The Cost of Debt - Wharton...

The Cost of Debt∗

Jules H. van Binsbergen

Fuqua School of Business

Duke University†

John R. Graham


Duke University‡

and NBER

Jie Yang


Duke University§

This version: March 2007

AbstractWe use panel data from 1980 to 2005 to estimate cost functions for corporatedebt. We start with Graham’s (2000) debt benefit curves and assume thatfor financially unconstrained firms, the benefit curve intersects the cost curveat the actual level of debt, on average. Using this equilibrium condition,exogenous shifts of the benefit curves enable us to identify the marginal costcurve. We recover marginal cost functions that are positively sloped, and,by integrating to determine the area between the benefit and cost functions,we estimate that the net benefit of debt equals about 4% of asset value. Ourfindings are consistent across industries and years and are robust to accountingfor fixed adjustment costs of debt. We show that both the intercept andthe slope of the marginal cost curve of debt depend on firm characteristicssuch as asset collateral, book-to-market, sales, cash flows, cash holdings,personal taxes, wages, and whether the firm pays dividends. As such, ourframework provides a new parsimonious environment within which we canexamine implications from competing capital structure theories. Our approachalso allows us to make recommendations about firm-specific optimal debt ratiosand estimate the cost of being under or overlevered.

JEL classifications: G30,G32Keywords: Capital Structure, Cost of Debt

∗We thank Ravi Bansal, Jonathan Berk, Michael Brandt, Alon Brav, Hui Chen, Cam Harvey, Han Hong,Ralph Koijen, Mike Lemmon, Rich Mathews, Bertrand Melenberg, Anamarıa Pieschacon, Adriano Rampini,Michael Roberts, David Robinson, Ilya Strebulaev, George Tauchen, Toni Whited, and seminar participantsat Duke University for helpful comments.

†Durham, NC 27708. Phone: (919) 660-7636. Email: [email protected].‡Durham, NC 27708. Phone: (919) 660-7857. Email: [email protected].§Durham, NC 27708. Phone: (919) 660-2939. Email: [email protected].

1 Introduction

Hundreds of papers investigate corporate financial decisions and the factors influencing

capital structure. Much theoretical work characterizes the choice between debt and equity

in a trade-off context: firms choose their optimal debt ratio by balancing the benefits and

costs. Traditionally, tax savings that occur because interest is deductible while equity payout

is not have been modeled as a primary benefit of debt (Kraus and Litzenberger, 1973).

Other benefits of debt include committing managers to operate efficiently (Jensen, 1984) and

engaging lenders to monitor the firm (Jensen and Meckling, 1976). The costs of debt include

the cost of financial distress (Scott, 1976), personal taxes (Miller, 1977), debt overhang

(Myers, 1977), and agency conflicts between managers and investors or among different

groups of investors. For the most part, these theoretical predictions have been tested using

reduced form regressions that include proxy variables to measure the hypothesized effects.

The empirical results are somewhat mixed but a number of empirical regularities have been

documented. Large firms with tangible assets and few growth options tend to use a relatively

large amount of debt (Rajan and Zingales, 2003; Frank and Goyal, 2004). Firms with high

corporate tax rates also tend to have higher debt ratios and use more debt incrementally

(Graham, Lemmon, and Schallheim, 1998).

One reason that tax incentives receive a lot of attention in the literature (and in the

classroom) is because they are relatively easy to quantify and because the gross benefits seem

large (e.g., 35 cents per dollar of interest if the corporate income tax rate is 35 percent).

Tax incentives are sufficiently quantifiable that it is possible to simulate the entire benefit

function associated with interest tax deductibility and integrate under the curve to estimate

the gross tax benefits of debt (Graham, 2000). Therefore, at least in the tax dimension, we

are able to specify the firm-specific benefits of debt.

It has been much more elusive to quantify the costs of debt. Warner (1977) and Miller

(1977) observe that the traditional costs of debt (e.g., direct bankruptcy costs) appear to be

low relative to the tax benefits, implying that other, unobserved, or hard to quantify costs

are important. Several recent papers argue that once these other debt costs are considered,

the marginal costs roughly equal the marginal (tax) benefits of debt (e.g., Berk, Stanton and

Zechner, 2006; Carlson and Lazrak, 2006). Notwithstanding these theoretical arguments,

debt costs are generally hard to measure empirically. To date, the costs of debt have not

been explicitly quantified to the same degree as have the benefits.

