System GMM
appears to perform slightly better than LSDVC because i) the goodness-of-fit scores of the system GMM models (0.761 and 0.729 in
Tables 3 and 4, respectively) are higher than those of LSDVC models (0.759 and 0.700 in Tables 3 and 4, respectively) and ii) LSDVC
estimates are reported to be the most accurate only in the absence of endogenous independent variables. Therefore, we use the system
GMM estimators of Blundell and Bond (1998) for target leverage estimation and other analyses in the rest of the paper. The estimated
target book leverage ratio and target market leverage ratio are denoted as BL and ML , respectively.
As the results based on market and book leverage ratios are qualitatively similar to each other, the following discussion will be
based on the results in Table 3. As predicted by Nickell (1981) and Bond (2002), the coefficients of the lagged dependent variable
GMM LSDVC OLS
estimated by system GMM (b1 = 0.770 ) and LSDVC (b1 = 0.742 ) comfortably fall between the OLS (b1 = 0.847 ) and WG
WG
(b1 = 0.609 ) estimates.28 System GMM results in Column (4) of Table 3 indicate that the overall book adjustment speed is ap-
proximately 23.0% per annum. The sensitivity of book targets to uncertainty is estimated to be −0.441, with statistical significance
at the 1% level. That is, a one-standard-deviation increase in uncertainty leads to a decrease in the book target leverage of 15.6% of
its one standard deviation.
We have similar results in Table 4 for market leverage models.29 The system GMM estimate of overall market adjustment speed is
approximately 24.2%, which is comparable to the literature. The system GMM estimate of the target-uncertainty sensitivity remains
significant at the 1% level, with the sensitivity being −0.485—suggesting that a one-standard-deviation increase in uncertainty leads
to a decrease in the market target leverage of 16.5% of its one standard deviation.
As discussed in Section 2, the effects of uncertainty through debt tax shields and agency benefits of debt cannot be determined a
26
If one replaces year fixed effects with year dummies, caution is required. To restore , one needs to adjust b0 by adding a constant to ensure
that the mean of year effects estimated using year dummies is zero. The adjusted b0 , or b0 , should be equal to .
27
Given the residual of the regression (i.e. it = i + i,t ), the fixed effects in actual leverage ( i ) can be estimated by calculating
within-firm average residuals. The fixed effects in target leverage ( i ) can be estimated by dividing the fixed effects in actual leverage
( i ) by the speed of adjustment estimate ( ).
28
The GMM-style instruments used in Column (4) include the fourth and all available further lags of a leverage measure (BL and ML in Tables 3 and 4,
respectively), the second to tenth lags of uncertainty and the second to tenth lags of all control variables for first-difference equations. Moreover, the third
lags of the change in leverage and the first lags of the changes in uncertainty and all control variables are used as instruments for level equations. The
Sargan–Hansen test of over-identifying restrictions does not reject this specification (p-value = 0.229), which supports the validity of our choice of
instruments. Arellano and Bond’s (1991) serial correlation tests find no significant evidence of the second-order serial correlation in the first-differenced
residuals (p-value = 0.860). Some additional lags are available as instruments for the model for book leverage, but we use the same lags as instruments as
the model for market leverage. However, the models with additional lags produce almost identical results.
29
The Sargan–Hansen test of over-identifying restrictions does not reject this specification (p-value = 0.118), which supports the validity of our
choice of instruments. Arellano and Bond’s (1991) serial correlation tests find significant evidence of the first- to third-order serial correlation but do
not find any significant evidence of the fourth-order serial correlation in the first-differenced residuals (p-value = 0.542).
9
Table 3
Estimation of the sensitivity of book leverage targets to uncertainty.
Dependent variable: Book leverage
(1) (2) (3) (4)
Estimation method OLS WG LSDVC System GMM
Lagged book leverage 0.847 ⁎⁎⁎

0.609 ⁎⁎⁎
0.742 ⁎⁎⁎
0.770⁎⁎⁎
(0.007) (0.011) (0.008) (0.019)
Asset volatility −0.082⁎⁎⁎ −0.080⁎⁎⁎ −0.061⁎⁎⁎ −0.101⁎⁎⁎
(0.017) (0.021) (0.019) (0.030)
Firm size 0.003⁎⁎⁎ 0.010⁎⁎⁎ 0.007⁎⁎⁎ 0.006⁎⁎⁎
(0.000) (0.002) (0.002) (0.002)
Market-to-book ratio 0.001⁎ 0.001 0.001 0.003⁎
(0.001) (0.001) (0.001) (0.001)
Profitability −0.019⁎⁎⁎ −0.024⁎⁎⁎ −0.020⁎⁎⁎ −0.021⁎
(0.006) (0.009) (0.006) (0.011)
Tangibility 0.009 0.019 0.013 0.026
(0.006) (0.018) (0.014) (0.022)
Depreciation −0.037 −0.239⁎⁎⁎ −0.260⁎⁎⁎ −0.394⁎⁎⁎
(0.042) (0.081) (0.065) (0.112)
R&D intensity −0.009 −0.013 −0.008 −0.006
(0.013) (0.026) (0.018) (0.030)
Zero R&D firm indicator 0.005⁎⁎ 0.014⁎⁎ 0.013⁎⁎ 0.009
(0.002) (0.006) (0.005) (0.008)
Industry median book leverage 0.028⁎⁎⁎ 0.043⁎⁎ 0.032 0.042
(0.010) (0.022) (0.023) (0.027)
Firm fixed effects No Yes Yes Yes
Year fixed effects Yes Yes Yes Yes
Number of observations 14,546 14,546 14,547 14,546
Number of firms 1,455 1,455 1,455 1,455
Goodness of fit score 0.765 0.747 0.759 0.761
m1 −16.93
(p-value) (0.000)
m2 −0.177
(p-value) (0.860)
Sargan–Hansen 1049
(p-value) (0.229)
Speed of adjustment ( ) 0.153⁎⁎⁎ 0.391⁎⁎⁎ 0.258⁎⁎⁎ 0.230⁎⁎⁎
(0.007) (0.011) (0.008) (0.019)
Target-uncertainty sensitivity ( ) −0.537⁎⁎⁎ −0.203⁎⁎⁎ −0.235⁎⁎⁎ −0.441⁎⁎⁎
(0.105) (0.054) (0.071) (0.131)
This table reports the results of the book target leverage estimation regressions using the OLS, WG, LSDVC and system GMM estimators. The
empirical model used is as follows: BLi,t = Constant + (1−λ)BLi,t−1 + λβ′Xi,t−1 + Year fixed effects + ηi + υi,t. The dependent variable is book
leverage (BL). A detailed description of the variables included in the models is provided in Panel A of Table 1. In the OLS and WG estimators,
standard errors are clustered by firm and displayed in parentheses below. In the LSDVC models, the Blundell-Bond estimator is chosen as an initial
estimator, and bootstrapped standard errors are reported. In the system GMM, we report two-step GMM coefficients and standard errors that are
asymptotically robust to both heteroskedasticity and serial correlation and that use the finite-sample correction proposed by Windmeijer (2005).
The IVs used in system GMM reported in Column (4) are the second to tenth lags of standardised uncertainty, the fourth to all available lags of
leverage and the second to tenth lags of firm-specific control variables for the equations in first-differences, as well as the change in standardised
uncertainty, the third lag of change in leverage and the first lag of change in all firm-specific control variables for level equations. Note that year
dummies are treated as instruments for the equations in levels only. m1 and m2 represent the test statistics of the Arellano–Bond tests for first-order
and second-order serial correlations in first-differenced residuals, respectively. Sargan–Hansen represents the test statistic of the Sargan–Hansen test
of over-identifying restrictions. Overall goodness-of-fit scores, measured as the square of the coefficient of correlation between the dependent
variable and its predicted value, are reported for OLS, WG, LSDVC and system GMM. Superscripts ⁎, ⁎⁎ and ⁎⁎⁎ indicate statistical significance at the
10%, 5% and 1% levels, respectively.
priori, although the effects of uncertainty through potential financial distress costs and debtholder–shareholder agency conflicts are
expected to be negative. The ultimate impact of uncertainty on target leverage, therefore, depends on which forces dominate. When
the sensitivity of a firm’s target leverage ratios to uncertainty is negative and both economically and statistically significant based on
both system GMM and LSDVC estimates, this suggests that either the effects of uncertainty through potential financial distress costs
and debtholder–shareholder agency conflicts offset the opposite effects through debt tax shields and disciplining role of debt, or the
effects of uncertainty through debt tax shields and/or agency benefits of debt are also played in the direction of lowering target
leverage ratios. We further analyse the mechanisms through which uncertainty decreases target leverage ratios in Section 4.3.
Having provided evidence about the marginal effects of uncertainty on target leverage ratios, we proceed to gain further insight
by comparing the target leverage ratios between high-uncertainty firms and low-uncertainty firms. Consistent with the analyses of the
marginal effects of uncertainty on target leverage ratios, high-uncertainty firms tend to have significantly lower target leverage ratios
10
Table 4
Estimation of the sensitivity of market leverage targets to uncertainty.
Dependent variable: Market leverage
(1) (2) (3) (4)
Estimation method OLS WG LSDVC System GMM
Lagged market leverage 0.815 ⁎⁎⁎

