Received: 22 June 2020 | Accepted: 23 August 2022

DOI: 10.1111/jfir.12300

ORIGINAL ARTICLE

Valuing the speed of adjustment of capital

structure

Brian Clark

Department of Finance, Lally School of

Management, Rensselaer Polytechnic Abstract
Institute, Troy, New York, USA
I estimate the optimal speed of adjustment (SOA) policy for
Correspondence a large sample of US firms by examining the dynamics of
Brian Clark, Department of Finance, Lally
the opportunity cost of leverage. Studying the dynamics of
School of Management, Rensselaer
Polytechnic Institute, Troy, NY 12180, USA. the opportunity cost of leverage as opposed to leverage
Email: [email protected]
ratios allows me to estimate the value of a given SOA policy
and hence calibrate the optimal SOA policy in terms
of firm value. Simulation results show that the optimal
SOA is between 31% and 37% when adjustment costs are
considered. The sensitivity of firm value to SOA is low for a
wide range of SOAs around the optimal level. However,
extremely high or low SOAs have substantial negative
impacts on firm value.

KEYWORDS
capital structure, corporate finance, speed of adjustment

JEL CLASSIFICATION
G30, G32

1 | INTRODUCTION

What is the optimal speed of adjustment (SOA) of capital structure? Several studies document statistically significant
mean reversion in leverage ratios, implying that firms actively manage toward target capital structures. However,
disagreement remains regarding the economic significance of this observed behavior because there is no direct link
between SOA estimates and firm value. For example, Graham and Leary (2011) note that although there is a clear
difference between SOA estimates of 0% and 100%, it is less clear whether there is a meaningful difference between
SOA estimates of 10%, 25%, and 50%. This ambiguity arises because the literature focuses on the SOA of leverage
ratios and has not linked SOA estimates to firm value as they do not specify a function to map changes in leverage

© 2022 The Southern Finance Association and the Southwestern Finance Association.

J Financ Res. 2022;45:855–879. wileyonlinelibrary.com/journal/JFIR | 855

 14756803, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jfir.12300 by Australian National University, Wiley Online Library on [10/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
856 | JOURNAL OF FINANCIAL RESEARCH

ratios to changes in firm value. As such, existing studies cannot directly examine the relation between SOA and firm
value and hence cannot examine what, if any, SOA policy is optimal in terms of maximizing firm value.
The purpose of this article, therefore, is to calibrate the empirical relation between SOA and firm value to
estimate the optimal SOA policy that maximizes firm value. To the best of my knowledge, this is the first article to
empirically calibrate this relation and address the question of what is the optimal SOA policy. Simulation results
show that the optimal SOA is between 31% and 37% when adjustment costs are considered. Whereas the
sensitivity of firm value to SOA is very low for a wide range of SOAs around the optimal level, extremely high or low
SOAs would have substantial negative impacts on firm value.
To calibrate the optimal SOA policy, I first estimate the opportunity cost of leverage (OC(L)) for each firm‐year
observation in my sample following the approach of Graham (2000) and Van Binsbergen et al. (2010). Intuitively,
OC(L) represents the difference between a firm's current value and its value at its optimal capital structure (i.e., the
value a firm is forgoing by being away from its optimal leverage ratio).1 The purpose of a positive SOA policy is to
minimize OC(L) and hence maximize firm value. Therefore, studying the dynamics of the estimated OC(L) functions
allows me to directly examine the link between changes in leverage and changes in firm value, because the dynamic
adjustment variable, OC(L), is inherently tied to firm value through the underlying marginal cost and benefit of
leverage functions. Importantly, such value functions are not present in any of the SOA studies that use leverage
ratios to study the dynamics of capital structure adjustment (see, e.g., Chang & Dasgupta, 2009; Fama &
French, 2002; Flannery & Rangan, 2006; Huang & Ritter, 2009).2 As such, this article offers a substantial departure
from other SOA studies that cannot answer the question: What is the economic benefit of one SOA policy relative
to another? Consequently, it is the first article to empirically calibrate the optimal SOA policy in terms of firm value.
The novelty of this article hence rests on the ability to link the dynamics of changes in capital structure to changes
in firm value. This is done by estimating OC(L) for each firm‐year observation as follows. First, estimate the firm‐year‐
specific marginal benefit of debt functions based on the tax deductibility of interest payments following Blouin et al.
(2010) who build on the methodology developed by Graham(1996a, 1996b, 2000). Second, use exogenous variation in
these marginal tax benefit of debt functions to identify firm‐year‐specific marginal cost of debt functions (Van
Binsbergen et al., 2010). Finally, integrate the area between these curves from a firm's current leverage ratio to the
optimal leverage ratio to estimate OC(L) for each firm‐year observation (Van Binsbergen et al., 2010).
This empirical process produces an estimate of OC(L) for each firm‐year observation. Given these estimates,
the next step is to estimate a partial‐adjustment model to test for mean reversion in OC(L). I estimate the SOA of
OC(L) (henceforth, SOAOC) using a standard partial‐adjustment model akin to what is commonly done in the
literature to estimate the SOA of leverage ratios (henceforth, SOAL) for a sample of nonfinancial, nonregulated
Compustat firms from 1971 to 2012. The mean estimate of SOAOC is a statistically significant 47.3% using the
Blundell and Bond system generalized method of moments (GMMBB), meaning that firms close about half the value
gap between their actual and target capital structure each year.3 By comparison, the mean estimate of SOAL is
29.1% using GMMBB, which is consistent with previous studies.4
Because the SOAOC estimates are based on OC(L) functions that link leverage to firm value, I can use the
resulting SOAOC estimates to simulate the distribution of OC(L) for various hypothetical values of SOAOC over

1
The opportunity cost measure is derived from papers such as Van Binsbergen et al. (2010) and Korteweg (2010) who quantify the net benefit of debt
financing.
2
Other papers use alternative approaches to test whether firms have target leverage ratios. For example, Harford et al. (2009) use a sample of large
acquisitions and show that the acquiring firm's initial leverage ratio affects the financing of the purchase and that firms tend to adjust back toward a target
leverage ratio post acquisition. Other papers examine the determinants of SOA. Devos et al. (2017) show that debt covenants affect SOA, and Yin and
Ritter (2020) estimate the effect of stock price fluctuations on SOA.
3
I also estimate all models using ordinary least squares (OLS) and firm fixed‐effects regressions. Consistent with previous studies, OLS estimates tend to
underestimate SOAOC, and firm fixed‐effects models tend to overestimate SOAOC.
4
The ratio of SOAOC to SOAL is about 1.7, which is within the range of what would be predicted by an analytical derivation of the relation between SOAOC
and SOAL. The basic reason is that OC(L) increases at an increasing rate as firms move away from their target leverage ratio. It is also consistent with
Mukherjee and Wang's (2013) finding that SOAL is an increasing function of the deviation from the target capital structure. See Section 2 for details.
 14756803, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jfir.12300 by Australian National University, Wiley Online Library on [10/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
VALUING THE SPEED OF ADJUSTMENT | 857

time and compute the expected OC(L) for any hypothetical SOAOC as a fraction of firm value. That is, I can simulate
the value gain associated with any SOAOC policy relative to random financing (i.e., passive capital structure
management, or SOAOC = 0%) and hence calibrate the empirical relation between SOAOC and firm value. In
particular, the residuals from the estimated partial‐adjustment models represent random shocks that push firms
away from their optimal capital structures to suboptimal leverage ratios, which can be mapped to costs through the
OC(L) functions. Firms that do not adjust (SOAOC = 0%) sacrifice the value of potentially reducing these random OC
(L) shocks. Firms that do choose a positive SOAOC actively adjust in response to these random shocks and hence
tend to operate closer to their optimal leverage ratios and have a lower OC(L) (i.e., greater firm value). By simulating
OC(L) for different SOAOC policies, I can therefore compare the difference between any two policies (e.g., the mean
observed SOAOC relative to the passive adjustment policy of SOAOC = 0%).
Because capital structure adjustment is costly, I also consider adjustment costs in the form of debt or equity
issuance spreads in the simulations. In the absence of adjustments costs, firms would obviously stand to benefit
from completely adjusting (i.e., SOAOC = 100%) as it would minimize OC(L). However, frequent adjustments to
capital structure in the form of issuing debt and/or equity are costly. Therefore, as these costs increase, the optimal
SOAOC policy decreases, making firm value a concave function of SOAOC. In my simulations, I include adjustment
costs in two ways. First, I consider the models of Altinkilic and Hansen (2000) who examine the behavior of
seasoned common stock and bond offering spreads paid to underwriters. Specifically, I use their reduced‐form
econometric models to estimate floatation costs that have both fixed and variable components. Second, for
robustness, I assume that floatation costs are completely variable (relative to the size of the issue) and about 5 times
larger for equity issues than debt issues, and I run the simulations for various levels of costs to establish a relation
between optimal SOAOC and adjustment costs.
The main findings are as follows. First, firm value is a concave function of SOAOC, so an optimal SOA policy exists
that maximizes firm value. Although unsurprising, mine is the first article to empirically calibrate this relation. Second, the
optimal SOAOC policy is decreasing in the level of adjustment costs but begins to level off at very high levels of
adjustment costs. This means that even for firms with very high adjustment costs, a positive SOAOC is still optimal. Third,
the simulations suggest that the optimal SOAOC policy is between 31% and 37% for realistic levels of adjustment costs.
Thus, the mean observed SOAOC policy of 47.3% is higher than the optimal (value‐maximizing) SOAOC policy implied by
my simulations. However, the difference in firm value between the observed SOAOC policy and the optimal SOAOCs
implied by the simulations is very small as firms stand to gain only up to about 0.3% of total assets by changing to the
simulated optimal SOAOC of 31%–37% from the mean observed SOAOC of 47.3%. Therefore, firms appear to be very
close to optimizing firm value, even though they seem to be adjusting slightly too fast. This is because firm value is not
very sensitive to SOAOC for a relatively wide range of SOAOCs around the optimum (i.e., the middle of the firm‐value vs.
SOAOC curve is relatively flat). However, extremely slow or fast SOAOC policies can have substantially negative impacts
on firm value. In other words, the relation between firm value and SOAOC follows an inverted U‐shape.
These results have several implications for the capital structure literature. First, the methodology allows for the
calibration of the benefits of capital structure adjustment as they relate SOAOC estimates to firm value. Second, the
results are consistent with evidence that firms target a range of leverage ratios (see, e.g., Graham & Harvey, 2001;
Leary & Roberts, 2005) and rebalance only when the leverage ratio migrates outside of a band of leverage ratios.
This is evidenced by the inverted U‐shape of the relation between firm value and SOAOC. These prior empirical
findings are at least partially driven by the fact that firm value is a concave function of leverage—meaning that the
benefits of adjusting increase nonlinearly as firms drift away from their target leverage ratios. This is alluded to and
tested in a static model by studies such as Van Binsbergen et al. (2010) and Korteweg (2010) but not explicitly
tested in a dynamic SOA model. I test this directly and empirically calibrate the relation between firm value and
SOAOC. Therefore, although the dynamics of leverage ratios may very well be described as lumpy (i.e., a slow SOA)
as in the previous studies, my results show that firms are actually operating an SOAOC that is slightly higher than the
optimal SOAOC but are still very close to maximizing firm value. The results support and extend the literature on the
net benefits of debt financing. For example, Korteweg (2010) shows that the maximum attainable net benefits of
 14756803, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jfir.12300 by Australian National University, Wiley Online Library on [10/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
858 | JOURNAL OF FINANCIAL RESEARCH