In this paper, we explicitly map out the debt cost function. To do this, we need to address

issues related to the standard econometric difficulties associated with identifying demand and

supply curves. In our case, we start with Graham’s simulated marginal tax benefit curves

1

and the observed (equilibrium) debt choices. Observing variation in the marginal benefit

curve over time and in the cross section gives us an important advantage over the standard

demand and supply framework where only equilibrium points are observed. Whereas in the

latter framework one has to use as an instrumental variable that proxies for shifts of the

demand (supply) curve to identify the supply (demand) curve, we have the advantage of

observing the actual shifts of the marginal benefit curve. What remains is to purge from

these shifts potential correlation with cost curve shifts (econometric details are provided in

Section 2) by using a set of control variables that proxy for cost effects. Then, we use the

variation due to pure benefit shifts to identify the cost curve.

As just described, we use marginal benefit variation in both the cross section and in the

time series to determine marginal cost curves. However, our proposed method does critically

depend on the right choice of control variables. As such, omitted variables may induce a bias

in our estimates. To ensure that our results are not driven by such econometric issues we

repeat our analysis using as the sole instrument the exogenous tax regime shifts between 1980

and 2005. We show that this pseudo-natural experiment leads to estimates that are very

similar to those obtained in our main specification. Furthermore, our primary conclusions

hold when estimated from 1998 to 2005. During this period there were no exogenous tax

regime shifts, meaning that for this time period the time series approach is infeasible, and

therefore the identification of the cost curve is mainly based on cross-sectional marginal

benefit variation. Reassuringly, the cross-sectional approach also corroborates our main

results. This indicates that our results are robust to identifying the cost curve based on (i)

cross-sectional variation, (ii) time series variation, or (iii) a combination of the two.

Our analysis produces cost curves that are steeply and positively sloped, as expected.

We estimate different curves for each industry, and the slopes of these curves vary in ways

that make sense economically. We relate both the slope and the intercept of the cost curve

to firm characteristics such as whether the firm’s assets are collateralizable, book-to-market,

sales, cash flows, cash holdings and whether the firm pays dividends. We then compare our

findings to existing theories on the costs and benefits of debt. We also produce easy-to-

implement algorithms to allow researchers and practitioners to explicitly map out the debt

cost function by industry and by firm. This fills a big void in the current state of affairs by

providing both explicit quantification of the cost of debt and specific recommendations to

corporations in terms of optimal capital structure.

Our results are robust to the presence of fixed adjustment costs. Recently it has been

argued (e.g. Fischer, Heinkel and Zechner, 1989; Leary and Roberts, 2005; and Strebulaev,

2006) that fixed adjustment costs prevent firms from responding instantaneously to changing

2

conditions, leading to infrequent capital structure adjustments. With adjustment costs,

capital structure policy can be modeled as following an (s,S) rule, where an adjustment only

occurs if the ‘optimal’ capital structure falls outside a bandwidth with lower bound s and

upper bound S. We repeat our analysis by only including those firm-year observations in

which a substantial rebalancing of capital structure occurs. Our results appear the same on

this sample.

Armed with simulated debt benefit functions and estimated cost functions, we calculate

firm-specific optimal capital structure at the intersection of the curves. We also integrate the

area between the curves to estimate the net benefits of debt financing and the deadweight

costs to deviating from the optimum. We form subsamples based on the level of financial

constraint and financial distress. We assume that, on average, unconstrained and non-

distressed firms make optimal capital structure choices. Among these firms, we estimate that

the net benefit of debt financing equals 2.9-3.9% of book value for firms at the calculated

optimum leverage positions, on average, compared to a gross benefit of 10.1-10.2% of book

value. The net benefit of debt financing equals 0.2-1.7% of firm value for firms at the

observed leverage positions, on average, compared to a gross benefit of 9.0% of firm value.

This implies deadweight losses of about 2.2-2.6% of firm value due to costs of being away

from the optimum.

The rest of the paper proceeds as follows. In section 2, we explain the main intuition of

our instrumental variables approach. In section 3 we give an extensive data description. In

section 4, we report and discuss our results. Section 5 presents case examples of our analysis

for select firms. In section 6 we calculate the benefits and costs of debt and analyze the

costs of being under or overlevered. Section 7 discusses several robustness checks. Section 8

concludes.