0.542 ⁎⁎⁎
0.664 ⁎⁎⁎
0.758⁎⁎⁎
(0.008) (0.013) (0.009) (0.019)
Asset volatility −0.062⁎⁎⁎ −0.093⁎⁎⁎ −0.077⁎⁎⁎ −0.118⁎⁎⁎
(0.020) (0.025) (0.021) (0.036)
Firm size 0.002⁎⁎⁎ 0.017⁎⁎⁎ 0.015⁎⁎⁎ 0.001
(0.000) (0.003) (0.002) (0.002)
Market-to-book ratio −0.002⁎⁎⁎ −0.002⁎ −0.001 −0.003⁎⁎⁎
(0.001) (0.001) (0.001) (0.001)
Profitability −0.011⁎ −0.017⁎ −0.011⁎ 0.002
(0.006) (0.009) (0.007) (0.013)
Tangibility 0.022⁎⁎⁎ 0.048⁎⁎ 0.042⁎⁎⁎ 0.059⁎⁎
(0.008) (0.019) (0.016) (0.024)
Depreciation −0.060 −0.182⁎ −0.215⁎⁎⁎ −0.349⁎⁎⁎
(0.048) (0.098) (0.073) (0.124)
R&D intensity −0.022⁎ 0.014 0.021 −0.047⁎
(0.012) (0.028) (0.020) (0.027)
Zero R&D firm indicator 0.011⁎⁎⁎ 0.013 0.010⁎ 0.013
(0.003) (0.008) (0.006) (0.009)
Industry median market leverage 0.009 0.053⁎⁎ 0.033 −0.047⁎
(0.012) (0.023) (0.022) (0.024)
Firm fixed effects No Yes Yes Yes
Number of firms 1,455 1,455 1,455 1,455
m1 −17.19
(p-value) (0.000)
m2 −4.939
(p-value) (0.000)
m3 6.120
(p-value) (0.000)
m4 −0.609
(p-value) (0.542)
Sargan–Hansen 1070
(p-value) (0.118)
(0.008) (0.013) (0.009) (0.019)
Target-uncertainty sensitivity ( ) −0.336⁎⁎⁎ −0.203⁎⁎⁎ −0.228⁎⁎⁎ −0.485⁎⁎⁎
(0.107) (0.054) (0.062) (0.152)
This table reports the results of the market target leverage estimation regressions using the OLS, WG, LSDVC and system GMM estimators. The
empirical model used is as follows: MLi,t = Constant + (1−λ)MLi,t−1 + λβ′Xi,t−1 + Year fixed effects + ηi + υi,t. The dependent variable is market
leverage (ML). A detailed description of the variables included in the models is provided in Panel A of Table 1. In the OLS and WG estimators,
standard errors are clustered by firm and displayed in parentheses below. In the LSDVC models, the Blundell-Bond estimator is chosen as an initial
estimator, and bootstrapped standard errors are reported. In the system GMM, we report two-step GMM coefficients and standard errors that are
asymptotically robust to both heteroskedasticity and serial correlation and that use the finite-sample correction proposed by Windmeijer (2005).
The IVs used in system GMM reported in Column (4) are the second to tenth lags of standardised uncertainty, the fourth to all available lags of
leverage and the second to tenth lags of firm-specific control variables for the equations in first-differences, as well as the change in standardised
uncertainty, the third lag of change in leverage and the first lag of change in all firm-specific control variables for level equations. Note that year
dummies are treated as instruments for the equations in levels only. m1, m2, m3 and m4 represent the test statistics of the Arellano–Bond tests for
first-order to fourth-order serial correlations in first-differenced residuals, respectively. Sargan–Hansen represents the test statistic of the Sar-
gan–Hansen test of over-identifying restrictions. Overall goodness-of-fit scores, measured as the square of the coefficient of correlation between the
dependent variable and its predicted value, are reported for OLS, WG, LSDVC and system GMM. Superscripts ⁎, ⁎⁎ and ⁎⁎⁎ indicate statistical
significance at the 10%, 5% and 1% levels, respectively.
than low-uncertainty firms. Panel A in Table 5 shows that the mean (median) difference in book target leverage ratios between high-
uncertainty and low-uncertainty firms is −10.1 (−13.7) percentage points, whereas Panel B in Table 5 shows that the mean
(median) difference in market target leverage ratios is −8.1 (−9.4) percentage points. Using Student’s t-test and Wilcoxon’s rank-
sum test, we show that the mean and median differences in both book targets and market targets are statistically significant at the 1%
level. The differences are also economically significant, especially given that we define high-uncertainty and low-uncertainty firms
based on the median of our uncertainty measure. The differences are even more significant if terciles are used to group firms as high-
uncertainty and low-uncertainty firms. This observation confirms our main results regarding the negative sensitivity of leverage
11
Table 5
Comparison of target leverage ratios by the level of uncertainty.
Panel A. Book target leverage
Book targets with firm fixed effects 14,546 0.185 0.150 0.000 0.054 0.169 0.278 0.462
Firm fixed effects in book targets 14,546 0.007 0.137 −0.169 −0.090 −0.017 0.075 0.257
Proportion of fixed effects in targets 10,039 0.361 0.245 0.028 0.160 0.320 0.530 0.826
Panel B. Market target leverage
Market targets with firm fixed effects 14,546 0.161 0.156 0.000 0.041 0.123 0.233 0.485
Firm fixed effects in market targets 14,546 0.007 0.144 −0.162 −0.087 −0.025 0.065 0.291
Proportion of fixed effects in targets 9,339 0.406 0.248 0.049 0.203 0.386 0.582 0.862
This table reports the summary statistics for book and market target leverage ratios in Panel A and Panel B, respectively. To estimate
target leverage ratios, we first obtain the system GMM estimates for the following dynamic panel regression model:
Li, t = + (1 ) L i, t 1 + Xi, t 1 + Year fixed effects + i + i,t , where ληi⋆ represents firm fixed effects in actual leverage (Li,t). Estimation re-
sults are reported in Tables 3 and 4. We then obtain , , and i following the procedure detailed in Section 3.4. Finally, we obtain book and
market target leverage estimates using the following equation: Li, t = + X i, t 1 + i . Each panel reports the summary statistics for target
leverage ratios including firm fixed effects (Li, t ) in the full sample and those for high- and low-uncertainty firms. A firm-year observation is classified
into a sub-sample of high-uncertainty firms (low-uncertainty firms) if asset volatility is above median (is equal to or below median). The differences
in mean and median target leverage ratios between the two groups are reported with the t-statistic of Student’s t test or z-statistic of Wilcoxon’s rank-
sum test. Superscript ⁎⁎⁎ indicates statistical significance at the 1% level. Each panel also reports the summary statistics for firm fixed effects in target
leverage ratios ( i ) and the proportion of firm fixed effects in target leverage ratios ( i / Li, t ).
targets to uncertainty reported in Tables 3 and 4.

Moreover, we examine the contribution made by firm fixed effects on the leverage targets. That is, to calculate the contribution
made by firm fixed effects, we decompose target leverage estimates with firm fixed effects into i) firm fixed effects ( i ) and ii)
leverage targets net of the effects. Table 5 reports the summary statistics for firm fixed effects in book and market targets and their
proportions in book and market targets. It shows that the mean (median) proportion of firm fixed effects on the target leverage ratios
is 36.1% (32.0%) for book targets and 40.6% (38.6%) for market targets. Thus, firm fixed effects constitute significant parts in both
book and market target leverage ratios. Table 5 also shows that firm fixed effects, whether positive or negative, are consistently
prominent across the distribution. The 5th and 95th percentiles of the firm fixed effects in book targets (market targets) are −16.9%
(−16.2%) and 25.7% (29.1%), respectively.
Furthermore, we implement the analysis of covariance (ANCOVA) to further examine the relative importance of various determi-
nants in capturing the variation in the ‘target’ leverage ratios estimated following the procedure detailed in Section 3.4, in contrast with
Lemmon et al.’s (2008) study which performs the ANCOVA for ‘actual’ leverage ratios. Table 6 presents the results of the variance
decompositions for several specifications. Each column in the table corresponds to a different model specification for target leverage
ratios. The numbers in the body of the table, excluding the last two rows, correspond to the contribution of each variable in a particular
model. That is, following Lemmon et al. (2008), we measure the contribution of each variable by dividing the partial sum of squares for
each effect by the aggregate partial sum of squares across all effects in the model so that the columns sum to one.
The results are summarised as follows. First, the ANCOVA results reported in the last columns of Panels A and B in Table 6 show
that the total variation of both book and market targets explained by uncertainty is 0.8%. This result suggests that the uncertainty
effects in both book and market targets are greater than the effects of firm size, market-to-book, profitability, assets tangibility, R&D
intensity, a zero R&D firm indicator and industry median leverage. In particular, uncertainty is the most important determinant of
market target leverage ratios among all time-varying determinants. Second, the ANCOVA results show that time-invariant firm-
specific effects are the major source of the total variation of leverage targets. The total variation of market targets explained by all
time-varying determinants is only 2.8%, whereas the total variation of book targets explained by all time-varying determinants is
only 3.2%. In other words, 97.2% and 96.8% of the total variations in market and book targets are explained by time-invariant firm
fixed effects, respectively. Intuitively, this finding suggests that much of the explanatory power of existing target leverage de-
terminants comes from the cross-sectional, as opposed to time-series, variation.
12
Table 6
Variance decomposition of target leverage.
Panel A. Book target leverage
(1) (2) (3) (4) (5) (6)
Variables BL BL BL BL BL BL
Firm fixed effects 0.987 0.989 0.965 0.968

Year fixed effects 0.022 0.003 0.025 0.000
Asset volatility 0.013 0.978 0.008 0.009 0.265 0.008
Firm size 0.004 0.374 0.003
Market-to-book ratio 0.003 0.014 0.003
Profitability 0.003 0.184 0.003
Tangibility 0.001 0.010 0.001
Depreciation 0.011 0.002 0.011
R&D intensity 0.000 0.011 0.000
Zero R&D firm indicator 0.001 0.015 0.001
Industry median book leverage 0.002 0.100 0.001
Number of observations 14,546 14,546 14,546 14,546 14,546 14,546
Adjusted R-squared 0.974 0.134 0.977 0.996 0.233 0.996
Panel B. Market target leverage
(1) (2) (3) (4) (5) (6)
Variables ML ML ML ML ML ML
Firm fixed effects 0.989 0.990 0.970 0.972

Year fixed effects 0.079 0.000 0.057 0.000
Asset volatility 0.011 0.921 0.009 0.009 0.196 0.008
Firm size 0.000 0.060 0.000
Market-to-book ratio 0.003 0.257 0.003
Profitability 0.000 0.193 0.000
Tangibility 0.005 0.048 0.004
Depreciation 0.007 0.006 0.007
R&D intensity 0.001 0.065 0.001
Zero R&D firm indicator 0.002 0.108 0.002
Industry median market 0.002 0.009 0.002
leverage
Adjusted R-squared 0.978 0.089 0.979 0.996 0.228 0.996
This table presents the variance decomposition results for the two leverage targets based on several different ANCOVA model specifications. The
ANCOVA results for book and market leverage targets are summarised in Panels A and B, respectively. To estimate target leverage ratios, we first
obtain system GMM estimates for the following dynamic panel regression model: Li, t = + (1 ) L i, t 1 + Xi, t 1 + Year fixed effects + i + i,t ,
where ληi⋆ represents firm fixed effects in actual leverage (Li,t). Estimation results are reported in Tables 3 and 4. We then obtain , , and i
following the procedure detailed in Section 3.4. Finally, we obtain book and market target leverage estimates using the following equation:
Li, t = + X i, t 1 + i . A detailed description of the variables is provided in Panel A of Table 1. The estimated target book and market leverage
ratios are denoted as BLi, t and MLi, t , respectively. The numbers in the body of the table, excluding the last two rows, correspond to the contribution
of each variable in a particular model. That is, we measure the contribution of each variable by dividing the partial sum of squares for each effect by
the aggregate partial sum of squares across all effects in the model so that the columns sum to one.
4.2. Mitigating potential endogeneity concerns
As discussed in the introduction, endogeneity is the obvious problem that we must address in identifying the directional effect of
uncertainty on target/optimal capital structure. Although we use asset volatility as an uncertainty measure to address the reverse
causality and measurement error concerns that arise when using equity volatility (as discussed in Section 3.2), further endogeneity
concerns remain. First, target leverage ratios may be influenced by some important omitted factors (e.g. managerial risk aversion).
Although we take into account the firm fixed effects in target leverage to minimise the omitted variables bias and measurement errors
arising from the omission of some important factors, there may be some omitted unobservable time-varying variables correlated with
the included target determinants. Second, both asset volatility and target/optimal capital structure are quite possibly driven by a
third factor (e.g. the amount and nature of investment opportunities and industry life cycles). Finally, the reverse causality problem
may not completely disappear even when we use asset volatility as an uncertainty measure.
To address these endogeneity concerns, we use IVs suggested by Alfaro et al. (2018). Our identification strategy exploits firms’
differential exposure to 10 aggregate sources of uncertainty shock, including the annual changes in the implied return volatilities of
13
currencies, energy and 10-year treasury bonds and the annual change in the realised volatility of the economic policy uncertainty (EPU)
index.30
We employ the first-differenced two-stage least squares (2SLS) estimator for the AR(1) panel data model proposed by Anderson
and Hsiao (1981, 1982). The second-stage regression model is specified as follows:
Li, t = constant + (1 ) Li , t 1 + i, t 1 + Ci , t 1 + i, t , (6)