debt are about 5.5% of firm value, but the average observed benefits of debt are around 4% of firm value.
Van Binsbergen et al. (2010) show a similar net benefit of debt financing. My results suggest that the dynamic
adjustment process accounts for a large portion of the net benefits of debt as firms would be about 2% worse off if
they did not actively adjust at all. Finally, adjustment costs play a large role in explaining the difference between the
maximum attainable and average observed net benefits of debt.

2 | USING T HE PARTIAL ‐A D J U S T M E N T M ODEL T O E STI M ATE SO A

The standard SOA partial‐adjustment model stems from the idea that firms experience random shocks that push
them away from their target capital structures. Consequently, models that assume firms are always in equilibrium
will have low explanatory power because firms are often operating out of equilibrium (Strebulaev, 2007). However,
if firms behave as the traditional trade‐off theory predicts and actively manage capital structure toward individual
targets, then even in the presence of random shocks, capital structure should exhibit mean reversion and we should
observe a positive SOA. Alternatively, if firms make random financing decisions or if capital structure is a result of
repeated attempts to time the market (Baker & Wurgler, 2002), then there should be no evidence of mean reversion
and the SOA should be equal to zero.
Despite the intuitive appeal of this model, it is not clear how to interpret the extant SOA estimates because
there is no direct link between changes in leverage ratios and changes in firm value. Therefore, my novel empirical
approach is to replace leverage ratios with a function that maps changes in leverage ratios to changes in firm value
to provide a link between changes in capital structure and firm value. I define this mapping function as OC(L) and it
represents the value that a firm is forgoing by being away from its optimal leverage ratio. To estimate OC(L), I follow
the methodology of Van Binsbergen et al. (2010) because it provides for the calibration of entire marginal cost and
benefit of debt functions for each firm‐year observation.5
Most SOA papers estimate a version of the following partial‐adjustment model:6

( )
Li, t +1 − Li, t = α + SOAL L*i, t − Li, t + εi, t +1, (1)

where Li,t is observed leverage for firm i at time t, L*i, t is the unobserved target leverage ratio, SOAL is the estimated
speed of adjustment, εi,t+1 is the error, and α is a constant that should be equal to zero. The novelty of this article is
to replace the leverage ratios in Equation (1) with the OC(L) and estimate the following:

 
OC(Li, t +1) − OC(Li, t) = α + SOA OC OC L*i, t − OC Li, t
 ( ) ( ) + εi,t+1, (2)

where OC(Li,t) is the opportunity cost of leverage for firm i at time t and OC(L*i, t) is the target opportunity cost of
leverage, which is equal to zero by definition. This new specification does not imply that estimates of SOAL based
on Equation (1) are incorrect; rather, it is a reformulation of the problem so that the model links capital structure
decisions to firm value.

5
One potential downside to this methodology is that it relies on the tax benefits of debt to identify the marginal benefit and hence marginal cost functions.
This may be problematic because there is debate as to the importance of taxes in determining capital structure. Some recent papers such as Li et al. (2016)
question whether taxes are a primary determinant of capital structure, whereas others such as Heider and Ljungqvist (2015) provide evidence that taxes
are a first‐order determinant of capital structure. However, few disagree that taxes play at least some role in determining capital structure. This is
important to note because even if taxes are not a first‐order determinant of capital structure, the Van Binsbergen et al. (2010) model is not necessarily
invalid. The reason is that their marginal “cost” functions contain variables that represent other (nontax) benefits of debt, so as long as these cost functions
can be properly identified, the model is valid in terms of the overall net benefits of debt. I expand on this point in Section 3.3 and Online Appendix C.
6
The most prominent econometric estimation techniques include OLS (Fama & French, 2002; Lemmon et al., 2008), firm fixed effects (Flannery &
Rangan, 2006), implied target (Welch, 2004), long difference estimator (Huang & Ritter, 2009), and GMMBB (Flannery & Hankins, 2013; Lemmon
et al., 2008).
 14756803, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jfir.12300 by Australian National University, Wiley Online Library on [10/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
VALUING THE SPEED OF ADJUSTMENT | 859

In addition to providing a link to firm value, there are other ancillary benefits to measuring SOAOC as opposed
to SOAL. The main benefit stems from the fact that firm value is a concave function of leverage and studies
overwhelmingly use linear partial‐adjustment models. This difference is highlighted in Table B.1 in Online
Appendix B. Notably, the magnitude of the difference between SOAOC and SOAL is greatest when firms are
operating on the steepest part of the firm value versus leverage curve (i.e., when they are far from their targets).
This is consistent with studies such as Mukherjee and Wang (2013) who find that SOAL is an increasing function of
the deviation from the target capital structure—meaning that the most popular SOAL partial‐adjustment models are
misspecified because they inherently assume that SOAL is independent of the distance from the target. Studying
the dynamics of OC(L) avoids the misspecification inherent in most SOAL studies. Moreover, because managers are
more likely to make capital structure decisions based on their expected impact on firm value, it is more appropriate
to study the dynamics of OC(L) directly, especially when using a linear partial‐adjustment model.

3 | ES TIMA TIN G OC(L )

This section first details the sample selection procedure and describes the data. It then describes the procedures for
estimating the marginal benefit functions, marginal cost functions, and OC(L).

3.1 | Data and sample selection

I start with all available Compustat firms from fiscal years 1971 through 2012 and drop firms in regulated or
financial industries, foreign firms, and firms with nonpositive total assets, equity, or sales. I also require 3 years of
historical earnings data to generate the marginal benefit functions. The final sample consists of 84,177 firm‐year
observations with nonmissing data. It includes 12,571 firms with an average tenure of about 12.5 years. Because
I am using a dynamic panel to estimate the partial‐adjustment models, the SOAOC regressions (Equation (2)) have
62,603 observations from 10,314 firms over 1975–2012.

3.2 | Marginal benefit of leverage functions

The approach to estimate the marginal benefit functions assumes that the primary benefit of debt financing is the
tax deductibility of interest rates. I follow Blouin et al. (2010) and simulate the firm‐year‐specific tax benefit of
leverage functions. Their work builds on Shevlin (1990), Mackie‐Mason (1990), Graham(1996a, 1996b, 2000), and
others. Assuming debt in perpetuity, the marginal benefit of debt is then equal to the marginal tax rate (MTR), which
is defined as the present value of expected taxes paid on an additional dollar of income.
The goal of the simulation procedure is to estimate firm‐year‐specific MTR curves as a function of debt to
proxy for the marginal benefit of leverage. This procedure is rather involved because the tax code permits
carryforwards and carrybacks of taxable income over time. Also, the tax code is nonlinear with respect to income.
Therefore, estimating the MTR requires an estimate of future income and knowledge of past income and taxes
paid.7 The details of this procedure are given in Online Appendix B.
Table 1 presents the results of the MTR simulations that form the marginal benefit of leverage functions. The
before‐financing MTRs (nointMTR) are the MTRs assuming the firms made zero interest payments (0% of current
interest), and the actual MTRs (intMTR) are the simulated MTRs calculated at the current interest expense levels

7
Because the US tax code allows carryforwards of up to 20 years and carrybacks of up to 2 years, in my sample, income 22 years ahead could conceivably
affect the current MTR.
 14756803, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jfir.12300 by Australian National University, Wiley Online Library on [10/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
860 | JOURNAL OF FINANCIAL RESEARCH

TABLE 1 Marginal tax rate simulations

My simulations BCG simulations

Full sample BCG sample BCG sample
Statistic nointMTR intMTR nointMTR intMTR nointMTR intMTR

Obs. 84,177 84,177 58,202 58,202 58,202 58,202

Mean 31.5% 28.5% 29.5% 26.2% 29.7% 26.2%

SD 11.8% 12.8% 10.8% 11.7% 11.5% 12.7%

Min. 0.0% 0.0% 0.0% 0.0% 0.0% 0.0%

P1 3.2% 1.8% 3.4% 1.9% 3.2% 1.3%

P5 8.7% 5.8% 8.4% 5.5% 7.3% 4.2%

P10 12.9% 9.1% 12.2% 8.5% 10.8% 6.8%

P25 26.1% 18.4% 23.8% 16.3% 23.8% 14.6%

P50 33.4% 31.2% 32.5% 29.4% 33.1% 30.2%

P75 39.9% 35.0% 35.0% 34.0% 35.0% 34.6%

P90 47.0% 46.0% 44.1% 40.0% 45.0% 42.9%

P95 48.0% 48.0% 46.0% 45.1% 46.0% 45.6%

P99 48.0% 48.0% 46.0% 46.0% 46.0% 46.0%

Max. 51.0% 51.0% 51.0% 51.0% 51.0% 51.0%

Note: This table describes the sample of simulated marginal tax rates (MTRs). My simulated MTRs are given in the first two
columns. I follow the methodology of Blouin et al. (2010) (BCG) to estimate the MTRs and give the descriptive statistics for
those authors’ estimated MTRs (available through WRDS) in the third and fourth columns for all firm‐year observations that
I could successfully merge. nointMTR is the before‐financing MTR, assuming the firm made zero interest payments; intMTR
is the MTR at the current level of interest expense.