2 Methodology of Estimating Marginal Cost Curves

Using the simulation techniques of Graham (2000), we create marginal tax benefit curves

of debt for a large panel of approximately 120,000 firm-years between 1980 and 2005. The

marginal benefit curve measures the marginal tax benefit for each dollar of incremental

interest deduction.1. The shape of these benefit curves varies by firm, but they are weakly

monotonic and typically horizontal for low levels of debt and become negatively sloped for

higher levels of debt (see Figure 1).

We know for each firm in each year the current level of debt. Henceforth, we will refer

1The mechanics of the marginal benefit curve simulations are described in section 3

3

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

x = interest/assets

MB(x)MC(x)

x*

y*

y =

MC

,MB

.

Figure 1: Capital structure equilibrium for a financially unconstrained firm. The figure shows the marginalbenefit curve of debt [MB(x)], the marginal cost curve of debt [MC(x)], and the equilibrium level of debt[x∗] where marginal cost and marginal benefit are equated. The equilibrium marginal benefit level at x∗

(which equals the cost level at x∗) is denoted by y∗.

to this current level of debt as the “equilibrium level of debt,” denoted by x∗i,t. That is,

we implicitly assume that for financially unconstrained firms, on average, the marginal cost

curve of debt (MC) intersects the marginal benefit curve of debt (MB) at the equilibrium

level. We refer to the observed corresponding marginal benefit level as the “equilibrium

benefit of debt,” denoted by y∗i,t. In equilibrium at xi,t = x∗i,t the following equality holds:

y∗i,t ≡ MCi,t

(x∗i,t

)= MBi,t

(x∗i,t

). (1)

The function fi,t, which is simulated, describes for firm i at time t the shape of the

marginal benefit curve of debt:

MBi,t = fi,t (xi,t) , (2)

where xi,t represents the level of debt, expressed as the ratio of interest over book value of

assets. Note that other measures of leverage, like the ratio of debt over the market value

of assets, can also be used. Figure 1 illustrates the equilibrium concept for a financially

unconstrained firm.

To recover the marginal cost function of debt from the equilibrium debt levels and the

equilibrium benefit levels, we need to identify ‘exogenous’ shifts of the marginal benefit

4

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0.5

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42x = interest/assets

y =

MC

,MB

.

MB1(x)

MB2(x)

MB3(x)

MC(x)

Figure 2: Identifying the cost function using shifts in the marginal benefit function. The figure shows threemarginal benefit curves of debt, each intersected by the marginal cost curve of debt. The three curvescan represent the marginal benefit curves of the same firm at three different points in time. The curvescan alternatively represent the marginal benefit curves of three different firms at the same point in time.Empirically, we use both cross-sectional and time-series variation in the marginal benefit curve to identifythe marginal cost function of debt.

curve. In this context, the word exogenous implies that a shift of the marginal benefit curve

is uncorrelated with a shift in the marginal cost curve. In other words, we need to identify

shocks to the marginal benefit curve of debt while holding the marginal cost curve constant.

The exogenous benefit shifts may result from actual time series shifts of the marginal benefit

curve of firm i, for example after a tax regime shift. However, exogenous benefit shifts may

also result from cross-sectional variation in the location of the marginal benefit curve of debt

at some time t. If two otherwise similar firms have different marginal benefit curves, and

hence a different equilibrium level of debt, this provides information about the marginal cost

curve of debt that we exploit to estimate the marginal cost curve as illustrated in Figure 2.

To identify exogenous shocks to the marginal benefit curve, which we use as the

identifying instrumental variable, we can take either of two approaches: bottom-up or top-

down. The bottom-up approach is the direct way of finding an instrument by considering

variables that proxy for, or events that cause, exogenous marginal benefit shifts. One obvious

event is the tax regime changes mentioned earlier. The shifts of the marginal benefit curve

induced by such natural experiments can provide a clean instrument that has no or very

5

little correlation with shifts of the marginal cost curve. The advantage of this approach,

which we further address in the robustness section, is that the instrument is relatively clean.

The disadvantage of this approach, however, is that the exogenous variation of the benefit

curve is limited to the variation of the tax regime changes. In particular, this approach does

not exploit the fact that there is variation of the marginal benefit curve both in the time

series and in the cross section.