where i, t 1 is the instrumented uncertainty shock, Li, t 1 is the instrumented lagged change in leverage and ΔCi,t−1 is a vector
containing the first-differences of all target determinants except asset volatility.
To address the endogeneity concerns related to the lagged leverage change (ΔLi,t−1), we instrument ΔLi,t−1 with the lagged
leverage level Li,t−2.31 Suppose that we have the following AR(1) panel data model:
Li , t = + (1 ) Li , t 1 + i, t 1 + C i, t 1 + i + i, t , (7)
where σi,t−1 is asset volatility and Ci,t−1 is a vector containing all the target determinants except asset volatility. The first-differencing
transformation eliminates the individual effects from the following model:
Li, t = (1 ) Li, t 1 + i, t 1 + Ci , t 1 + i, t . (8)
The dependence of Δυi,t on υi,t−1 implies that the OLS estimates of (1−λ) in the first-differenced model are not consistent.
However, a consistent estimate of (1−λ) can now be obtained using 2SLS with IVs that are both correlated with ΔLi,t−1 and or-
thogonal to Δυi,t. Together with the assumption that the disturbances υi,t are serially uncorrelated, the predetermined initial condi-
tions imply that the lagged level Li,t−2 will be uncorrelated with υi,t and thus be available as an IV for the first-differenced equation
(Bond, 2002). Thus, we instrument ΔLi,t−1 with the lagged leverage level Li,t−2.32
To address the endogeneity concerns related to the uncertainty shock (Δσi,t−1), we instrument Δσi,t−1 with the lagged industry-
level (i.e. SIC 3-digit) non-directional exposure to 10 aggregate sources of uncertainty shocks. Following Alfaro et al. (2018), we also
control for the lagged directional exposure to the 10 aggregate uncertainty shocks.33
Table 7 reports the 2SLS regression results based on book and market leverage ratios in Columns (1) through (3) and in Columns
(4) through (6), respectively. Columns (1) and (2) (Columns (4) and (5)) report the first-stage regression results for a model for the
first-difference in book (market) leverage. We test for weak instruments in each specification. The reported Wald F-statistics based on
Kleibergen and Paap (2006) are higher than the Stock-Yogo critical value (i.e. 5% maximal IV relative bias), indicating that the
included instruments are not weak instruments at the 5% significance level.
The second-stage results in Columns (3) and (6) are consistent with the system GMM results reported in Tables 3 and 4, re-
spectively. The results show that the instrumented uncertainty shock has negative effects on the first-difference in book leverage
(ΔBLi,t−1) and the first-difference in market leverage (ΔMLi,t−1) with significance at the 1% level. The estimated sensitivity of target
leverage to uncertainty ( ) is −2.192 for book targets and −4.309 for market targets, and both sensitivities are statistically sig-
nificant at the 1% level. The estimated sensitivities for book targets (market targets) are five times (nine times) larger than those
obtained from the system GMM estimation of the dynamic panel regression model, as specified in Eq. (4).34 Using IV regressions, we
find that high-uncertainty firms have significantly lower leverage targets than their low-uncertainty counterparts. Therefore, the
analysis in this section allays our major endogeneity concerns regarding the main results reported in Section 4.1.
4.3. Mechanisms through which uncertainty lowers long-run leverage targets
In Section 2, we have identified four potential mechanisms through which uncertainty affects optimal/target leverage ratios, i.e.
debt tax shields, financial distress costs, agency costs of debt and agency benefits of debt. The effects of uncertainty on leverage
targets through financial distress costs and debtholder–shareholder agency conflicts are expected to be negative, but the effects
through debt tax shields and agency benefits of debt can be either negative or positive. Thus, we have investigated whether un-
certainty increases or decreases leverage targets in Section 4.1, finding that uncertainty lowers optimal/target leverage ratios. In this
30
Alfaro et al. (2018) use the following factors to estimate the industry-level sensitivities to the 10 aggregate uncertainty shocks. As currency
factors, they use the growth in the exchange rates of the Federal Reserve Board’s seven major currencies: Australian Dollar (AUD), Japanese Yen
(JPY), Canadian Dollar (CAD), Swiss Franc (CHF), British Pound (GBP), Swedish Krona (SEK) and Euro (EUR). As the energy factor, they use the
growth in crude oil prices. As the treasury bond factor, they use the return on the US 10-year treasury bond. As the EPU factor, they use the growth
in the EPU index proposed by Baker et al. (2016). Refer to Alfaro et al. (2018) for details on how to construct non-directional and directional
exposure to the 10 aggregate uncertainty shocks. The data for the exposure to the 10 aggregate sources of uncertainty shock are available at the
following website: https://www.policyuncertainty.com/firm_uncertainty.html.
31
We are grateful to an anonymous referee for suggesting this. The second-stage control variables are also included in the first-stage regressions.
32
The difference GMM (Arellano and Bond, 1991) and system GMM (Arellano and Bover, 1995; Blundell and Bond, 1998) estimators are more
efficient estimators; however, the endogeneity concerns related to the uncertainty shock (Δσi,t−1) cannot be straightforwardly addressed using the
GMM estimators.
33
The first-differenced model in a 2SLS framework is estimated because the uncertainty shock (Δσi,t−1), rather than the uncertainty level (σi,t−1),
is instrumented using the IVs suggested by Alfaro et al. (2018).
34
The estimates of the overall book and market adjustment speeds are similar to those obtained from the system GMM estimation of the dynamic
panel regression model, as specified in Eq. (4).
14
subsection, we examine through which mechanisms uncertainty lowers leverage targets. The results are presented in Tables 8 and 9.
4.3.1. Through debt tax shields

The effects of uncertainty on target leverage ratios through debt tax shields can be negative or positive depending on the magnitude
of two conflicting effects. The negative effect is directly related to the impact of uncertainty on debt tax shields. A high-uncertainty
firm that has lower and more volatile earnings is expected to have a higher chance of having no taxable income. Consequently, its
expected tax rate and expected payoff from interest tax shields will be lower. Thus, uncertainty can lower target leverage ratios by
lowering debt tax shields. The positive effect is derived from the impact of uncertainty on non-debt tax shields. A high-uncertainty
firm that has lower R&D expenditures and depreciation expenses is expected to benefit more from debt tax shields if the amount of
debt is given, in which case uncertainty can raise target leverage ratios by increasing debt tax shields.
Given that we find evidence that uncertainty lowers target leverage ratios, we investigate whether some of the effects of un-
certainty can be attributed to debt tax shields. A simple way to test this is to see if uncertainty reduces the present value of tax shields
Table 7
The effect of uncertainty on target leverage ratios—IV regressions.
Change in book leverage Change in market leverage
(1) (2) (3) (4) (5) (6)
Variables Δσi,t−1 ΔLi,t−1 ΔLi,t Δσi,t−1 ΔLi,t−1 ΔLi,t
i, t 1
−0.542 ⁎⁎⁎
−1.632⁎⁎⁎
(0.101) (0.207)
Li, t 1
0.753⁎⁎⁎ 0.621⁎⁎⁎
(0.038) (0.051)
Exposure to ΔσAUD 0.204⁎⁎ 0.100 0.155⁎ 0.229
(0.087) (0.111) (0.086) (0.155)
JPY
Exposure to Δσ 0.504⁎⁎⁎ 0.110 0.455⁎⁎⁎ 0.300
(0.161) (0.157) (0.159) (0.243)
Exposure to ΔσCAD 0.656⁎⁎⁎ 0.083 0.623⁎⁎⁎ 0.224
(0.087) (0.195) (0.086) (0.280)
Exposure to ΔσCHF 0.141⁎⁎⁎ 0.061 0.128⁎⁎ 0.204⁎
(0.051) (0.074) (0.051) (0.119)
Exposure to ΔσGBP 0.116 0.244 0.114 0.078
(0.093) (0.238) (0.095) (0.288)
Exposure to ΔσSEK 0.591⁎⁎⁎ −0.085 0.570⁎⁎⁎ 0.288⁎⁎
(0.066) (0.079) (0.067) (0.127)
Exposure to ΔσEUR 0.108⁎⁎⁎ 0.000 0.103⁎⁎⁎ 0.156⁎⁎
(0.038) (0.034) (0.038) (0.074)
Exposure to ΔσEPU 26.667 75.777 −15.610 192.272⁎
(23.377) (50.282) (29.004) (116.804)
Exposure to ΔσOil 0.386⁎⁎⁎ 0.141⁎ 0.306⁎⁎⁎ 0.702⁎⁎⁎
(0.101) (0.082) (0.102) (0.228)
Exposure to ΔσTreasury 0.000⁎⁎⁎ 0.000 0.000⁎⁎⁎ 0.000⁎⁎
(0.000) (0.000) (0.000) (0.000)
Li,t−2 0.011⁎⁎⁎ −0.101⁎⁎⁎ 0.016⁎⁎⁎ −0.114⁎⁎⁎
(0.001) (0.009) (0.002) (0.006)
ΔFirm size −0.006⁎⁎⁎ 0.087⁎⁎⁎ −0.066⁎⁎⁎ −0.004⁎⁎ 0.093⁎⁎⁎ −0.056⁎⁎⁎
(0.002) (0.014) (0.014) (0.002) (0.014) (0.013)
ΔMarket-to-book ratio 0.003⁎⁎⁎ −0.003⁎⁎ 0.004⁎⁎⁎ 0.004⁎⁎⁎ −0.015⁎⁎⁎ 0.018⁎⁎⁎
(0.001) (0.001) (0.001) (0.001) (0.003) (0.004)
ΔProfitability 0.005 −0.124⁎⁎⁎ 0.102⁎⁎⁎ 0.005 −0.175⁎⁎⁎ 0.147⁎⁎⁎
(0.004) (0.019) (0.019) (0.003) (0.029) (0.033)
ΔTangibility −0.015 0.083⁎⁎⁎ 0.003 −0.017 0.126⁎⁎⁎ −0.055
(0.016) (0.027) (0.035) (0.015) (0.023) (0.042)
ΔDepreciation 0.043 0.074 −0.314 0.037 0.203⁎ −0.218
(0.057) (0.129) (0.238) (0.057) (0.106) (0.155)
ΔR&D intensity −0.031⁎⁎⁎ 0.052⁎⁎ −0.011 −0.029⁎⁎⁎ −0.073⁎⁎⁎ 0.060⁎
(0.007) (0.021) (0.035) (0.007) (0.022) (0.034)
ΔZero R&D firm indicator −0.008⁎⁎ 0.021⁎⁎ −0.011 −0.007⁎⁎ 0.013 −0.025⁎
(0.004) (0.010) (0.011) (0.004) (0.012) (0.013)
ΔIndustry median leverage 0.049⁎⁎⁎ 0.323⁎⁎⁎ −0.196⁎⁎⁎ 0.090⁎⁎⁎ 0.530⁎⁎⁎ −0.249⁎⁎⁎
(0.019) (0.028) (0.043) (0.017) (0.037) (0.057)
Kleibergen–Paap Wald F-statistic 23.07 54.07
(critical value at 5%) (20.54) (20.54)
Speed of adjustment ( ) 0.247⁎⁎⁎ 0.379⁎⁎⁎
(0.038) (0.051)
(continued on next page)
15
Table 7 (continued)
Change in book leverage Change in market leverage
(1) (2) (3) (4) (5) (6)
Variables Δσi,t−1 ΔLi,t−1 ΔLi,t Δσi,t−1 ΔLi,t−1 ΔLi,t
Target-uncertainty sensitivity ( ) −2.192 ⁎⁎⁎