(100% of current interest). I report detailed summary statistics of the nointMTR and intMTR for brevity, but the
MTRs are calculated at 19 discrete points. For comparison, I also report summary statistics for the MTRs from
Blouin et al. (2010) that I am able to successfully merge with my sample.8
There is wide dispersion in MTRs as they range from 0% to 51%. Although not explicit in Table 1, this difference
arises from both cross‐sectional and times‐series variation. Comparing nointMTR with intMTR shows that the mean
firm in my sample decreases its MTR by roughly 3.0% (31.5% − 28.5%) by writing off interest expense. This
suggests that the mean firm is currently at a level of debt that lies on the downward‐sloping portion of the marginal
benefit of debt curve. Finally, the distribution of the estimated MTRs closely resembles that of Blouin et al. (2010).

3.3 | Marginal cost of leverage functions

With the marginal benefit functions defined, the next step is to estimate the marginal cost functions. I follow
Van Binsbergen et al. (2010) and use exogenous variation in the marginal benefit functions to identify the
marginal cost functions. I assume that, on average, observed debt ratios represent the level of debt that
equates the marginal cost and benefit functions. I then use cross‐sectional and times‐series variation in
the tax benefit functions to identify the marginal cost functions. The identification strategy requires the cost

8
I thank the authors for making their MTRs available. Note that the difference in sample size is primarily because the Blouin et al. (2010) sample spans
1980–2007.
 14756803, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jfir.12300 by Australian National University, Wiley Online Library on [10/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
VALUING THE SPEED OF ADJUSTMENT | 861

curve to remain fixed while the tax benefit function varies. To ensure this is the case, I include control
variables associated with the costs and benefits of debt.9 This implies that even if the tax deductibility of
interest is not the primary benefit of debt financing (e.g., Li et al., 2016), the model is valid so long as the
marginal cost functions can be properly identified from exogenous shifts in the tax benefit functions, which
requires only that taxes play a nontrivial role in debt decisions. The details of this procedure are given in
Online Appendix B.
Table 2 presents descriptive statistics for the variables used to identify the marginal cost functions for the full
sample (Panel A) and a financially unconstrained, nondistressed sample (Panel B). I estimate the marginal cost
functions on a financially unconstrained, nondistressed sample of firms because they are the subset of observations
most likely to be making optimal capital structure choices. To be included in this sample, an observation must be in
the top tercile of the sample in terms of debt or equity issuance or repurchase (DISS, DRED, EISS, and/or ERED
must be in the top tercile of the full sample) and have a ZSCORE in the top tercile of the sample. This is the same
definition as in Van Binsbergen et al. (2010). All variables are defined in the Appendix and are winsorized at the
1st and 99th percentiles.
One difference in my sample is that the estimated AREA_I variable (0.023 for my full sample in Panel A of
Table 2) is smaller than that reported by Van Binsbergen et al. (2010; 0.033 in their table II, panel A). This is
expected as the nonparametric method I am using to forecast income produces marginal benefit of leverage
functions that indicate lower total benefits of debt, and hence adjusting, than those reported by Graham (2000) and
Van Binsbergen et al. (2010).10 In fact, Blouin et al. (2010) report that the nonparametric methodology produces net
benefit of debt estimates that are about 43% of those produced by Graham's (2000) methodology used by Van
Binsbergen et al. (2010).
Table 3 presents estimation results for the marginal cost of leverage functions. The results in each column
can be interpreted as a function of interest over book value of assets (IOB) and an intercept. AREA_I is used as
the instrument for IOB, so the coefficient on IOB represents the slope of the marginal cost function. The
control variables determine the firm‐year‐specific intercept of the marginal cost function. I estimate three
specifications for each sample: instrumental variable regression (IVREG), firm fixed effects (Firm FE), and year
fixed effects (Year FE).
I find positive and significant coefficients on IOB in each column of Table 3, which means that the marginal
cost function is upward sloping. The slope of the line exhibits some variability and is somewhat smaller for the
unconstrained, nondistressed sample compared to the full sample. This is consistent with Van Binsbergen et al.
(2010), and it makes sense that unconstrained firms are less sensitive to fluctuations in capital structure.
Table B.1 in Online Appendix B reports marginal cost and benefit functions for the mean firm in the full and
unconstrained, nondistressed samples. Overall, the results are reasonable in that the marginal benefit
functions are downward sloping whereas the marginal cost functions are upward sloping. The difference
between the marginal cost and benefit functions is also greater for the full sample than for the unconstrained,

9
The vector of control variables includes well‐established costs and benefits of debt. Therefore, the estimated marginal cost functions are not
strictly marginal cost curves. Rather, they also account for nontax benefits of debt. Because I am ultimately interested in the opportunity cost, this
should not influence the results. The reason is that assuming that the calibrated cost function is a linear function of leverage, the control variables
affect only the intercept, and the variables with negative coefficients can be interpreted as benefits of debt financing. Because the opportunity cost
is the area between the marginal cost and benefit functions, it does not matter whether the marginal benefit function is shifted down or the marginal
cost function is shifted up; the area between the curves will be the same so long as the independent variables are uncorrelated with the instrument
for leverage.
10
In addition to the full and unconstrained, nondistressed samples, I estimate the marginal cost functions using a cross‐validation approach
common in the machine learning literature. The approach is to estimate the marginal cost functions on a randomly selected 10% subsample of
firms and then estimate the partial‐adjustment model on the 90% holdout sample in the following step. This is repeated 10 times (once for each
subsample). The rationale is that it is possible that estimating the marginal cost functions on the same sample that is being used to fit the partial‐
adjustment models in the next step could bias the results. The results are not materially different from the reported results and are available upon
request.
 14756803, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jfir.12300 by Australian National University, Wiley Online Library on [10/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
862 | JOURNAL OF FINANCIAL RESEARCH

TABLE 2 Descriptive statistics

Variable N Mean σ Min. Median Max.

Panel A: Full sample

IOB 84,177 0.022 0.021 0.000 0.017 0.102

MKL 84,177 0.255 0.244 0.000 0.190 0.887

BKL 84,177 0.224 0.185 0.000 0.205 0.716

COL 84,177 0.458 0.241 0.015 0.481 0.920

LTA 84,177 5.092 2.065 0.792 4.981 10.300

BTM 84,177 0.853 0.822 0.030 0.602 4.754

INTANG 84,177 0.096 0.154 0.000 0.020 0.693

CF 84,177 0.071 0.210 −1.006 0.117 0.398

DDIV 84,177 0.343 0.475 0.000 0.000 1.000

AREA_I 84,177 0.025 0.020 0.000 0.022 0.095

ZSCORE 84,177 1.332 3.019 −15.050 1.959 5.968

UNFC 84,177 0.763 0.425 0.000 1.000 1.000

DISS 84,177 0.084 0.163 0.000 0.008 0.975

DRED 84,177 0.076 0.154 0.000 0.020 1.002

EISS 84,177 0.066 0.181 0.000 0.003 1.046

ERED 84,177 0.011 0.032 0.000 0.000 0.202

Panel B: Unconstrained sample

IOB 19,604 0.017 0.016 0.000 0.014 0.102

MKL 19,604 0.209 0.213 0.000 0.141 0.887

BKL 19,604 0.175 0.144 0.000 0.162 0.716

COL 19,604 0.500 0.210 0.015 0.524 0.920

LTA 19,604 5.313 1.699 0.792 5.209 10.300

BTM 19,604 0.824 0.730 0.030 0.598 4.754

INTANG 19,604 0.057 0.094 0.000 0.013 0.693

CF 19,604 0.183 0.091 −1.006 0.178 0.398

DDIV 19,604 0.530 0.499 0.000 1.000 1.000

AREA_I 19,604 0.029 0.022 0.000 0.026 0.095

ZSCORE 19,604 3.448 0.812 2.556 3.205 5.968

UNFC 19,604 1.000 0.000 1.000 1.000 1.000

DISS 19,604 0.084 0.172 0.000 0.009 0.975

DRED 19,604 0.081 0.174 0.000 0.020 1.002

 14756803, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jfir.12300 by Australian National University, Wiley Online Library on [10/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
VALUING THE SPEED OF ADJUSTMENT | 863

TABLE 2 (Continued)

Variable N Mean σ Min. Median Max.

EISS 19,604 0.035 0.098 0.000 0.004 1.046

ERED 19,604 0.021 0.043 0.000 0.001 0.202

Note: This table gives the descriptive statistics for the variables used to identify the marginal cost of debt (MC(L))
functions. Panel A presents the full sample. Panel B includes only unconstrained, nondistressed firms. To be included
in Panel B, an observation must be in the top tercile of the sample in terms of equity or debt issuance or repurchase
(DISS, DRED, EISS, and/or ERED must be in the top tercile of the full sample) and have a ZSCORE in the top tercile of
the sample. IOB is interest expense over book value of assets, MKL is market leverage, BKL is book leverage, COL is
collateralizable assets over total book value of assets, LTA is log of total assets (2000 dollars), BTM is book‐to‐market
value, INTANG is intangible assets over total assets, CF is cash flow over total assets, and DDIV is a dummy equal to 1
if the firm paid a dividend, and 0 otherwise. AREA_I is the area below the marginal tax rate curves from 0% to 2000%
of IOB. ZSCORE is Altman's Z‐score. UNFC is an indicator equal to 1 if the observation is in the unconstrained,
nondistressed sample. DISS, DRED, EISS, and ERED are long‐term debt issuance, long‐term debt reduction, equity
issuance, and equity reduction, respectively, all given as a fraction of total assets. All variables are winsorized at the
1st and 99th percentiles.

nondistressed sample, which implies that OC(L) is less for an unconstrained, nondistressed firm. Again, this
makes intuitive sense and suggests that financially constrained or distressed firms should benefit more relative
to unconstrained, nondistressed firms from the same SOA policy.