The top-down approach exploits the fact that, unlike the standard framework of

identifying demand and supply curves where only equilibrium points are observed, we observe

the entire simulated marginal benefit curve. In other words, apart from measurement errors

(which we assume to be idiosyncratic), we directly observe the cross-sectional and time series

variation (i.e., shifts) in the benefit curve, which we use to identify the cost function. What

then remains is to purge this variation from any cost effects. The result is an exogenous

benefit shifter.

To purge cost effects, we follow two steps. First, we compute for each firm in each year

the total potential tax benefit of debt, Ai,t, which is equal to the area under the marginal

tax benefit curve:

Ai,t =

∞∫

0

fi,t (xi,t) dxi,t. (3)

Empirically, this area provides an accurate description of the location of the marginal benefit

curve. If the marginal benefit curve shifts upward (downward), then the area under the curve

will increase (decrease). Henceforth, we will therefore interpret variation in this area measure

as variation (shifts) of the marginal benefit curve.2

We next purge the benefit measure Ai,t of potential cost effects. To accomplish this, we

include a set of control variables that are theorized to be correlated with the location of

the marginal cost curve of debt: proxies for a firm’s collateralizable assets (PPEi,t), sales

(LSALESi,t), book-to-market (BTMi,t), whether the firm pays dividends (DDIVi,t), the

firm’s cash flow (CFi,t), cash holdings (CASHi,t), and the personal tax penalty (PTPi,t)

that the firm faces. Define C as the set of cost control variables that drive the location of

the MC curve:

C ≡ {PPE, LSALES, BTM, DDIV,CF, CASH,PTP}. (4)

2Using other measures that proxy for shifts of the marginal benefit curve, such as the y-intercept of themarginal benefit curve, partitions of the area measure, or the coefficients of some polynomial approximationof each curve, leads to similar qualitative results to those we present below. For ease of exposition we focuson the area measure.

6

The purpose of our analysis is to estimate the marginal cost curve of debt, which we

assume to be linear in both interest-over-book (IOB), denoted by xi,t, and the cost control

variables, C.3 Under these assumptions, the marginal cost curve of debt is given by

MCi,t = a + bxi,t +∑c∈C

δcci,t + ξi,t. (5)

Define the error function gi,t as

gi,t = y∗i,t − a− bx∗i,t −∑c∈C

δcci,t. (6)

We then estimate the coefficients, in an exactly identified system of equations, using

generalized method of moments (GMM). The moments are obtained by interacting the

error function above with the following instruments: the constant term, the variation of

the marginal benefit curve Ai,t, and each of the control variables.

In equation (5) we assume that the control variables only cause parallel shifts of the

marginal cost curve of debt and that they do not change its slope. If we also allow the slope

of the marginal cost curve to depend on the control variables, the equation for the marginal

cost curve is given by


δcci,t +∑c∈C

θcci,txi,t + ξi,t (7)

and the error function becomes


δcci,t −∑c∈C

θcci,txi,t. (8)

In this setting, since we have to identify the coefficient on xi,t as well as the coefficients on

ci,txi,t for each c ∈ C, we construct additional instruments by taking the product of Ai,t and

each of the control variables. As we argue later in the paper, separately identifying intercept

and slope effects can be challenging. We therefore focus on the case where the intercept is

fixed across firms and only the slope is allowed to vary, conditional on the cost variables, C.

In this case the equation of the marginal cost curve is given by:


θcci,txi,t + ξi,t (9)

3Note that the linearity of the marginal cost of debt implies that the total cost of debt is a quadraticfunction of xi,t and with a positive slope on xi,t in the marginal cost function, the total cost curve is convex.

7

and the error function is given by


θcci,txi,t (10)

We estimate our coefficients by GMM for efficiency and correct standard errors. However,

to better explain the underlying intuition of our estimation, we now present the two-stage

least squares equivalent of equations 5 and 6.