−4.309⁎⁎⁎
(0.549) (0.631)
This table reports the results of the 2SLS regression analyses designed to examine the effect of uncertainty on target leverage ratios. The second-stage
model is specified as follows: Li, t = constant + (1 ) Li, t 1 + i, t 1 + Ci, t 1 + i, t , where i, t 1 is the instrumented uncertainty shock,
Li, t 1 is the instrumented lagged change in leverage and ΔCi,t−1 is a vector containing the first-differences of all target determinants except asset
volatility. To address the endogeneity concerns related to the lagged leverage change (ΔLi,t−1), we instrument ΔLi,t−1 with the lagged leverage level Li,t−2.
To address the endogeneity concerns related to the uncertainty shock (Δσi,t−1), we use IVs suggested by Alfaro et al. (2018). In particular, we instrument
the uncertainty shock (Δσi,t−1) with the lagged industry-level (i.e. SIC 3-digit) non-directional exposure to 10 aggregate sources of uncertainty shocks, i.e.
the annual changes in the implied return volatilities of currencies, energy and 10-year treasury bonds and the annual change in the realised volatility of
the EPU index. When we estimate the industry-level sensitivities to the 10 shocks, we use the following factors. As currency factors, we use the growth in
the exchange rates of the Federal Reserve Board’s seven major currencies: Australian Dollar (AUD), Japanese Yen (JPY), Canadian Dollar (CAD), Swiss
Franc (CHF), British Pound (GBP), Swedish Krona (SEK) and Euro (EUR). As the energy factor, we use the growth in crude-oil prices. As the treasury bond
factor, we use the return on the U.S. 10-year treasury bond. As the EPU factor, we use the growth in the EPU index proposed by Baker et al. (2016).
Following Alfaro et al. (2018), we also control for the lagged directional exposure to the 10 aggregate uncertainty shocks. Refer to Alfaro et al. (2018) for
details on how to construct non-directional and directional exposure to the 10 aggregate uncertainty shocks. The table also reports the Kleibergen–Paap
Wald F-statistic for a weak identification test. The Stock–Yogo critical value for the F-test (i.e. 5% maximal IV relative bias) is also reported. A detailed
description of other control variables included in the models is provided in Panel A of Table 1. Standard errors are clustered at industry level. Superscripts
, and ⁎⁎⁎ indicate statistical significance at the 10%, 5% and 1% levels, respectively.
⁎ ⁎⁎
(Stage 1) and if the reduction in the present value of tax shields lowers target leverage (Stage 2), using a 2SLS procedure.35 To
measure the present value of debt tax shields, we first calculate tax rate, τ, as corporate tax payments divided by pre-tax income.
Under the assumption of a perpetual debt, the present value of tax shields is computed as tax-deductible debt (the sum of long-term
debt and short-term debt) multiplied by the tax rate (τ). We then scale the present value of tax shields by total assets to obtain a
measure of debt tax shields, TXSHLDi,t−1.
The first two columns in Tables 8 and 9 report the results of 2SLS analyses for book leverage targets and market leverage targets,
respectively. The first-stage regression results show that heightened uncertainty (σi,t−1) reduces the present value of debt tax shields
(TXSHLDi,t) with significance at the 1% level. The second-stage regression results show that a higher present value of tax shields
(TXSHLDi, t ) leads to a higher target leverage ratio (BLi, t and MLi, t ) with significance at the 1% level. The results provide evidence
that high uncertainty leads to a lower optimal/target leverage ratio by decreasing the present value of debt tax shields.
4.3.2. Through financial distress costs

As a firm faced with higher uncertainty tends to have higher expected bankruptcy costs, the effects of uncertainty on target
leverage ratios through potential financial distress costs are expected to be negative. Given that we find evidence that uncertainty
lowers target leverage ratios, we further investigate whether some of the effects of uncertainty can be attributed to financial distress
costs. Again, we employ a 2SLS procedure to examine if uncertainty increases bankruptcy costs (Stage 1) and if the increase in
bankruptcy costs lowers target leverage (Stage 2). To implement this test, we employ the modified Altman’s Z-score (MZ) proposed by
Graham et al. (1998) and used by Chava et al. (2008) and Im (2012).36
Columns (3) and (4) in Tables 8 and 9 report the results of 2SLS analyses designed to test the financial distress costs channel for
book leverage targets and market leverage targets, respectively. The first-stage regression results show that heightened uncertainty
(σi,t−1) lowers the modified Altman’s Z-score (MZi,t) (i.e. increases financial distress costs) with significance at the 1% level. The
second-stage regression results show that a higher modified Altman’s Z-score (MZi, t ) leads to a higher target book leverage ratio (BLi, t
35
We employ the 2SLS procedure by assuming that the present value of tax shields is endogenous but asset volatility is exogenous. The first-stage
regression can verify whether asset volatility increases or decreases the present value of tax shields, whereas the second-stage regression can verify
whether the estimated present value of tax shields increases or decreases target leverage.
36
The modified Altman’s Z-score that is inversely related to bankruptcy costs is calculated as follows: MZ = 1.2X1 + 1.4X2 + 3.3X3 + X4, where
X1 is the ratio of working capital to total assets, X2 is the ratio of retained earnings to total assets, X3 is the ratio of earnings before interests and taxes
to total assets and X4 is the ratio of sales to total assets. The modified Altman’s Z-score which does not include leverage is appropriate as we use it to
predict target/optimal leverage.
16
Table 8
Mechanisms through which uncertainty affects book leverage targets.
Tax Shields Bankruptcy costs Financial constraints Agency costs Agency benefits
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Variables TXSHLDi,t MZi,t SAi,t VIOLYRi,t SUMBLKSi,t

BLi,t BLi,t BLi,t BLi,t BLi,t
σi,t−1 −0.300 ⁎⁎⁎

−23.867 ⁎⁎⁎
3.044 ⁎⁎⁎
0.428 ⁎⁎⁎
−0.071
(0.020) (1.917) (0.178) (0.123) (0.059)
TXSHLDi, t 3.660⁎⁎⁎
(0.250)
MZi,t 0.044⁎⁎⁎
(0.004)
SAi, t −0.352⁎⁎⁎
(0.028)
VIOLYRi, t −2.067⁎⁎⁎
(0.631)
SUMBLKSi,t 16.744
(14.071)
Industry fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Number of observations 11,738 11,738 12,766 12,766 12,851 12,851 4,053 4,053 5,421 5,421
This table reports the results of the 2SLS analyses used to examine the mechanisms through which uncertainty affects book leverage targets. To
estimate book target leverage ratios, we first obtain system GMM estimates for the following dynamic panel regression model:
BLi, t = + (1 ) BLi, t 1 + X i,t 1 + Year fixed effects + i + i, t , where ληi⋆ represents firm fixed effects in actual book leverage. The esti-
mation result is reported in Column (4) of Table 3. We then obtain , , and i following the procedure detailed in Section 3.4. Finally, we obtain
book target leverage estimates using the following equation: BLi, t = + X i, t 1 + i . TXSHLD is the value of tax shields scaled by total assets,
where the value of the tax shield is estimated as the tax-deductible debt (i.e. the sum of long-term debt and short-term debt) multiplied by the
corporate tax rate. MZ is the modified Z-score proposed by Graham et al. (1998) and used by Chava et al. (2008) and Im (2012), which is calculated
as follows: Z = 1.2X1 + 1.4X2 + 3.3X3 + X4, where X1 is the ratio of working capital to total assets, X2 is the ratio of retained earnings to total
assets, X3 is the ratio of earnings before interests and taxes to total assets and X4 is the ratio of sales to total assets. SA is a financial constraints index
proposed by Hadlock and Pierce (2010). VIOLYR is a proxy for the likelihood of covenant violation proposed by Nini et al. (2009), which equals one
if a firm reports a loan covenant violation in an SEC 10-K or 10-Q filing for a given year and zero otherwise. SUMBLKS is the percentage of shares
held by all blockholders. Firm-clustered standard errors are reported in parentheses. Superscripts ⁎, ⁎⁎ and ⁎⁎⁎ denote statistical significance at the
and MLi, t ) with significance the 1% level.37 The results provide evidence that heightened uncertainty leads to a lower optimal/target
leverage ratio by increasing financial distress costs and exacerbating financial constraints.
4.3.3. Through agency costs of debt