3.4 | Opportunity cost of leverage

The final step before estimating SOAOC is to compute OC(L). It is computed for each firm‐year observation by first
finding the level of debt that minimizes the difference between the marginal cost and benefit of leverage functions.
This point is defined as L*. I integrate the area between the marginal cost and benefit functions from the current
interest expense to the interest at the optimal leverage ratio, L*, to get OC(L) for each firm‐year observation. OC(L)
is defined as a fraction of total assets.
Table 4 presents descriptive statistics for OC(L) as a fraction of total assets. To benchmark the results to Van
Binsbergen et al. (2010), a few points are necessary. First, in their table VIII, Van Binsbergen et al. (2010) refer to
what I am terming OC(L) as the equilibrium net benefit of debt (NBDe). The relevant summary statistics are given in
the last row of their panel A. Second, they present descriptive statistics as a fraction of total book assets in
perpetuity by dividing the annual NBDe by the mean Moody's bond rating, which is about 9% for my sample. Finally,
they define the NBDe as a strictly positive value.
I report descriptive statistics for the raw annual OC(L) in Panel A of Table 4. To transform my reported
statistics to perpetual values, divide by the mean Moody's Baa bond rate over the sample period, which is
about 9%. This results in a mean perpetual OC(L) of between 1.04% and 1.50%, which is only 30%–43% of the
3.52% reported by Van Binsbergen et al. (2010) in their table VIII. The reason for the difference is that Blouin
et al.'s (2010) nonparametric approach used in this article produces flatter marginal benefit of debt functions,
implying a lower OC(L). As such, my results are likely to, if anything, understate the benefits of adjusting
capital structure. Finally, although Blouin et al. (2010) do not estimate OC(L), they do report that their method
of estimating the marginal tax benefit functions results in net benefits of debt that are about 43% of those
reported by Van Binsbergen et al.11 As such, my estimates of OC(L) are reasonable when benchmarked to the
literature.
 14756803, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jfir.12300 by Australian National University, Wiley Online Library on [10/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
864 | JOURNAL OF FINANCIAL RESEARCH

TABLE 3 Marginal cost of leverage estimation

Unconstrained, nondistressed sample Full sample

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

IOB 3.242 3.213 0.918 3.585 4.669 1.884

(0.124) (0.130) (0.085) (0.081) (0.076) (0.059)

COL −0.015 −0.003 −0.009 −0.033 −0.025 −0.023

(0.002) (0.002) (0.001) (0.001) (0.001) (0.001)

LTA 0.024 −0.003 0.028 0.024 0.018 0.024

(0.002) (0.003) (0.001) (0.001) (0.001) (0.001)

BTM 0.019 0.017 −0.003 0.009 0.008 −0.002

(0.002) (0.001) (0.001) (0.001) (0.001) (0.001)

INTANG −0.024 −0.013 −0.006 −0.027 −0.02 −0.015

(0.002) (0.002) (0.001) (0.001) (0.001) (0.001)

CF 0.086 0.076 0.051 0.065 0.058 0.057

(0.003) (0.002) (0.002) (0.001) (0.001) (0.001)

DDIV 0.052 0.038 0.019 0.085 0.06 0.053

(0.002) (0.002) (0.002) (0.002) (0.002) (0.001)

Constant 0.219 0.236 0.417 0.178 0.163 0.364

(0.003) (0.003) (0.004) (0.002) (0.002) (0.003)

Fixed effects No Firm Year No Firm Year

Obs. 19,604 84,177

No. of firms 4806 12,571

Note: Each column gives the results of the marginal cost of leverage (MC(L)) estimates. AREA_I is the identifying instrument
for the interest expense over book assets ratio (IOB), x*i, t ; C is the set of control variables; and y*i, t is the observed marginal
benefit of debt (MTR). Columns 1–3 are estimated using the unconstrained, nondistressed sample, and Columns 4–6 are
estimated on the full sample. COL is collateralizable assets over total book value of assets, LTA is log of total assets (year
2000 dollars), BTM is book‐to‐market value, INTANG is intangible assets over total assets, CF is cash flow over total assets,
and DDIV is a dummy equal to 1 if the firm paid a dividend, and 0 otherwise. All of the control variables, except DDIV, are
standardized to have 0 mean and 1 SD. Robust, clustered standard errors are given in parentheses.

4 | ES TIMA TIN G SOA

In estimating SOAOC, I assume that the target OC(L) is zero, which is to say, firms attempt to maximize value. One
benefit of this assumption is that it simplifies Equation (2) as the target (OC(L*) = 0) drops out. However, it also
creates a new challenge because OC(L) is always positive by definition. This is problematic because testing for mean
reversion implies that the target is the mean. Because the target is equal to zero for value‐maximizing firms, the
partial‐adjustment model given by Equation (2) would be misspecified if the mean opportunity cost is greater
than zero.
To circumvent this problem, I define OC(L) to be nonnegative for underlevered observations and nonpositive
for overlevered observations.12 This allows for the possibility that the estimated SOAOC exceeds 100%, which goes

11
The 43% is calculated as the weighted average in column (6) of panel A of table 7 in Blouin et al. (2010), which is the ratio NP benefit/RW benefit. NP
benefit refers to Blouin et al. (2010) parametric method, and RW benefit refers to Van Binsbergen et al.'s (2010) random walk method.
 14756803, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jfir.12300 by Australian National University, Wiley Online Library on [10/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
VALUING THE SPEED OF ADJUSTMENT | 865

TABLE 4 Distribution of opportunity cost of leverage

Unconstrained, nondistressed sample Full sample

IVREG Firm FE Year FE IVREG Firm FE Year FE
Statistic (1) (2) (3) (4) (5) (6)

Panel A: Distribution of raw OC(L)

Mean 0.124% 0.135% 0.097% 0.116% 0.128% 0.094%

SD 0.223% 0.262% 0.189% 0.200% 0.238% 0.160%

Percentile

1st 0.000% 0.000% 0.000% 0.000% 0.000% 0.000%

5th 0.000% 0.000% 0.000% 0.000% 0.000% 0.000%

10th 0.001% 0.001% 0.000% 0.001% 0.001% 0.001%

25th 0.009% 0.009% 0.007% 0.010% 0.011% 0.008%

Med. 0.036% 0.039% 0.024% 0.042% 0.044% 0.032%

75th 0.125% 0.134% 0.088% 0.132% 0.140% 0.104%

90th 0.332% 0.361% 0.277% 0.303% 0.317% 0.256%

95th 0.597% 0.606% 0.501% 0.500% 0.490% 0.435%

99th 1.287% 1.554% 1.104% 1.183% 1.465% 0.898%

Panel B: Distribution of positive/negative OC(L)

Mean 0.050% 0.075% 0.045% 0.018% 0.032% 0.041%

SD 0.255% 0.285% 0.208% 0.231% 0.269% 0.186%

Percentile

1st ‐0.621% ‐0.359% ‐0.497% ‐0.522% ‐0.432% ‐0.554%

5th ‐0.180% ‐0.165% ‐0.132% ‐0.245% ‐0.237% ‐0.226%

10th ‐0.105% ‐0.102% ‐0.069% ‐0.159% ‐0.160% ‐0.136%

25th ‐0.025% ‐0.025% ‐0.018% ‐0.056% ‐0.059% ‐0.046%

Med. 0.000% 0.000% 0.000% ‐0.002% ‐0.005% ‐0.003%

75th 0.054% 0.065% 0.037% 0.026% 0.023% 0.016%

90th 0.253% 0.321% 0.210% 0.182% 0.217% 0.130%

95th 0.496% 0.606% 0.436% 0.401% 0.490% 0.307%

99th 1.287% 1.554% 1.104% 1.183% 1.465% 0.898%

Obs. L > L* 30,147 30,260 29,890 24,898 23,744 23,490

Obs. L = L* 5495 5606 5971 5223 4761 5511

Obs. L < L* 26,961 26,737 26,742 32,482 34,098 33,602

Note: This table gives the distribution of the opportunity cost of leverage (OC(L)) for the full sample of firms. Panel A gives
the raw (positive) estimates. In Panel B, negative values represent observations in which the firm is underlevered (Li,t < L*)
and positive values represent observations that are overlevered. For each column, the marginal cost of leverage (MC(L))
function used to calculate OC(L) corresponds to the analogous column in Table 3. The marginal benefit (MB(L)) curves used
are derived from the marginal tax rate estimates. The OC(L) functions are estimated by integrating the area between MB(L)
and MC(L) from the current level of interest expense over book assets (IOB) to the level of IOB that would maximize firm
value. The statistics describe the absolute value of the OC(L) for comparison to Van Binsbergen et al.'s (2010). IVREG is
instrumental variable regression, Firm FE is firm fixed effects, and Year FE is year fixed effects.
 14756803, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jfir.12300 by Australian National University, Wiley Online Library on [10/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
866 | JOURNAL OF FINANCIAL RESEARCH