First, we need to purge Ai,t of possible cost effects by running the following regression:

Ai,t = β0 +∑c∈C

βcci,t + εi,t (11)

By construction, the error term εi,t of this regression is orthogonal to the regressors (i.e., the

control variables). To the extent that the regressors span the information set that describes

the location of the marginal cost curve of debt, the error term εi,t can be interpreted as the

exogenous variation of the marginal benefit curve of debt that is not correlated with shifts

of the MC curve. We use this variation as the instrumental variable to identify the marginal

cost curve of debt. It is important to note that this variation of the marginal benefit curve

includes the tax regime shifts described earlier, and also includes other marginal benefit

shifters both in the time series and in the cross-section. The advantage of this approach is

that the exogenous variation (both time series and cross-sectional) in the benefit curve is large

and not limited to tax regime shifts. The disadvantage is that when purging the cost effects,

we may not have controlled for all possible cost variables. As in all model specifications, this

could possibly lead to an omitted variable bias. To ensure that our results are not driven

by such econometric issues, we also perform our analysis using the bottom-up approach

described above, in which we only use tax regime shifts as the instrument.

Armed with the error term εi,t from equation 11, we can then estimate the coefficients

through an instrumental variables (IV) analysis. This error term, which captures pure

benefit shifts, is the main identifying instrument. The first stage of the IV analysis involves

projecting firms’ equilibrium debt levels x∗i,t onto a constant, εi,t, and the control variables

in C:

x∗i,t = β0 + βεεi,t +∑c∈C

βcci,t + ηi,t, (12)

where ηi,t is the error term of the first-stage regression.4 The fitted values of this regression,

denoted by x∗i,t, represents the variation in the equilibrium debt level due to exogenous shifts

4Econometrically, given the presence of the control variables in the first stage, εi,t can be replaced in

8

of the marginal benefit curve, while holding the marginal cost curve fixed. In the second

stage, we regress y∗i,t on a constant, x∗i,t, and the control variables in C to obtain the slope

and the intercept of the marginal cost curve.

y∗i,t = a + bx∗i,t +∑c∈C

δcci,t + ξi,t, (13)

where ξi,t is the error term of the second-stage regression, which is uncorrelated with xi,t

by construction. Including the control variables in both stages of the analysis serves two

purposes. First, as mentioned in the previous paragraph, it allows us to control for shifts

in the location of the marginal cost curve. Second, it allows us to separately examine the

contribution of each control variable to the estimated functional form of the marginal cost

curve.

3 Data and Summary Statistics

3.1 Marginal benefit curves

Our marginal benefit curves are derived as in Graham (2000). Each point on these benefit

functions measures the present value tax benefit of a dollar of interest deduction. To

illustrate, ignore for this paragraph the dynamic features of the tax code such as tax loss

carryforwards and carrybacks and other complexities. The first point on the tax benefit

function measures the tax savings associated with deducting the first dollar of interest.

Additional points on the function measure the tax savings from deducting a second dollar

of interest, a third dollar, etc. Based on the current statutory federal tax schedule, each of

these initial interest deductions would be worth $0.35 for a profitable firm, where 0.35 is the

corporate marginal income tax rate. At some point, all taxable income would be shielded

by interest deductions, and incremental deductions would be worthless. Therefore, ignoring

the complexities of the tax code, a static tax benefit function would be a step function that

has an initial value of 0.35 and eventually drops to 0.0.

The dynamic and complex features of the tax code have a tendency to stretch out and

smooth the benefit function. First, consider dynamic features of the tax code, such as

tax loss carryforwards. At the point at which all current taxable income is shielded by

current interest deductions, an extra dollar of interest leads to a loss today, and this loss is

equation (12) by Ai,t. We have presented equation (11) merely to convey the intuition that εi,t captures theexogenous variation of the marginal benefit curve, as this residual is by construction orthogonal to the costcontrol variables.

9

carried forward to shield profits in future years. For example, for a loss firm that will soon

become profitable, an extra dollar of interest today effectively shields income next year, and

saves the firm $0.35 one year from today. Therefore, the present value tax savings from an

incremental dollar of interest today is worth the present value of $0.35 today, or about $0.33.

Once carryforwards are considered, therefore, rather than stepping straight down to zero at

the point of surplus interest deductions, the benefit function is sloped smoothly downward,

reaching zero more gradually. Other features of the tax code that we consider, including tax

loss carrybacks, the alternative minimum tax, and investment tax credits also smooth the

tax benefit function (see Graham and Smith, 1999 for details).