The effects of uncertainty on target leverage ratios through agency costs of debt are expected to be negative. A high-uncertainty
firm faces more severe under-investment and asset substitution problems because uncertainty makes both assets in place and in-
vestment projects riskier. In response to the exacerbated agency conflicts, a high-uncertainty firm is expected to choose a lower
optimal/target leverage ratio. Given that this study finds evidence that uncertainty lowers target leverage ratios, we further in-
vestigate whether some of the effects of uncertainty can be attributed to the exacerbated debtholder–shareholder conflicts. To in-
vestigate whether uncertainty increases agency costs of debt and whether the increase in agency costs lowers target leverage, we
employ a 2SLS procedure. As a proxy for the debtholder–shareholder agency conflicts, we utilise the likelihood of covenant violation
following Nini et al. (2009). To measure the likelihood of covenant violation VIOLYR, we use a dummy variable that equals one if a
firm reports a loan covenant violation in an SEC 10-K or 10-Q filing for a given year and zero otherwise.38
Columns (7) and (8) in Tables 8 and 9 report the results of 2SLS analyses for book leverage targets and market leverage targets,
respectively. The first-stage regression results show that heightened uncertainty (σi,t−1) increases the likelihood of covenant violation
37
Following an anonymous referee’s suggestion, we consider Hadlock and Pierce’s (2010) financial constraints index (a.k.a. SA index) as another
proxy of financial distress costs. Although financial constraints and financial distress are two different concepts, a more financially constrained firm
is clearly more likely to have a higher probability of bankruptcy. We obtain similar results using the SA index, and they are reported in Columns (5)
and (6) of Tables 8 and 9. The first-stage regression result shows that heightened uncertainty (σi,t−1) increases the SA index (SAi,t) (i.e. exacerbates
financial constraints) with significance at the 1% level. The second-stage regression result shows that a higher SA index (SAi,t) leads to a lower target
leverage ratio (BLi,t⋆ and MLi,t⋆) with significance at the 1% level.
38
Data regarding loan covenant violations are available at Amir Sufi’s website: http://faculty.chicagobooth.edu/amir.sufi/data.html. We con-
struct an annualised loan covenant violation measure by modifying an original quarterly measure. The firm-year specific loan covenant violation
measure equals one if there is at least one violation across four quarters in that year and zero otherwise.
17
Table 9
Mechanisms through which uncertainty affects market leverage targets.
Tax shields Bankruptcy costs Financial constraints Agency costs Agency benefits
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Variables TXSHLDi,t MZi,t SAi,t VIOLYRi,t SUMBLKSi,t

MLi, t MLi, t MLi, t MLi, t MLi, t
σi,t−1 −0.300 ⁎⁎⁎

−23.867 ⁎⁎⁎
3.044 ⁎⁎⁎
0.428 ⁎⁎⁎
−0.071
(0.020) (1.917) (0.178) (0.123) (0.059)
TXSHLDi, t 2.657⁎⁎⁎
(0.239)
MZi,t 0.033⁎⁎⁎
(0.004)
SAi, t −0.263⁎⁎⁎
(0.026)
VIOLYRi, t −1.692⁎⁎⁎
(0.550)
SUMBLKSi,t 13.038
(10.899)
Industry fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Number of observations 11,738 11,738 12,766 12,766 12,851 12,851 4,053 4,053 5,421 5,421
This table reports the results of the 2SLS analyses used to examine the mechanisms through which uncertainty affects market leverage targets. To
estimate market target leverage ratios, we first obtain system GMM estimates for the following dynamic panel regression model:
MLi, t = + (1 ) MLi, t 1 + X i,t 1 + Year fixed effects + i + i, t , where ληi⋆ represents firm fixed effects in actual market leverage. The es-
timation result is reported in Column (4) of Table 4. We then obtain , , and i following the procedure detailed in Section 3.4. Finally, we
obtain market target leverage estimates using the following equation: MLi, t = + Xi, t 1 + i . TXSHLD is the value of tax shields scaled by total
assets, where the value of the tax shield is estimated as the tax-deductible debt (i.e. the sum of long-term debt and short-term debt) multiplied by the
corporate tax rate. MZ is the modified Z-score proposed by Graham et al. (1998) and used by Chava et al. (2008) and Im (2012), which is calculated
as follows: Z = 1.2X1 + 1.4X2 + 3.3X3 + X4, where X1 is the ratio of working capital to total assets, X2 is the ratio of retained earnings to total
assets, X3 is the ratio of earnings before interests and taxes to total assets and X4 is the ratio of sales to total assets. SA is a financial constraints index
proposed by Hadlock and Pierce (2010). VIOLYR is a proxy for the likelihood of covenant violation proposed by Nini et al. (2009), which equals one
if a firm reports a loan covenant violation in an SEC 10-K or 10-Q filing for a given year and zero otherwise. SUMBLKS is the percentage of shares
held by all blockholders. Firm-clustered standard errors are reported in parentheses. Superscripts ⁎, ⁎⁎ and ⁎⁎⁎ denote statistical significance at the
(VIOLYRi,t) with significance at the 1% level. The second-stage regression results show that a higher likelihood of covenant violation
(VIOLYRi, t ) leads to a lower target leverage ratio (BLi, t and MLi, t ) with significance at the 1% level. The results provide evidence that
heightened uncertainty leads to a lower optimal/target leverage ratio by exacerbating debtholder–shareholder conflicts.
4.3.4. Through agency benefits of debt

Jensen (1986) shows that agency costs increase with free cash flows and that debt may reduce the free cash flow agency problem
by ensuring that managers are disciplined. However, the effect of uncertainty on target leverage ratios through agency benefits of
debt is not clear-cut. If a firm has many profitable assets in place, the profitability of its assets in place will be lower when it is faced
with high uncertainty. For the firm, high uncertainty will lead to less free cash flows and therefore lower shareholder–manager
agency costs. Thus, high uncertainty will lower the value of debt as a disciplining device to the firm, implying that the effect of
uncertainty on the optimal leverage ratio will be negative. However, if a firm has many profitable investment opportunities, a high-
uncertainty firm will have a higher value of the real option to wait (Bloom et al., 2007; Gulen and Ion, 2015), leading to more free
cash flows and therefore higher shareholder–manager agency costs. Thus, it is possible for a high-uncertainty firm to receive more
benefits from the disciplining role of debt.
Given that we find evidence that uncertainty lowers target leverage ratios, we investigate whether some of the effects of un-
certainty are attributed to the reduction in agency benefits of debt driven by increased uncertainty. A 2SLS procedure is employed to
examine if uncertainty reduces shareholder–manager agency conflicts (Stage 1) and if the reduction in shareholder–manager conflicts
lowers target leverage (Stage 2). To implement this test, we employ the percentage of shares held by blockholders, SUMBLKS, which
is inversely related to the degree of the shareholder–manager agency conflicts.39
Columns (9) and (10) in Tables 8 and 9 report the results of 2SLS analyses for book leverage targets and market leverage targets,
respectively. The first-stage regression results show that heightened uncertainty (σi,t−1) does not have a significant impact on the
39
Data regarding blockholders’ shareholdings used by Dlugosz et al. (2006) are available at Andrew Metrick’s website: http://faculty.som.yale.
edu/andrewmetrick/data.html.
18
percentage of shares held by blockholders (SUMBLKSi,t) (i.e. the agency conflicts between shareholders and managers). The second-
stage regression results also show that a higher percentage of shares held by blockholders (SUMBLKSi, t ) does not have a significant
influence on target leverage (BLi, t and MLi, t ). Thus, the results do not provide sufficient evidence that heightened uncertainty leads to
a lower optimal/target leverage ratio by lowering agency benefits related to the disciplining role of debt.
The analyses reported above show evidence supporting the financial distress costs channel and the debtholder–shareholder
conflicts channel as well as the debt tax shields channel. The results suggest that uncertainty decreases debt tax shields, increases
potential financial distress costs and exacerbates debtholder–shareholder conflicts (e.g. under-investment and risk-shifting problems),
thereby leading to a lower optimal/target leverage ratio.
4.4. Additional analyses
4.4.1. Moderating role of managerial risk aversion

In this subsection, we test whether the effect of uncertainty is greater for firms whose managers are more risk averse. Lambrecht
and Myers (2017) suggest that the negative association between uncertainty and target leverage is moderated by managerial risk
aversion. That is, when top managers are more risk averse, uncertainty may have a more pronounced negative effect on target
leverage. To shed light on the moderating role of managerial risk aversion, we conduct a subsample analysis. Motivated by stock
option research in which risk-averse and undiversified executives may choose to exercise options early (Hall and Murphy, 2002), we
identify risk aversion based on their stock option holdings.40 For the analysis, we focus on the risk aversion of CEOs. We classify a
CEO as being risk averse if he/she holds options that are less than 67% in the money or if he/she chooses to exercise the stock options
in a year. We further classify the CEOs without any option grant as being risk averse. Consistent with the notion that risk aversion is a
persistent personal trait, the CEOs identified as being risk averse remain to be so for the entire sample period.41 The threshold of 67%
is chosen following Malmendier and Tate’s (2005) study in which Hall and Murphy’s (2002) model is calibrated. Option moneyness is
estimated as the ratio of realisable value per option to exercise price (Campbell et al., 2011), where exercise price is estimated as the
per-option realisable value net of stock price at the end of each fiscal year and the per-option realisable value is the total realisable
value of exercisable options divided by the number of exercisable options.
The system GMM estimation results are presented in Table 10. In Column (1) and Column (2), the dependent variable is book
leverage, whereas in Column (3) and Column (4), the dependent variable is market leverage. We partition the sample based on the
option-based risk aversion measure into firms with more risk-averse CEOs and firms with less risk-averse CEOs. Columns (1) and (2)
(Columns (3) and (4)) show that the negative effect of asset volatility on book (market) target leverage is more pronounced for firms
with more risk-averse CEOs. In addition, the estimated sensitivities of both book and market target leverage to asset volatility are
statistically significant for firms with more risk-averse CEOs but not for firms with less risk-averse CEOs. Thus, our results in Table 10
are consistent with Lambrecht and Myers’ (2017) model predicting that managerial risk aversion moderates the negative association
between firm uncertainty and target leverage.
4.4.2. Uncertainty and leverage under alternative capital structure theories