against economic intuition because it should not be possible to overadjust, at least not in terms of firm value.
Therefore, to the extent that firms are consistently overadjusting, the SOAOC estimates may be biased upward.
However, if this were a significant concern, then as shown in Section 2, SOAOC should be less than SOAL, which is
contrary to my findings. Additionally, this transformation helps mitigate problems associated with estimating partial‐
adjustment models with a bounded ratio as a dependent variable (for a description of the problem, see, e.g., Chang
& Dasgupta, 2009; Elsas & Florysiak, 2015; Iliev & Welch, 2010).13 The transformed sample used to estimate
Equation (2) is described in Panel B of Table 4.
Although there is a large literature devoted to dynamic panel estimation techniques, the focus in my article is
not to determine or derive an optimal or unbiased estimator but rather to estimate the partial‐adjustment models
using GMMBB. I conduct robustness tests using several of the most prominent techniques in the literature so I can
benchmark the results and provide a range of estimates. In particular, I test OLS and firm fixed‐effects models
because Flannery and Hankins (2013) show that in the context of estimating SOA, OLS produces a negative bias,
firm fixed effects produces a positive bias, and GMMBB provides the most accurate estimate of SOA. Iliev and
Welch (2010) report similar results. Therefore, the OLS and firm fixed‐effects results should provide lower and
upper bounds on the true SOA, and GMMBB should produce the most accurate estimate. I report only the GMMBB
results; the OLS and firm fixed‐effects model results are available upon request.
Table 5 presents the SOAOC results for the full sample. All models are estimated using GMMBB. Panel A
presents the results for the full sample. For robustness, I estimate Equation (2) for each combination of the two
samples used to fit the marginal cost functions (full and unconstrained, nondistressed) and the three econometric
techniques used to fit each of these marginal cost functions (IVREG, Firm FE, and Year FE). Estimated values of
SOAL using Equation (1) are reported in the last row. The variables used to instrument for the target leverage ratio
are suppressed for brevity.
Mean SOAOC is 47.3% across the six models for the full sample. The estimates range from 41.3% to 53.5%,
depending on the specification used to identify the marginal cost of leverage functions. To benchmark to the
literature, I estimate SOAL to be 29.1% for the full sample, which is similar to numbers reported in the literature.
These numbers imply that the ratio of SOAOC to SOAL is about 1.6, depending on the estimation methodology. In
Table B.1 of Online Appendix B, I show that this ratio should be about 1.7 assuming linear marginal cost and benefit
of leverage functions, so the SOAOC estimates appear reasonable and the differences are a result of the change in
measure.
When interpreting SOA estimates, it is common to discuss the results in terms of the half‐life, or the length of
time it takes to close half the gap between current leverage and target leverage. In this context, it is not uncommon
for researchers to reasonably assume that if the half‐life is less than one, firms are relatively active in terms of
adjustment. The SOAOC estimates in Table 5 correspond to a half‐life of about 1.1 years. What this means is that a
firm that suffers a shock to its capital structure that causes the firm's value to become suboptimal will recoup about
half of that lost value in the year immediately following the shock.
In Panels B and C of Table 5, I reestimate the SOAOC models using restricted samples. I first drop all
zero‐leverage firms from the sample because there is a group of firms that persistently carry no leverage even when
they are profitable and hence forgo lucrative tax benefits of debt. The literature has yet to come to a consensus as
to why this phenomenon occurs, and it is beyond the scope of this article to solve this problem (see, e.g., Strebulaev
& Yang, 2013). The results are virtually unchanged from the full sample. In particular, the mean SOAOC is about
46.3%. Therefore, zero‐leverage firms are not biasing the results.

12
This is analogous to Van Binsbergen et al.'s (2010) definition of NBDe. The adjustment has to be made only because I refer to both positive and negative
deviations to leverage as positive opportunity costs.
13
OC(L) is practically unbounded when defined as nonnegative for Lt < L* and nonpositive for Lt > L*. Technically, it is bounded by +1 and −1 as firm value
cannot be negative, but the OC(L) in my sample never approaches either of these bounds.
 14756803, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jfir.12300 by Australian National University, Wiley Online Library on [10/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
VALUING THE SPEED OF ADJUSTMENT | 867

TABLE 5 Speed of adjustment results

Adjustment variable MC(L) sample MC(L) model Half‐life SOA S.E. a S.E.

Panel A: Full sample (N = 62,603)

SOAOC estimates

OC(L) Full IVREG 1.06 47.9% 0.015 0.000 0.000

OC(L) Full Firm FE 1.18 44.4% 0.017 0.000 0.000

OC(L) Full Year FE 0.91 53.5% 0.013 0.000 0.000

OC(L) Unconstrained IVREG 1.12 46.1% 0.016 0.000 0.000

OC(L) Unconstrained Firm FE 1.30 41.3% 0.017 0.000 0.000

OC(L) Unconstrained Year FE 0.99 50.5% 0.014 0.000 0.000

SOAL estimates

BKL N/A N/A 2.02 29.1% 0.028 ‐0.093 0.031

Panel B: Drop zero‐leverage firms (N = 53,413)

SOAOC estimates

OC(L) Full IVREG 1.11 46.6% 0.016 0.000 0.000

OC(L) Full Firm FE 1.23 43.1% 0.018 0.000 0.000

OC(L) Full Year FE 0.93 52.6% 0.015 0.000 0.000

OC(L) Unconstrained IVREG 1.15 45.4% 0.017 0.000 0.000

OC(L) Unconstrained Firm FE 1.34 40.5% 0.017 0.000 0.000

OC(L) Unconstrained Year FE 1.01 49.8% 0.015 0.000 0.000

SOAL estimates

BKL N/A N/A 2.49 24.3% 0.03 ‐0.085 0.035

Panel C: Long‐life firms (N = 25,098)

SOAOC estimates

OC(L) Full IVREG 1.08 47.4% 0.022 0.000 0.000

OC(L) Full Firm FE 1.23 43.0% 0.023 0.000 0.000

OC(L) Full Year FE 0.89 54.0% 0.019 0.000 0.000

OC(L) Unconstrained IVREG 1.12 46.0% 0.022 0.000 0.000

OC(L) Unconstrained Firm FE 1.27 42.0% 0.022 0.000 0.000

OC(L) Unconstrained Year FE 0.91 53.4% 0.02 0.000 0.000

SOAL estimates

BKL N/A N/A 2.74 22.3% 0.028 ‐0.039 0.036

Note: This table summarizes the speed of adjustment (SOA) estimates for the full sample, dropping zero‐leverage firms, and
firms with at least 10 consecutive observations in Panels A, B, and C, respectively. SOAOC is the SOA estimated on a cost
basis and is defined by Equation (2). SOAL is estimated on a leverage ratio basis and is defined by Equation (1). MC(L) is
marginal cost of leverage, and OC(L) is opportunity cost of leverage. All models are estimated using the Blundell and Bond
(1998) system generalized method of moments (GMMBB). I report the SOA coefficients and constant terms for each
equation for brevity. Full estimation results are available upon request (i.e., the vector of β coefficients in SOAL models).
Half‐life is defined as log(0.5)/log(1 − SOA OC ) or log(0.5)/log(1 − SOAL ). IVREG is instrumental variable regression, Firm FE
is firm fixed effects, and Year FE is year fixed effects.
 14756803, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jfir.12300 by Australian National University, Wiley Online Library on [10/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
868 | JOURNAL OF FINANCIAL RESEARCH

I next retain only sequences where a firm appears in the data for at least 10 consecutive years. The reason is
that fitting a partial‐adjustment model to a panel with insufficient time‐series observations can lead to biased
estimates (see Flannery & Hankins, 2013). I refer to this as the “long‐life” sample, and the results are given in Panel
C. The findings are consistent with my primary results as the mean SOAOC is 47.6%. Overall, the results appear to
be robust to a wide range of alternative settings and are consistent with the hypothesis that firms actively adjust
toward target capital structures.

5 | O P T I M A L SO A

The preceding results show that firms actively adjust toward target capital structures as the mean SOAOC is found
to be about 47%. However, my primary purpose is to examine the relation between SOAOC and firm value and
ultimately examine what, if any, SOAOC is optimal. In this section, I use a simulation to answer these questions.
The simulation is built on the predicate that the residuals from the partial‐adjustment regressions (Equation (2))
represent the effect of random shocks to capital structure on OC(L). They are therefore orthogonal to OC(L) and
hence the SOAOC estimates. As such, they can be used to proxy shocks to capital structure and to simulate a
random time series of OC(L) for any desired SOAOC parameter value (i.e., SOAOC = 0%, 1%, …, 100%). The value of a
simulated SOAOC policy relative to random financing (i.e., SOAOC = 0%) should be equal to the difference between
the simulated OC(L) and what OC(L) would have been had the firm not adjusted (i.e., the value gain of adjusting).
Estimating the observed annual OC(Li,t) is straightforward as I have computed it for each observation in the sample.
However, estimating the hypothetical value of OC(Li,t) in a world without adjustment requires simulation because
there is no directly observable counterfactual sample of firms that follow a random financing strategy. Therefore, to
simulate the economic value of a given SOAOC policy, I assume firms start out at their target capital structures so
OC(Li,0) = 0 and do the following:

1. Estimate Equation (2) and save the parameter estimates and residuals.
2. Choose a hypothetical SOAOC policy, SOAOC* (e.g., SOAOC* = 47.9%).
3. Bootstrap the residuals from the SOAOC estimations given in Table 5, substitute into Equation (2) to get a
distribution of OC(Li,1), and repeat recursively to get a series of i paths of OC(Li,t) for t = 1, 2, 3, …, 30 years.14
Compute the net present value (NPV) of the absolute value of each path of OC(Li,t) using the mean Moody's Baa
discount rate over my sample period (9%). This creates a distribution of the gross cost of a given adjustment
policy (SOAOC*), NPV(SOA*).
4. Repeat Steps 2–4 for SOAOC* values ranging from random financing through complete adjustment to get a
series of distributions of NPV(j), for j = 0%, 1%, …, 100%.

Figure 1 illustrates the process to help visualize the simulation procedure. Each of the six subplots shows a
series of random realizations of OC(Li,t) described in Step 3 for 30 years of shocks to capital structure. The shocks to
capital structure are a series of randomly drawn residuals from Equation (2) and are the same within each subplot.
The different lines in each subplot represent realizations of OC(Li,t) under different SOAOC* assumptions from 0% to
100%. The most volatile lines represent random financing, and the least volatile lines represent complete
adjustment. Thus, each line represents one of the i realizations of OC(Li,t) for one of the j SOAOC*s. As shown,
the benefit of increasing SOAOC is to reduce the volatility of OC(Li,t) and hence reduce the value loss resulting from
the random shocks to capital structure.