Some studies maintain that whether firms tend to adjust towards target leverage is inconclusive (Chang and Dasgupta, 2009;
Welch, 2004). In addition, Graham and Harvey (2001) report that 81%, not 100%, of firms consider a target debt ratio or range when
making their capital structure decisions. Therefore, some firms’ capital structure policies may not be consistent with what the trade-
off theory predicts. Thus, we first discuss the theoretical predictions of the effect of uncertainty on (optimal) leverage under two
alternative capital structure theories (i.e. pecking order theory and market timing theory) and then empirically test whether the
negative effect of uncertainty on (optimal) leverage is more pronounced for firms whose leverage adjustment patterns are more
consistent with a dynamic trade-off theory.
Unlike the static/dynamic trade-off theory, the pecking order and market timing theories both suggest that uncertainty has a
positive effect on (optimal) leverage. First, in the pecking-order world with asymmetric information between managers and outside
investors (Myers, 1984; Myers and Majluf, 1984), higher-uncertainty firms are likely to face more severe information asymmetry and
thus have higher information production costs. Thus, higher-uncertainty firms are expected to prefer debt over equity, implying a
positive association between uncertainty and leverage. Second, in the market-timing world (Baker and Wurgler, 2002; Graham and
Harvey, 2001), managers are more likely to issue (repurchase) equity when their market values are high (low) relative to book and
past market values. Thus, higher-uncertainty firms are less likely to issue equity, implying a positive association between uncertainty
and leverage.
To examine whether the negative effect of uncertainty on (optimal) leverage is more pronounced for firms
whose leverage adjustment patterns are more consistent with a dynamic trade-off theory, we classify firms into two groups based on
the first-order autoregressive regression coefficient (i.e. 1−λ) of each firm’s first-order autoregressive (i.e. AR(1)) model:
40
In an untabulated analysis, we define risk aversion based on a political-ideology-based measure following Hutton et al. (2014) and Deng et al.
(2018). In particular, we consider a CEO as being risk averse if he/she exclusively makes political contributions to the Republican Party. We find
that the results are robust to the use of the alternative measures.
41
A similar method is found in the studies of Hirshleifer et al. (2012) and Hribar and Yang (2016), wherein overconfidence is considered as a
persistent personal trait.
19
Table 10
Moderating effect of managerial risk aversion on target leverage ratios.
Book leverage Market leverage
(1) (2) (3) (4)
Sample More risk averse Less risk averse More risk averse Less risk averse
Lagged leverage 0.770 ⁎⁎⁎

0.851 ⁎⁎⁎
0.691 ⁎⁎⁎
0.708⁎⁎⁎
(0.021) (0.025) (0.024) (0.030)
Asset volatility −0.146⁎⁎⁎ −0.082 −0.235⁎⁎⁎ −0.110
(0.054) (0.068) (0.067) (0.081)
Other target determinants Yes Yes Yes Yes
Firm fixed effects Yes Yes Yes Yes
Number of firms 743 520 743 520
m1 −11.24 −9.542 −10.67 −8.001
(p-value) (0.000) (0.000) (0.000) (0.000)
m2 −0.802 −1.106 −1.502 −3.145
(p-value) (0.422) (0.269) (0.133) (0.002)
m3 3.141 2.511
(p-value) (0.002) (0.012)
m4 −1.493 1.568
(p-value) (0.136) (0.117)
Sargan–Hansen 554.3 413.9 570.5 423.3
(p-value) (0.982) (0.076) (0.945) (0.040)
(0.021) (0.025) (0.024) (0.030)
Target-uncertainty sensitivity ( ) −0.632⁎⁎⁎ −0.553 −0.759⁎⁎⁎ −0.378
(0.230) (0.440) (0.215) (0.281)
This table reports the sub-sample analyses designed to examine whether the effect of uncertainty on target leverage is greater for firms whose
managers are more risk averse. In Column (1) and Column (2), the dependent variable is book leverage, whereas in Column (3) and Column (4), the
dependent variable is market leverage. We adopt the system GMM estimation method. The empirical models used are similar to those used in Tables
3 and 4. We use an option-based measure to identify CEO’s risk aversion. In particular, a CEO is classified as being more risk averse if he/she holds
stock options that are less than 67% in the money at least twice over the sample period. Otherwise, the CEO is classified as being less risk averse.
Option moneyness is estimated as the ratio of realisable value per option to estimated exercise price (Campbell et al., 2011). Exercise price is
estimated as the per-option realisable value net of stock price at the end of each fiscal year, where the per-option realisable value is the total
realisable value of exercisable options divided by the number of exercisable options. A detailed description of all variables included in the models is
provided in Panel A of Table 1. We report two-step GMM coefficients and standard errors that are asymptotically robust to both heteroskedasticity
and serial correlation and that use the finite-sample correction proposed by Windmeijer (2005). The IVs are selected considering the Arellano–Bond
tests and Sargan–Hansen tests. Note that year dummies are treated as instruments for the equations in levels only. m1, m2, m3 and m4 represent the
test statistics of the Arellano–Bond tests for first-order to fourth-order serial correlations in first-differenced residuals, respectively. Sargan–Hansen
represents the test statistic of the Sargan–Hansen test of over-identifying restrictions. The overall goodness-of-fit score, measured as the square of the
coefficient of correlation between the dependent variable and its predicted value, is reported in each column. Superscripts ⁎, ⁎⁎ and ⁎⁎⁎ indicate
statistical significance at the 10%, 5% and 1% levels, respectively.
Lt = + (1 ) Lt 1 + t, (9)
for firm i = 1, ⋯, N.42 In particular, we classify firms with in the range of (0,1) as ‘trade-off firms’ and firms with outside the
range as ‘non-trade-off firms’. We conduct a sub-sample analysis designed to examine whether the effect of uncertainty on target
leverage is different between these two groups of firms.
Table 11 reports the system GMM estimation results for the models similar to those in Tables 3 and 4. Consistent with our predictions
stated above, the effect of uncertainty on target leverage is negative and significant only for ‘trade-off firms’. In the model for book (market)
leverage, the coefficient of uncertainty for ‘trade-off firms’ is negative and statistically significant at the 5% (1%) level. However, the
coefficients of uncertainty for ‘non-trade-off firms’ are not statistically significant at conventional levels in both models. We also report that
the sensitivity of book (market) target leverage to uncertainty for ‘trade-off firms’ is negative and significant at the 5% (1%) level. The
magnitudes of the target-uncertainty sensitivities are similar to those reported in Tables 3 and 4. However, target-uncertainty sensitivities are
not significant for ‘non-trade-off firms’ in both book and market leverage models. Overall, the negative effect of uncertainty on target leverage
is much more pronounced for firms whose capital structure decisions are consistent with the dynamic trade-off theory.
42
A firm-by-firm AR(1) model is essentially the same as a partial adjustment model of leverage with a constant target leverage ratio. Thus, (1−λ)
can be interpreted as each firm’s speed of leverage adjustment. There are substantial sampling errors owing to limited sample size, but it makes sense
to use the estimates to identify firms whose capital structure policies are more (or less) consistent with the dynamic trade-off theory. We are grateful
to an anonymous referee for suggesting this analysis.
20
Table 11
Estimation of the sensitivity of leverage targets to uncertainty: ‘trade-off firms’ versus ‘non-trade-off firms’.
Book leverage Market leverage
(1) (2) (3) (4)
Sample Trade-off firms Non-trade-off firms Trade-off firms Non-trade-off firms
Lagged leverage 0.801 ⁎⁎⁎

0.793 ⁎⁎⁎
0.686 ⁎⁎⁎
0.758⁎⁎⁎
(0.013) (0.088) (0.023) (0.066)
Asset volatility −0.081⁎⁎ 0.023 −0.166⁎⁎⁎ 0.144
(0.033) (0.082) (0.035) (0.094)
Other target determinants Yes Yes Yes Yes
Firm fixed effects Yes Yes Yes Yes
Number of firms 689 123 689 123
m1 −15.49 −3.826 −14.30 −4.151
(p-value) (0.000) (0.000) (0.000) (0.000)
m2 −0.220 0.951 −5.011 −2.125
(p-value) (0.826) (0.341) (0.000) (0.034)
m3 5.651 1.906
(p-value) (0.000) (0.057)
m4 −0.920 0.836
(p-value) (0.358) (0.403)
Sargan–Hansen 644.8 111.1 682.4 110.0
(p-value) (0.323) (0.660) (0.996) (0.428)
Speed of adjustment ( ) 0.199⁎⁎⁎ 0.207⁎⁎ 0.314⁎⁎⁎ 0.242⁎⁎⁎
(0.013) (0.088) (0.023) (0.066)
Target-uncertainty sensitivity ( ) −0.405⁎⁎ 0.111 −0.530⁎⁎⁎ 0.596
(0.164) (0.386) (0.117) (0.409)
This table reports a sub-sample analyse designed to examine whether the effect of uncertainty on target leverage varies between two groups of firms:
‘trade-off firms’ versus ‘non-trade-off firms’. To examine whether the negative effect of uncertainty on optimal leverage is more pronounced for firms
whose leverage adjustment patterns are more consistent with a dynamic trade-off theory, we classify firms into two groups based on the first-order
autoregressive regression coefficient (i.e. 1−λ) of each firm’s first-order autoregressive (i.e. AR(1)) model: Lt = λα + (1−λ)Lt−1 + εt for firm i = 1,
⋯, N. In particular, we classify firms with in the range of (0,1) as ‘trade-off firms’ and firms with outside the range as ‘non-trade-off firms’. In
Column (1) and Column (2), the dependent variable is book leverage, whereas in Column (3) and Column (4), the dependent variable is market
leverage. We adopt the system GMM estimation method. The empirical models used are similar to those used in Tables 3 and 4. A detailed
description of all variables included in the models is provided in Panel A of Table 1. We report two-step GMM coefficients and standard errors that
are asymptotically robust to both heteroskedasticity and serial correlation and that use the finite-sample correction proposed by Windmeijer (2005).
The IVs are selected considering the Arellano–Bond tests and Sargan–Hansen tests. Note that year dummies are treated as instruments for the
equations in levels only. m1, m2, m3 and m4 represent the test statistics of Arellano–Bond tests for first-order to fourth-order serial correlations in
first-differenced residuals, respectively. Sargan–Hansen represents the test statistic of the Sargan–Hansen test of over-identifying restrictions. The
overall goodness-of-fit score, measured as the square of the coefficient of correlation between the dependent variable and its predicted value, is
reported in each column. Superscripts ⁎, ⁎⁎ and ⁎⁎⁎ indicate statistical significance at the 10%, 5% and 1% levels, respectively.
4.4.3. Robustness tests

As discussed in Section 3.2, asset volatility as an uncertainty measure has several advantages over alternative uncertainty
measures, particularly when investigating the effect of uncertainty on actual/optimal leverage. However, many alternative un-
certainty measures have been used in the literature. As alternative uncertainty measures, we consider the following four uncertainty
proxies: volatility index (VIX) (Brenner and Galai, 1989), economic policy uncertainty (EPU) index (Baker et al., 2016), equity
volatility (Bloom et al., 2007) and implied volatility (Alfaro et al., 2018). Among them, the first two are macro-level uncertainty
measures whereas the remaining two are firm-level uncertainty measures. We perform several robustness tests using those proxies.
The robustness tests presented in the internet appendix (Section A) suggest that our main finding—asset volatility lowers target
leverage—is not driven by the correlation between asset volatility and macroeconomic uncertainty and that asset volatility has a
significant incremental impact on target leverage in excess of the impact of macroeconomic uncertainty. Furthermore, we show that a
firm-level business uncertainty measure (independent of capital structure decisions)—asset volatility—has a significantly larger effect
on target leverage than the two alternative firm-level uncertainty measures.43
Furthermore, we examine whether our results are driven by the use of manufacturing firms with detailed debt data. First, we
estimate the dynamic panel regression models (in Section 4.1) using a less restricted sample. Following Fama and French (2002),
43
As discussed earlier, the two alternative firm-level proxies based on stock returns may be influenced by capital structure decisions. Thus, the
impact of uncertainty on target leverage cannot be estimated consistently when equity volatility or implied volatility is used as an uncertainty
measure.
21
Flannery and Rangan (2006) and Faulkender et al. (2012), we include all industries except financial services and regulated utilities in
our sample. Next, we estimate the IV regression models (in Section 4.2) using the less restricted sample and using equity volatility as
an uncertainty measure. By doing so, we include firms that do not have detailed debt data into the sample. The results are presented
in the internet appendix (Section B). We find that the negative impact of uncertainty is still valid in a more generalised sample,
suggesting that our main results are not driven by the selection of extreme firms or availability of detailed debt data.
5. Conclusion
This study investigates how uncertainty affects firms’ optimal capital structures using a panel data set of U.S. public manu-
facturers between 2003 and 2018. In this paper, we address this question by first identifying the directional effect of increased
uncertainty on leverage targets and then considering through which mechanisms uncertainty affects target leverage ratios. Using
asset volatility as an uncertainty measure and target/optimal leverage ratios obtained by estimating a partial adjustment model, we
find that uncertainty lowers firms’ target leverage ratios. High-uncertainty firms have 10.1 (8.1) percentage points lower mean book
(market) targets than low-uncertainty firms. This study also shows that the effect of uncertainty on leverage targets is greater than the
impact of firm size, market-to-book ratio, assets tangibility, R&D intensity and industry median leverage, making uncertainty the
most crucial determinant among all time-varying determinants of leverage targets. This paper also explores several possible me-
chanisms through which uncertainty could influence target leverage ratios by empirically testing the effects of uncertainty on debt
tax shields, financial distress costs and agency costs and benefits. The results suggest that heightened uncertainty leads to a lower
optimal leverage ratio by decreasing debt tax shields, increasing potential financial distress costs and exacerbating debt-
holder–shareholder conflicts.
Appendix A. Approximating loan returns
For firm-month observations without loan sale information, we approximate loan returns using bond returns following Choi
(2013). First, we classify firms into three groups based on their S&P credit ratings. Second, we estimate the following regression
model by the three groups:
RL RF = + 1 (R
B RF ) + 2 (R
T RF ) + ,
where RT is the return on a one-year treasury bond, RB is the return on bonds, RL is the return on bank loans, and RF is one-month T-
bill rate. The inclusion of RT−RF allows us to consider the difference in interest sensitivities between bonds and loans. In addition,
estimating separate regression models by credit rating groups allows us to take into account the differential correlation between bond
returns and loan returns with respect to the varying credit risks of the underlying assets. Presumably, we expect to see a stronger
predictive power of bond returns on loan returns for firms with a lower credit rating. Regression results are reported in Table A1.
Table A1. Regression results used to approximate loan returns.
Dependent variable: RL−RF
(1) (2) (3)
BBB and above Between BBB and B Below B or unrated
B F
R −R 0.232 ⁎⁎⁎
0.336 ⁎⁎⁎
0.381⁎⁎⁎
(0.002) (0.001) (0.002)
RT−RF −1.298⁎⁎⁎ −1.293⁎⁎⁎ −1.275⁎⁎⁎
(0.009) (0.009) (0.015)
Adjusted R-squared 0.325 0.592 0.432
Appendix B. Supplementary data
The internet appendix to this article can be found online at https://doi.org/10.1016/j.jcorpfin.2020.101642.
References
Alfaro, I., Bloom, N., Lin, X., 2018. The Finance Uncertainty Multiplier. Tech.rep.. National Bureau of Economic Research.
Anderson, T.W., Hsiao, C., 1981. Estimation of dynamic models with error components. J. Am. Stat. Assoc. 76 (375), 598–606.
Anderson, T.W., Hsiao, C., 1982. Formulation and estimation of dynamic models using panel data. J. Econ. 18 (1), 47–82.
Andrade, G., Kaplan, S.N., 1998. How costly is financial (not economic) distress? Evidence from highly leveraged transactions that became distressed. J. Financ. 53 (5),
1443–1493.
Antoniou, A., Guney, Y., Paudyal, K., 2008. The determinants of capital structure: capital market-oriented versus bank-oriented institutions. J. Financ. Quant. Anal. 43
(1), 59–92.
Arellano, M., Bond, S., 1991. Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. Rev. Econ. Stud. 58 (2),
277–297.
22
Arellano, M., Bover, O., 1995. Another look at the instrumental variable estimation of error-components models. J. Econ. 68 (1), 29–51.
Baker, M., Wurgler, J., 2002. Market timing and capital structure. J. Financ. 57 (1), 1–32.
Baker, S.R., Bloom, N., Davis, S.J., 2016. Measuring economic policy uncertainty. Q. J. Econ. 131 (4), 1593–1636.
Bernanke, B.S., 1983. Nonmonetary effects of the financial crisis in the propagation of the Great Depression. Am. Econ. Rev. 73 (3), 257–276.
Bhandari, L.C., 1988. Debt/equity ratio and expected common stock returns: empirical evidence. J. Financ. 43 (2), 507–528.
Bharath, S.T., Shumway, T., 2008. Forecasting default with the merton distance to default model. Rev. Financ. Stud. 21 (3), 1339–1369.
Black, F., Scholes, M., 1973. The pricing of options and corporate liabilities. J. Polit. Econ. 81 (3), 637–654.
Bloom, N., 2009. The impact of uncertainty shocks. Econometrica 77 (3), 623–685.
Bloom, N., Bond, S., Van Reenen, J., 2007. Uncertainty and investment dynamics. Rev. Econ. Stud. 74 (2), 391–415.
Bloom, N., Floetotto, M., Jaimovich, N., Saporta-Eksten, I., Terry, S.J., 2018. Really uncertain business cycles. Econometrica 86 (3), 1031–1065.
Blouin, J., Core, J.E., Guay, W., 2010. Have the tax benefits of debt been overestimated? J. Financ. Econ. 98 (2), 195–213.
Blundell, R., Bond, S., 1998. Initial conditions and moment restrictions in dynamic panel data models. J. Econ. 87 (1), 115–143.
Bond, S.R., 2002. Dynamic panel data models: a guide to micro data methods and practice. Port. Econ. J. 1 (2), 141–162.
Bradley, M., Jarrell, G.A., Kim, E.H., 1984. On the existence of an optimal capital structure: theory and evidence. J. Financ. 39 (3), 857–878.
Brenner, M., Galai, D., 1989. New financial instruments for hedge changes in volatility. Financ. Anal. J. 45 (4), 61–65.
Bruno, G.S., 2005. Approximating the bias of the LSDV estimator for dynamic unbalanced panel data models. Econ. Lett. 87 (3), 361–366.
Campbell, T.C., Gallmeyer, M., Johnson, S.A., Rutherford, J., Stanley, B.W., 2011. CEO optimism and forced turnover. J. Financ. Econ. 101 (3), 695–712.
Chang, X., Dasgupta, S., 2009. Target behavior and financing: how conclusive is the evidence? J. Financ. 64 (4), 1767–1796.
Chava, S., Livdan, D., Purnanandam, A., 2008. Do shareholder rights affect the cost of bank loans? Rev. Financ. Stud. 22 (8), 2973–3004.
Chen, H., 2010. Macroeconomic conditions and the puzzles of credit spreads and capital structure. J. Financ. 65 (6), 2171–2212.
Choi, J., 2013. What drives the value premium?: the role of asset risk and leverage. Rev. Financ. Stud. 26 (11), 2845–2875.
Choi, J., Richardson, M., 2016. The volatility of a firm’s assets and the leverage effect. J. Financ. Econ. 121 (2), 254–277.
Christie, A.A., 1982. The stochastic behavior of common stock variances: value, leverage and interest rate effects. J. Financ. Econ. 10 (4), 407–432.
Çolak, G., Gungoraydinoglu, A., Öztekin, Ö., 2018. Global leverage adjustments, uncertainty, and country institutional strength. J. Financ. Intermed. 35, 41–56.
DeAngelo, H., Masulis, R.W., 1980. Optimal capital structure under corporate and personal taxation. J. Financ. Econ. 8 (1), 3–29.
Deng, M., Ho, J.L., Li, S., 2018. Does managerial risk aversion affect earnings management? Evidence from CEO political ideology. Baruch College Zicklin School of
Business Research Paper (2018-05).
Dick-Nielsen, J., 2009. Liquidity biases in TRACE. J. Fixed Income 19 (2), 43–55.
Dick-Nielsen, J., 2014. How to Clean Enhanced TRACE Data. (Available at SSRN 2337908).
Dlugosz, J., Fahlenbrach, R., Gompers, P., Metrick, A., 2006. Large blocks of stock: prevalence, size, and measurement. J. Corp. Finan. 12 (3), 594–618.
Drucker, S., Puri, M., 2008. On loan sales, loan contracting, and lending relationships. Rev. Financ. Stud. 22 (7), 2835–2872.
Easterbrook, F.H., 1984. Two agency-cost explanations of dividends. Am. Econ. Rev. 74 (4), 650–659.
Edwards, A.K., Harris, L.E., Piwowar, M.S., 2007. Corporate bond market transaction costs and transparency. J. Financ. 62 (3), 1421–1451.
Elsas, R., Flannery, M.J., Garfinkel, J.A., 2014. Financing major investments: information about capital structure decisions. Rev. Finance 18 (4), 1341–1386.
Fama, E.F., French, K.R., 1997. Industry costs of equity. J. Financ. Econ. 43 (2), 153–193.
Fama, E.F., French, K.R., 2002. Testing trade-off and pecking order predictions about dividends and debt. Rev. Financ. Stud. 15 (1), 1–33.
Fama, E.F., Miller, M.H., 1972. The Theory of Finance. Holt Rinehart & Winston.
Faulkender, M., Flannery, M.J., Hankins, K.W., Smith, J.M., 2012. Cash flows and leverage adjustments. J. Financ. Econ. 103 (3), 632–646.
Fischer, E.O., Heinkel, R., Zechner, J., 1989. Dynamic capital structure choice: theory and tests. J. Financ. 44 (1), 19–40.
Flannery, M.J., Hankins, K.W., 2013. Estimating dynamic panel models in corporate finance. J. Corp. Finan. 19, 1–19.
Flannery, M.J., Rangan, K.P., 2006. Partial adjustment toward target capital structures. J. Financ. Econ. 79 (3), 469–506.
Frank, M.Z., Goyal, V.K., 2009. Capital structure decisions: which factors are reliably important? Financ. Manag. 38 (1), 1–37.
Friend, I., Lang, L.H., 1988. An empirical test of the impact of managerial self-interest on corporate capital structure. J. Financ. 43 (2), 271–281.
George, T.J., Hwang, C.-Y., 2010. A resolution of the distress risk and leverage puzzles in the cross section of stock returns. J. Financ. Econ. 96 (1), 56–79.
Graham, J.R., 1996. Proxies for the corporate marginal tax rate. J. Financ. Econ. 42 (2), 187–221.
Graham, J.R., 2000. How big are the tax benefits of debt? J. Financ. 55 (5), 1901–1941.
Graham, J.R., Harvey, C.R., 2001. The theory and practice of corporate finance: evidence from the field. J. Financ. Econ. 60 (2–3), 187–243.
Graham, J.R., Kim, H., 2009. Simulating Corporate Marginal Income Tax Rates and Implications for Corporate Debt Policy. Duke University Working Paper.
Graham, J.R., Lemmon, M.L., Schallheim, J.S., 1998. Debt, leases, taxes, and the endogeneity of corporate tax status. J. Financ. 53 (1), 131–162.
Gulen, H., Ion, M., 2015. Policy uncertainty and corporate investment. Rev. Financ. Stud. 29 (3), 523–564.
Gungoraydinoglu, A., Çolak, G., Öztekin, Ö., 2017. Political environment, financial intermediation costs, and financing patterns. J. Corp. Finan. 44, 167–192.
Hadlock, C.J., Pierce, J.R., 2010. New evidence on measuring financial constraints: moving beyond the KZ index. Rev. Financ. Stud. 23 (5), 1909–1940.
Hall, B.J., Murphy, K.J., 2002. Stock options for undiversified executives. J. Account. Econ. 33 (1), 3–42.
Hamada, R.S., 1972. The effect of the firm’s capital structure on the systematic risk of common stocks. J. Financ. 27 (2), 435–452.
Harris, M., Raviv, A., 1991. The theory of capital structure. J. Financ. 46 (1), 297–355.
Hirshleifer, D., Low, A., Teoh, S.H., 2012. Are overconfident ceos better innovators? J. Financ. 67 (4), 1457–1498.
Hribar, P., Yang, H., 2016. Ceo overconfidence and management forecasting. Contemp. Account. Res. 33 (1), 204–227.
Hutton, I., Jiang, D., Kumar, A., 2014. Corporate policies of republican managers. J. Financ. Quant. Anal. 49 (5–6), 1279–1310.
Im, H.J., 2012. Essays in Corporate Finance. Ph.D.thesis. Oxford University, UK.
Im, H.J., 2019. Asymmetric peer effects in capital structure dynamics. Econ. Lett. 176, 17–22.
Jensen, M.C., 1986. Agency costs of free cash flow, corporate finance, and takeovers. Am. Econ. Rev. 76 (2), 323–329.
Jensen, M.C., Meckling, W.H., 1976. Theory of the firm: managerial behavior, agency costs and ownership structure. J. Financ. Econ. 3 (4), 305–360.
Keefe, M.O., Yaghoubi, M., 2016. The influence of cash flow volatility on capital structure and the use of debt of different maturities. J. Corp. Finan. 38, 18–36.
Kellogg, R., 2014. The effect of uncertainty on investment: evidence from Texas oil drilling. Am. Econ. Rev. 104 (6), 1698–1734.
Kim, H., Kung, H., 2016. The asset redeployability channel: how uncertainty affects corporate investment. Rev. Financ. Stud. 30 (1), 245–280.
Kim, W.S., Sorensen, E.H., 1986. Evidence on the impact of the agency costs of debt on corporate debt policy. J. Financ. Quant. Anal. 21 (2), 131–144.
Kleibergen, F., Paap, R., 2006. Generalized reduced rank tests using the singular value decomposition. J. Econ. 133 (1), 97–126.
Lambrecht, B.M., Myers, S.C., 2017. The dynamics of investment, payout and debt. Rev. Financ. Stud. 30 (11), 3759–3800.
Leahy, J.V., Whited, T.M., 1996. The effect of uncertainty on investment: some stylized facts. J. Money Credit Bank. 28 (1).
Leary, M.T., Roberts, M.R., 2005. Do firms rebalance their capital structures? J. Financ. 60 (6), 2575–2619.
Leland, H.E., 1994. Corporate debt value, bond covenants, and optimal capital structure. J. Financ. 49 (4), 1213–1252.
Lemmon, M.L., Roberts, M.R., Zender, J.F., 2008. Back to the beginning: persistence and the cross-section of corporate capital structure. J. Financ. 63 (4), 1575–1608.
Li, X., Qiu, M., 2018. How important is economic policy uncertainty for capital structure decisions? Evidence from US Firms.
Malmendier, U., Tate, G., 2005. CEO overconfidence and corporate investment. J. Financ. 60 (6), 2661–2700.
Merton, R.C., 1974. On the pricing of corporate debt: the risk structure of interest rates. J. Financ. 29 (2), 449–470.
Miles, J.A., Ezzell, J.R., 1985. Reformulating tax shield valuation: a note. J. Financ. 40 (5), 1485–1492.
Modigliani, F., Miller, M., 1958. The cost of capital, corporation finance and the theory of finance. Am. Econ. Rev. 48 (3), 291–297.
23
Myers, S.C., 1977. Determinants of corporate borrowing. J. Financ. Econ. 5 (2), 147–175.
Myers, S.C., 1984. The capital structure puzzle. J. Financ. 39 (3), 574–592.
Myers, S.C., Majluf, N.S., 1984. Corporate financing and investment decisions when firms have information that investors do not have. J. Financ. Econ. 13 (2),
187–221.
Nickell, S., 1981. Biases in dynamic models with fixed effects. Econometrica 1417–1426.
Nini, G., Smith, D.C., Sufi, A., 2009. Creditor control rights and firm investment policy. J. Financ. Econ. 92 (3), 400–420.
Stulz, R., 1990. Managerial discretion and optimal financing policies. J. Financ. Econ. 26 (1), 3–27.
Welch, I., 2004. Capital structure and stock returns. J. Polit. Econ. 112 (1), 106–131.
Windmeijer, F., 2005. A finite sample correction for the variance of linear efficient two-step GMM estimators. J. Econ. 126 (1), 25–51.
24