14
I simulate i = 62,603 paths because that is the number of observations in my sample. The results are stable if I increase the number of simulated paths.
 14756803, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jfir.12300 by Australian National University, Wiley Online Library on [10/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
VALUING THE SPEED OF ADJUSTMENT | 869

F I G U R E 1 Opportunity cost of leverage OC(L) simulation example. This figure plots six random realizations of
OC(L) using the results of the speed of adjustment (SOA) estimates in the first row in Panel A of Table 5 under
varying SOA policies. Each subplot shows the realization of the OC(L) for 30 years of random shocks to capital
structure for SOA policies ranging from 0%, 10%, 20%, …, 100%. Within the subplots, the random shocks are the
same for each assumed SOA path. The most volatile paths are those where the SOA is assumed to be random
financing or 0% SOA. The least volatile paths represent complete adjustment or 100% SOA [Color figure can be
viewed at wileyonlinelibrary.com]

The next step in valuing a given SOAOC* policy is to apply Step 4 to each line in each subplot and find the
discounted sum of the absolute value of the random realizations of OC(L). This results in a distribution of size i of
the values of the gross OC(L) for each of the j SOAOC*s, or NPV(SOAj). Finally, the mean value of a given SOAOC*
policy is equal to the difference between NPV(SOAj) and NPV(SOAj=0%). That is, the value of a given SOAOC* policy
represents the marginal savings compared to a random financing strategy without adjustment.
Although the simulation process is conceptually simple, it does require a few assumptions about the data. First,
Equation (2) is fit to a sample of observations where OC(L) is defined as nonnegative for overlevered firms and
nonpositive for underlevered firms. However, in reality, it is always a positive cost. I only define it in such a way to
 14756803, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jfir.12300 by Australian National University, Wiley Online Library on [10/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
870 | JOURNAL OF FINANCIAL RESEARCH

F I G U R E 2 Mean annual opportunity cost of leverage (OC(L)) by speed of adjustment (SOA) policy. The vertical
axis is expressed as a percentage of total assets. All simulations assume that firms start out at their optimal capital
structure at time 0. [Color figure can be viewed at wileyonlinelibrary.com]

fit the partial‐adjustment models. Therefore, I simulate OC(L) for various hypothetical values of SOAOC* using the
positive/negative definition in Step 3 but then discount the absolute value of OC(L) when computing the NPV of
each path in Step 4.
Second, completely random financing (i.e., SOAOC = 0%) implies a unit‐root process, which means that the
variance of OC(Li,t) goes to infinity as the number of years of simulated data (t) goes to infinity. For illustration,
Figure 2 plots the mean annual OC(L) for various simulated values of SOAOC*. The uppermost curve (SOAOC* = 0%)
highlights the unit‐root process as the mean annual OC(L) increases steadily through 30 years (and beyond). The
other paths all reach steady‐state values within the time horizon shown. The reason that the paths start at about
0.1% of total assets is that the mean absolute residual from Equation (2) is about 0.1% of total assets. Therefore,
even a firm that completely adjusts annually is subject to these random shocks and is, on average, operating slightly
out of equilibrium.15
I do not make any assumptions about the specific distribution of the residuals from Equation (2) other than to
assume they are independent and identically distributed. I could assume the residuals are, for example, normally
distributed but choose not to because so long as the residuals are not biased and are independent and identically
distributed, the GMMBB models are well specified. Therefore, bootstrapping from the residuals gives the most
realistic simulation.
To recap, the result of the simulation is a series of distributions of the NPV of OC(L) for any SOAOC policy
ranging from random financing through complete adjustment, expressed as a percentage of total assets. The main
advantage of this methodology is that I can use these simulated distributions to compare the relative benefit
between any two adjustment policies. Most notably, I can compare the cost of the observed SOAOC policy to the
cost of any hypothetical simulated adjustment policy by taking the difference between the expected values of the
simulated distributions (i.e., E[NPV(j)] − E[NPV(47.9%)]). I refer to this as the net benefit of the observed SOAOC
policy, and it represents the expected value loss or gain that a firm would incur if it were to adhere to a different

15
I address the unit root when computing the NPV of the simulated annual OC(L) paths as follows. I first simulate the distribution of the OC(L) for 30 years
and assign a terminal value beyond 30 years by assuming that the OC(L) in year 30 will continue in perpetuity. Although unrealistic for random financing, it
ends up being a relatively minor issue because the opportunity costs are discounted at 9% annually.
 14756803, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jfir.12300 by Australian National University, Wiley Online Library on [10/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
VALUING THE SPEED OF ADJUSTMENT | 871

F I G U R E 3 Benefit of adjusting versus speed of adjustment (SOA). This figure plots the mean net benefit of the
observed adjustment policy of 47.9% versus simulated SOA policies from 0% to 100%. The vertical axis represents
the relative benefit expressed as a percentage of total assets. A positive (negative) value on the net benefit curve
means the firm is better (worse) off adhering to the observed adjustment policy. All simulations assume that firms
start out at their optimal capital structure at time 0

SOAOC policy. A positive (negative) value means the firm is better (worse) off adhering to its actual SOAOC policy
vis‐à‐vis the comparison policy.
Figure 3 plots the net benefit of the observed adjustment policy of 47.9% versus simulated values of SOAOC*
ranging from 0% to 100%. The vertical axis represents the relative benefit expressed as a percentage of total assets.
The horizontal axis represents the comparison SOAOC* policy. Several aspects of this figure stand out. First, the net
benefit curve crosses 0% benefits at a value of SOAOC* of 47.9%, which is by definition. More important are the
extremes of the plot. The left side of the plot shows that if firms were to adhere to a policy of random financing, it
would cost them an additional 3.3% of total assets. However, moving in the opposite direction, the benefit declines
rapidly, and moving to a complete adjustment SOAOC* policy of 100% would net firms only an additional 0.40% of
total assets. This suggests that the mean firm chooses a SOAOC policy that locks in the majority of the benefits of
adjustment.

5.1 | Adjustment costs

The preceding simulations show the relation between OC(L) and SOAOC. The value of adjusting is decreasing at a
decreasing rate, and the results imply that firms benefit from complete adjustment. However, the preceding
simulation results do not include adjustment costs. As such, they overstate the benefits of adjusting and hence
imply that the optimal SOAOC is 100% because firms would obviously benefit from complete adjustment if doing so
were costless.
Therefore, I incorporate adjustment costs into the analysis in this section. The goal is to calibrate a realistic
relation between OC(L) and SOAOC such that firm value is a concave function of SOAOC and there exists an optimal
SOAOC. However, there are some practical limitations and key assumptions that need to be made to include
adjustment costs, which are summarized next.
First, there are several estimates of adjustment costs in the literature. Most studies focus on net issuance costs.
Altinkilic and Hansen (2000), for example, examine the behavior of seasoned equity offering (SEO) and bond
offering spreads paid to underwriters. They fit several reduced‐form econometric models that estimate bond and
 14756803, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jfir.12300 by Australian National University, Wiley Online Library on [10/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
872 | JOURNAL OF FINANCIAL RESEARCH

SEO spreads as functions of individual firm risk factors, issue size, and relative firm size. Hennessy and Whited
(2005) calibrate a structural dynamic model of capital structure to estimate equity floatation costs. Other papers
such as Lee and Masulis (2009) study the determinants of SEOs and provide robust estimates of equity floatation
costs.
I start with Altinkilic and Hansen's (2000) methodology for several reasons. First, they estimate spreads on debt
and equity issues separately, which is beneficial in the context of my study because adjustment toward an optimal
capital structure involves a net debt or equity issuance or reduction. Second, their reduced‐form models consider
both fixed and variable costs of adjustment, so the size of the adjustment matters. Therefore, ceteris paribus, firms
with higher SOAOC and/or more severe shocks to capital structure face greater costs of adjustment. Finally, the
models are tractable and can be incorporated into my simulation study.
For robustness, I use two model specifications to incorporate adjustment costs based on equations (9) and (10)
in Altinkilic and Hansen (2000). The first model is defined as

s = γ0 + γ1ln(x1), (3)

where s is the spread paid to equity or debt holders in percentage terms and x1 is the issuance size in millions of
dollars. The second model further includes firm size and is defined as

1 x1
s = β0 + β1 + β2 , (4)
x1 x2

where x2 is firm size as measured by the market value of equity in millions of dollars. The two models consider both
fixed (γ0 and β0) and variable costs of adjustment. The second model captures the effect of the relative size of the
issue with the third term. In my simulations, I use the parameter estimates from Altinkilic and Hansen's (2000)
Table 2 for equity issues and Table 5 for debt issues, as shown in Table 6.
Because the preceding models are only two estimates of adjustment costs, I repeat the simulations for various
levels of adjustment costs to gauge the sensitivity of the results to the chosen level of adjustment costs. In
particular, I assume adjustment costs to be completely variable and a fixed percentage of the size of the net debt or
equity issue. Consistent with Altinkilic and Hansen (2000), I assume the cost of issuing equity is about 5 times
greater than the cost of issuing debt. I then simulate OC(L) for various adjustment costs ranging from 0 to 200 basis
points for debt and 0 to 1000 basis points for equity issues. This essentially puts reasonable bound around the
adjustment costs while capturing the fact that the costs are variable and relative to the size of the issue.
With the functional form of the adjustment costs defined, the question remains as to how much of a firm's
actual capital structure adjustments are sunk costs and how much can be attributed to a given SOAOC policy—that
is, how much of the incurred adjustment costs are active versus passive (see Iliev & Welch, 2010). Fortunately, this