Journal of Corporate Finance: Hyun Joong Im, Ya Kang, Janghoon Shon T

Uploaded by

Copyright:

Available Formats

Journal of Corporate Finance: Hyun Joong Im, Ya Kang, Janghoon Shon T

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Journal of Corporate Finance: Hyun Joong Im, Ya Kang, Janghoon Shon T

Uploaded by

Copyright:

Available Formats

Journal of Corporate Finance 64 (2020) 101642

Contents lists available at ScienceDirect

Journal of Corporate Finance

How does uncertainty influence target capital structure?☆

ARTICLE INFO ABSTRACT

2. Predictions from theory

3.1. Sample selection

3.2. Measurement of variables

3.2.1. Measuring uncertainty

3.2.2. Measuring control variables

3.3. Descriptive statistics

Panel B. Summary statistics

Variables Obs. Mean S.D. P05 P25 Median P75 P95

3.4. Research design

Variables Obs. Mean S.D. P05 P25 Median P75 P95

Variables Obs. Mean S.D. P05 P25 Median P75 P95

Li, t = + Xi, t 1 + i (3)

Li, t = b0 + b1 Li, t 1 + b2 Xi, t 1 + t + i + i, t , (5)

4.1. The effect of uncertainty on long-run leverage targets

(1) (2) (3) (4)

Estimation method OLS WG LSDVC System GMM

Lagged book leverage 0.847 ⁎⁎⁎

(1) (2) (3) (4)

Estimation method OLS WG LSDVC System GMM

Lagged market leverage 0.815 ⁎⁎⁎

Variables Obs. Mean S.D. P05 P25 Median P75 P95

Panel B. Market target leverage

Variables Obs. Mean S.D. P05 P25 Median P75 P95

targets to uncertainty reported in Tables 3 and 4.

(1) (2) (3) (4) (5) (6)

Firm fixed effects 0.987 0.989 0.965 0.968

Panel B. Market target leverage

(1) (2) (3) (4) (5) (6)

Firm fixed effects 0.989 0.990 0.970 0.972

4.2. Mitigating potential endogeneity concerns

Li, t = constant + (1 ) Li , t 1 + i, t 1 + Ci , t 1 + i, t , (6)

4.3. Mechanisms through which uncertainty lowers long-run leverage targets

4.3.1. Through debt tax shields

(1) (2) (3) (4) (5) (6)

Variables Δσi,t−1 ΔLi,t−1 ΔLi,t Δσi,t−1 ΔLi,t−1 ΔLi,t

Change in book leverage Change in market leverage

(1) (2) (3) (4) (5) (6)

Variables Δσi,t−1 ΔLi,t−1 ΔLi,t Δσi,t−1 ΔLi,t−1 ΔLi,t

Target-uncertainty sensitivity ( ) −2.192 ⁎⁎⁎

4.3.2. Through financial distress costs

Variables TXSHLDi,t MZi,t SAi,t VIOLYRi,t SUMBLKSi,t

σi,t−1 −0.300 ⁎⁎⁎

4.3.3. Through agency costs of debt

Variables TXSHLDi,t MZi,t SAi,t VIOLYRi,t SUMBLKSi,t

σi,t−1 −0.300 ⁎⁎⁎

4.3.4. Through agency benefits of debt

4.4. Additional analyses

4.4.1. Moderating role of managerial risk aversion

4.4.2. Uncertainty and leverage under alternative capital structure theories

(1) (2) (3) (4)

Lagged leverage 0.770 ⁎⁎⁎

(1) (2) (3) (4)

Sample Trade-off firms Non-trade-off firms Trade-off firms Non-trade-off firms

Lagged leverage 0.801 ⁎⁎⁎