TABLE 6 Adjustment cost coefficients

Coefficient Equity coefficient Debt coefficient

γ0 8.69 2.60

γ1 −0.88 −0.30

β0 4.04 0.50

β1 25.65 25.17

β2 2.64 4.63

Note: This table presents the coefficients used to parameterize the adjustment cost models from Altinkilic & Hansen, (2000;
Equations (3) and (4) in the text). The equity coefficients are from Altinkilic and Hansen's (2000) Table 2, and the debt
coefficients are from their Table 5.
 14756803, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jfir.12300 by Australian National University, Wiley Online Library on [10/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
VALUING THE SPEED OF ADJUSTMENT | 873

issue is implicitly addressed in the preceding simulations because I compare the difference between the actual
adjustment implied by a policy of SOAOC* to completely passive management (SOAOC = 0%). As such, I consider
only the cost associated with the active portion of the capital structure adjustment associated with a given SOAOC*
policy and effectively net out the random shocks to capital structure that drive the simulation.
Finally, direct issuance costs apply only when a firm conducts a net equity or debt issuance. In many cases,
overlevered (underlevered) firms reduce leverage by paying down debt (repurchasing shares). In these cases, the
capital structure adjustment should be essentially costless. I address this issue in the simulations by drawing a
random variable to indicate whether the capital structure adjustment is a new issuance or reduction of debt or
equity. The random variable is drawn from a binomial distribution calibrated on my sample data. For overlevered
firms, the mean parameter is calibrated by comparing the likelihood of a net equity issuance (EISS) versus a net debt
reduction (DRED). The probability that a delevering adjustment is passive is 0.551, meaning that an overlevered
firm that adjusts toward its target does so by paying down debt about 55.1% of the time (i.e., adjustment costs are
incurred 44.9% of the time). For underlevered firms, the mean parameter is calibrated by comparing the likelihood
of a net debt issuance (DISS) versus a net equity reduction (ERED). The probability that an up‐levering adjustment is
passive is 0.296, meaning that an underlevered firm that adjusts toward its target does so by paying down equity
about 29.6% of the time. Importantly, this aspect of the simulation means that adjustment costs are incurred only if
a firm issues debt or equity.

5.2 | Optimal SOA results with adjustment costs

The results of the simulations that include adjustment costs are shown in Figures 4 and 5. The first set of results
plots the relation between OC(L) and SOAOC* for the two Altinkilic and Hansen (2000) models. The adjustment
costs in the top subfigure are calibrated using Equation (3) and the bottom subfigure costs are calibrated using
Equation (4). In each plot, the red line represents the simulated OC(L) associated with the SOAOC* policy on the
horizontal axis. The solid vertical lines represent the optimal SOAOC policy in each subfigure based on the
respective simulations. The dotted vertical lines represent the observed SOAOC of 47.9% from Table 5. Importantly,
for each adjustment cost model, OC(L) is a convex function of SOAOC, so there is an optimal SOAOC. Although this
result is not surprising, I am the first to empirically calibrate this curve.
The results in Figure 4 imply that the observed SOAOC policy of 47.9% is actually higher than the simulated
optimal SOAOC policy of 31.3%–34.9% shown in Figure 4. However, the observed SOAOC policies are on the
relatively flat portion of the curves, suggesting that firms are close to their optimal SOAOC policies, at least in terms
of firm value. For example, in the top subfigure of Figure 4, the optimal SOAOC policy of 31.3% implies an OC(L) of
2.84% of total assets, whereas the observed SOAOC policy of 47.9% implies an OC(L) of 2.95% of total assets. As
such, firms would stand to gain only about 0.11% in value of total assets by changing to the optimal SOAOC policy.
For the second model in the bottom subfigure, the optimal SOAOC policy of 34.9% implies an OC(L) of 3.13% of
total assets, whereas the observed SOAOC policy of 47.9% implies an OC(L) of 3.19% of total assets for a difference
of only 0.06%.
The main takeaway from Figure 4, therefore, is that although the simulated optimal SOAOC policy is slightly
lower than the observed SOAOC policy of 47.9%, the difference in terms of firm value is negligible. That is, once
adjustment costs are considered, there is a wide range of SOAOC policies that do not have a large effect on firm
value, as indicated by the relatively flat portions of the curves around the optimal levels.
Next, to gauge the sensitivity of optimal SOAOC policies to different levels of adjustment costs, I repeat the
preceding analysis for a wide range of adjustment costs and estimate the optimal SOAOC policy conditional on each
level of adjustment costs. The conditional optimal SOAOC policy is defined as the SOAOC that minimizes OC(L) for a
given level of adjustment costs. As shown next, the results provide a reasonable range of optimal SOAOC policies for
 14756803, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jfir.12300 by Australian National University, Wiley Online Library on [10/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
874 | JOURNAL OF FINANCIAL RESEARCH

F I G U R E 4 Opportunity cost of leverage (OC(L)) as a percentage of total assets versus various speed of
adjustment estimated on a cost basis (SOAOC) policies ranging from 0% to 100% accounting for adjustment costs.
Adjustment costs are calibrated using Equations (3) and (4) for the top and bottom figures, respectively. All
simulations assume that adjustment costs are incurred only if a net debt or equity issuance is conducted to move a
firm toward its target capital structure. The solid vertical line represents the optimal SOAOC policy in each subfigure
based on the respective simulations. The dotted vertical lines represent the observed SOAOC speed of 47.9% from
Table 5 [Color figure can be viewed at wileyonlinelibrary.com]

varying values of adjustment costs. In essence, the analysis in Figure 5 puts reasonable bounds around the point
estimates in Figure 4.
In Figure 5, the adjustment costs are assumed to be completely variable and a fixed percentage of the
respective net debt or equity issue implied by the change in the OC(L) that can be attributed to a given SOAOC
policy versus taking no action. Therefore, the costs are strictly increasing for higher SOAOC policies. Again, the
adjustment costs range from 0 to 200 basis points for net debt issues and 0 to 1000 basis points for net equity
issues. In the plots, the color of the lines represents the level of adjustment costs ranging from low costs (orange
lines) to high costs (pink lines). The black dots along each line represent the optimal SOAOC for each curve.
 14756803, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jfir.12300 by Australian National University, Wiley Online Library on [10/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
VALUING THE SPEED OF ADJUSTMENT | 875

F I G U R E 5 Opportunity cost of leverage (OC(L)) versus speed of adjustment (SOA) including adjustment costs.
This figure plots the relation between OC(L) and SOA estimated on a cost basis (SOAOC) for various levels of
adjustment costs. The adjustment costs range from 0 to 200 basis points for net debt issues and 0 to 1000 basis
points for net equity issues. The color of the lines represents the level of the adjustment costs ranging from low
(orange lines) to high costs (pink lines). The black dots along each line represent the optimal SOAOC that minimizes
OC(L) for each curve. The black lines indicate the OC(L) at the observed SOAOC of 47.9%, and the red lines
represent the optimal SOAOC assuming 100 basis points and 500 basis points issuance costs for debt and equity,
respectively [Color figure can be viewed at wileyonlinelibrary.com]

The optimal SOAOC clearly decreases as adjustment costs increase. For example, the optimal SOAOC policy
ranges from SOAOC = 100% when there are no adjustment costs to a minimum of about 21.5% at the highest levels
of adjustment costs (200 basis points for debt and 1000 basis points for equity). This plot also implies that for the
observed SOAOC policy of 47.9% to be the unconditional optimal policy, the net issuance costs would have to be
about 66 basis points for debt and 325 basis points for equity, as indicated by the black lines. This SOAOC policy
would equate to an overall cost of about 2.2% of total assets.
Based on the literature, this level of adjustment costs seems too low, especially for equity issues. Altinkilic and
Hansen (2000) show SEO spreads of about 5.4% and Lee and Masulis (2009) show a mean SEO spread of about
5.1% in their sample. Therefore, Figure 5 also highlights the optimal SOAOC associated with more reasonable
issuance costs of 100 basis points for debt and 500 basis points for equity. In this case, the optimal SOAOC drops to
about 37.5%, which is consistent with the results in Figure 4 based on the Altinkilic and Hansen (2000) models. But
again, the differential effect to firm value is minimal as the OC(L) associated with the optimal SOAOC of 37.5% at
these levels of issuance costs drops by only 0.3% to 2.2% of total assets.
Overall, the main conclusions from the simulation results are as follows. First, firm value is a concave function
of SOAOC. Second, the optimal SOAOC policy is decreasing in the level of adjustment costs. Again, although these
results may be unsurprising, this is the first article to empirically calibrate these curves. Third, the observed SOAOC
policy of 47.9% is higher than the optimal SOAOC policy implied by the preceding simulations at reasonable levels of
adjustment costs. However, the difference in terms of firm value between the observed SOAOC policy and the
optimal SOAOC implied by the simulation study is very small as firms would stand to gain up to only about
0.1%–0.3% of total assets by changing to the simulated optimal SOAOC speeds of 31%–37%. Therefore, firms
appear to be very close to optimizing value, even though they seem to be adjusting too fast. Moreover, once
adjustment costs are taken into consideration, firm value is not very sensitive to SOAOC for a relatively wide range
 14756803, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jfir.12300 by Australian National University, Wiley Online Library on [10/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
876 | JOURNAL OF FINANCIAL RESEARCH

of SOAOCs around the optimum. However, extremely slow or fast SOAOC policies can have substantially negative
impacts on firm value and the curves are all convex.
These are important findings because they suggest that firms are not only actively targeting capital structures,
but they are also choosing SOAOC policies that lock in the bulk of potential savings. Moreover, this is the first article
to calibrate the empirical relation between SOAOC policies and firm value and derive an optimal dynamic capital
structure adjustment policy.

6 | SU MM A R Y A N D CO N CL U S IO N S

I estimate the speed at which firms adjust toward target capital structures on a cost basis. Previous studies that
estimate SOAL produce mixed findings that are impossible to interpret in terms of their relation to firm value. I
extend this literature by calibrating functions that provide a direct link between capital structure decisions and firm
value. This allows for the calibration of the empirical relation between SOAOC policies and firm value, and my
simulation results suggest an optimal SOAOC of between 31% and 37%.
Moreover, the mean observed SOAOC policy of 47.9% is higher than the optimal SOAOC policy implied by the
simulations at reasonable levels of adjustment costs. However, the difference in terms of firm value between the
observed SOAOC policy and the optimal SOAOC implied by the simulation study is very small as firms would stand to
gain up to only about 0.1%–0.3% of total assets by changing to the simulated optimal SOAOC speeds of 31%–37%.
Therefore, firms appear to be very close to optimizing value even though they seem to be adjusting too fast. In
other words, firm value is not very sensitive to SOAOC for a relatively wide range of SOAOCs around the optimum as
the relation between firm value and SOAOC follows an inverted U‐shape. However, extremely slow or fast
SOAOC policies can have substantially negative impacts on firm value because firm value versus SOAOC is convex.

A C KN O W L E D G M E N T S
I thank Alireza Ebrahim, Bill Francis, Iftekhar Hasan, Jonathan Jones, Thomas Shohfi, Akhtar Siddique, Majeed Simaan,
and participants at the American Economic Association and Financial Management Association annual meetings for
helpful comments and suggestions. All errors and views expressed herein are those of the author alone.

ORCID
Brian Clark http://orcid.org/0000-0002-4819-1200

REFERENCES
Altinkilic, O., & Hansen, R. (2000). Are there economies of scale in underwriting fees? Evidence of rising external financing
costs. Review of Financial Studies, 13, 191–218.
Baker, M., & Wurgler, J. (2002). Market timing and capital structure. Journal of Finance, 57, 1–30.
Blouin, J., Core, J., & Guay, W. (2010). Have the tax benefits of debt been overestimated? Journal of Financial Economics, 98,
195–213.
Blundell, R., & Bond, S. (1998). Initial conditions and moment restrictions in dynamic panel data models. Journal of
Econometrics, 87, 115–143.
Chang, X., & Dasgupta, S. (2009). Target behavior and financing: How conclusive is the evidence? Journal of Finance, 64,
1767–1796.
Devos, E., Rahman, S., & Tsang, D. (2017). Debt covenants and the speed of capital structure adjustment. Journal of
Corporate Finance, 45, 1–18.
Elsas, R., & Florysiak, D. (2015). Dynamic capital structure adjustment and the impact of fractional dependent variables.
Journal of Financial and Quantitative Analysis, 50, 1105–1133.
Fama, E., & French, K. (2002). Testing trade‐off and pecking order predictions about dividends and debt. Review of Financial
Studies, 15, 1–33.
Flannery, M., & Hankins, K. (2013). Estimating dynamic panel models in corporate finance. Journal of Corporate Finance, 19,
1–19.
 14756803, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jfir.12300 by Australian National University, Wiley Online Library on [10/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
VALUING THE SPEED OF ADJUSTMENT | 877

Flannery, M., & Rangan, K. (2006). Partial adjustment toward target capital structures. Journal of Financial Economics, 79,
469–506.
Graham, J. (1996a). Debt and the marginal tax rate. Journal of Financial Economics, 41, 41–73.
Graham, J. (1996b). Proxies for the marginal tax rate. Journal of Financial Economics, 42, 187–221.
Graham, J. (2000). How big are the tax benefits of debt? Journal of Finance, 55, 1901–1941.
Graham, J., & Harvey, C. (2001). The theory and practice of corporate finance: Evidence from the field. Journal of Financial
Economics, 60, 187–243.
Graham, J., & Leary, M. (2011). A review of capital structure research and directions for the future. Annual Review of
Financial Economics, 3, 309–345.
Harford, J., Klasa, S., & Walcott, N. (2009). Do firms have leverage targets? Evidence from acquisitions. Journal of Financial
Economics, 93, 1–14.
Heider, F., & Ljungqvist, A. (2015). As certain as debt and taxes: Estimating the tax sensitivity of leverage from state tax
changes. Journal of Financial Economics, 118, 684–712.
Hennessy, C., & Whited, T. (2005). Debt dynamics. Journal of Finance, 60, 1129–1165.
Huang, R., & Ritter, J. (2009). Testing theories of capital structure and estimating the speed of adjustment. Journal of
Financial and Quantitative Analysis, 44, 237–271.
Iliev, P., & Welch, I. (2010). Reconciling estimates of the speed of adjustment of leverage ratios (Working Paper). Available at
SSRN: https://ssrn.com/abstract=1542691 or https://doi.org/10.2139/ssrn.1542691
Korteweg, A. (2010). The net benefits of leverage. Journal of Finance, 66, 2137–2170.
Leary, M., & Roberts, M. (2005). Do firms rebalance their capital structures? Journal of Finance, 60, 2575–2619.
Lee, G., & Masulis, R. (2009). Seasoned equity offerings: Quality of accounting information and expected floatation costs.
Journal of Financial Economics, 92, 443–469.
Lemmon, M., Roberts, M., & Zender, J. (2008). Back to the beginning: Persistence and the cross‐section of corporate capital
structures. Journal of Finance, 63, 1575–1608.
Li, S., Whited, T. M., & Wu, Y. (2016). Collateral, taxes, and leverage. Review of Financial Studies, 29, 1453–1500.
Mackie‐Mason, J. (1990). Do taxes affect corporate financing decisions? Journal of Finance, 44, 1471–1493.
Mukherjee, T., & Wang, W. (2013). Capital structure deviation and the speed of adjustment. Financial Review, 48, 597–615.
Shevlin, T. (1990). Estimating corporate marginal tax rates with asymmetric tax treatment of gains and losses. Journal of the
American Taxation Association, 12, 51–67.
Strebulaev, I. (2007). Do tests of capital structure theory mean what they say? Journal of Finance, 52, 1747–1787.
Strebulaev, I., & Yang, B. (2013). The mystery of zero‐leverage firms. Journal of Financial Economics, 109, 1–23.
Van Binsbergen, J., Graham, J., & Yang, J. (2010). The cost of debt. Journal of Finance, 65, 2089–2136.
Welch, I. (2004). Capital structure and stock returns. Journal of Political Economy, 112, 106–131.
Yin, Q. E., & Ritter, J. R. (2020). The speed of adjustment to the target market value leverage is slower than you think.
Journal of Financial and Quantitative Analysis, 55, 1946–1977.

S UP P O R T I N G I N F O R M A T I O N
Additional supporting information can be found online in the Supporting Information section at the end of this
article.

How to cite this article: Clark, B. (2022). Valuing the speed of adjustment of capital structure. Journal of
Financial Research, 45, 855–879. https://doi.org/10.1111/jfir.12300
 14756803, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jfir.12300 by Australian National University, Wiley Online Library on [10/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
878 | JOURNAL OF FINANCIAL RESEARCH

A P P E N D IX A : V A R I A B L E DE F I N I T I O N S
Variable definitions with Compustat codes in brackets:

Total Interest and Related Expense[XINT]

IOB =
Total Assets[AT]

Current Liabilities[DLC] + Long Term Debt[DLTT]

MKL =
Current Liabilities[DLC]+
Long Term Debt[DLTT] + Market Equity[CSHPRI × PRCC]

Current Liabilities[DLC] + Long Term Debt[DLTT]

BKL =
Total Assets[AT]

Total Inventories[INVT] + Net Property, Plant, and Equipment[PPENT]

COL =
Total Assets[AT]

LTA = log(Total Assets[AT])

Commom Equity [CEQ]

BTM =
Market Equity[CSHPRI × PRCC]

Total Intangible Assets[INTAN]

INTANG =
Total Assets[AT]

Operating Income Before Depreciation[OIBDP]

CF =
Total Assets[AT]

1, Common Dividends[DVC] > 0


DDIV = 


 0,Otherwise

3.3(Pretax Income[PI]) + 1.0(Net Sales[SALE])

+1.4(Retained Earnings[RE]) + 1.2(Working Capital[WCAP])
ZSCORE =
Total Assets[AT]

Long‐Term Debt Issuance[DLTIS]

DISS =
Total Assets[AT]

Long‐Term Debt Reduction[DLTR]

DRED =
Total Assets[AT]

Sale of Common and Preferred Stock[SSTK]

EISS =
Total Assets[AT]

Purchase of Common and Preferred Stock[PRSTKC]

ERED =
Total Assets[AT]
 14756803, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jfir.12300 by Australian National University, Wiley Online Library on [10/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
VALUING THE SPEED OF ADJUSTMENT | 879

A P P E N D IX B : V A R I A B L E DE F I N I T I O N S
Variable definitions with Compustat codes in brackets:

Total Interest and Related Expense[XINT]

IOB =
Total Assets[AT]

Current Liabilities[DLC] + Long Term Debt[DLTT]

MKL =
Current Liabilities[DLC]+
Long Term Debt[DLTT] + Market Equity[CSHPRI*PRCC]

Current Liabilities[DLC] + Long Term Debt[DLTT]

BKL =
Total Assets[AT]

Total Inventories[INVT] + Net Property, Plant, and Equipment[PPENT]

COL =
Total Assets[AT]

LTA = log(Total Assets[AT])

Commom Equity[CEQ]
BTM =
Market Equity[CSHPRI*PRCC]

Total Intangible Assets[INTAN]

INTANG =
Total Assets[AT]

Operating Income Before Depreciation[OIBDP]

CF =
Total Assets[AT]

1, Common Dividends [DVC] > 0


DDIV = 


 0, Otherwise

3.3(Pretax Income[PI]) + 1.0(Net Sales[SALE])

+1.4(Retained Earnings[RE]) + 1.2(Working Capital[WCAP])
ZSCORE =
Total Assets[AT]

Long Term Debt Issuance[DLTIS]

DISS =
Total Assets[AT]

Long Term Debt Reduction[DLTR]

DRED =
Total Assets[AT]

Sale of Common and Preferred Stock[SSTK]

EISS =
Total Assets[AT]

Purchase of Common and Preferred Stock[PRSTKC]

ERED =
Total Assets[AT]