Second, consider an uncertain world in which the probability of profitability is between

zero and one. Say, for example, that there is a one-half chance that a firm will be profitable

and a one-half chance that it will be unprofitable. In this case, even with a simple, static

tax code, the expected tax benefit is $0.175 for one dollar of interest deduction. Therefore,

we simulate tax benefit functions so that the tax benefit of interest deductions at any given

point is conditional on the probability that the firm will be taxable.

More specifically, we calculate one point on a tax benefit function for one firm in one year

as follows. (Recall that each point on the function represents the expected corporate marginal

tax rate (MTR) for that level of income and deduction.) The first step for a given firm-year

involves calculating the historic mean and variance of the change in taxable income for each

firm. The second step uses this historic information to forecast future income many years

into the future, to allow for full effects of the tax carryforward feature of the tax code (e.g.,

in 2005, tax losses can be carried forward 20 years into the future, so we forecast 20 years

into the future when simulating the 2005 benefit curves). These forecasts are generated with

random draws from a normal distribution, with mean and variance equal to that gathered

in the first step; therefore, many different forecasts of the future can be generated for each

firm. In particular, we produce 50 forecasts of the future for each firm in each year.

The third step calculates the present value tax liability along each of the 50 income paths

generated in the second step, accounting for the tax-loss carryback and carryforward features

of the tax code. The fourth step adds $1 to current year income and recalculates the present

value tax liability along each path. The incremental tax liability calculated in the fourth

step, minus that calculated in the third step, is the present value tax liability from earning

an extra dollar today; in other words, the economic MTR. A separate marginal tax rate is

calculated along each of the forecasted income paths to capture the different tax situations

a firm might experience in different future scenarios. The idea is to mimic the different

planning scenarios that a manager might consider. The fifth step averages across the MTRs

10

from the 50 different scenarios to calculate the expected economic marginal tax rate for a

given firm-year.

These five steps produce the expected marginal tax rate for a single firm-year, for a

given level of interest deduction. To calculate the entire benefit function (for a given firm

in a given year), we replicate the entire process for 17 different levels of interest deductions.

Expressed as a proportion of the actual interest that a firm deducted in a given firm-year,

these 17 levels are 0%, 20%, 40%, 60%, 80%, 100%, 120%, 160%, 200%, 300%, 400%, ...,

1000%. To clarify, 100% represents the actual level of deductions taken, so this point on the

benefit function represents that firm’s actual marginal tax rate in a given year, considering

the present value effects of the dynamic tax code. The marginal tax benefit function is

completed by “connecting the dots” created by the 17 discrete levels of interest deduction.

Also, note that the area under the benefit function up to the 100% point represents the gross

tax benefit of debt for a given firm in a given year, ignoring all costs.

These steps are replicated for each firm for each year, to produce a panel of firm-year tax

benefit functions for each year from 1980 to 2005. The benefit functions in this panel vary

across firms. They can also vary through time for a given firm as the tax code or a firm’s

circumstances change.

3.2 Corporate financial statement data

We obtain corporate financial statement data from Standard & Poor’s COMPUSTAT

database from 1980 to 2005. Merging the tax benefit functions with COMPUSTAT based on

the eight digit firm CUSIP leaves 119,199 firm-year observations.5 For each firm, we create

empirical measures of the control variables described in the previous section, which includes

plant, property and equipment over total book assets (PPE), log of net sales (LSALES), book

equity to market equity (BTM), indicator for a dividend paying firm (DDIV), cash flow over

total book assets (CF), cash holdings over total book assets (CASH), and personal tax

penalty (PTP) as calculated in Graham (1999). We measure financial distress by Altman’s

(1968) Z-score (ZSCORE). Firms are defined to be non-distressed if they have Z-scores

above the median. We measure financial constraint according to four definitions offered in

the literature: (i) firms with no or limited leverage adjustments, (ii) the Kaplan and Zingales

(1997) index (KZ), (iii) the Cleary (1999) index (CL), and (iv) the Whited and Wu (2005)

index (WW). We discuss the four measures in the next subsection. Appendix A provides a

5To avoid issues involving changes in firm CUSIP across time, we track firms across time usingCOMPUSTAT’s GVKEY variable, which is created for this purpose. However, merging by firm CUSIPwithin each year will not be affected by this issue.

11

detailed description on the construction of each variable.

We normalize equilibrium interest expense by total book assets, which hereafter we refer

to as interest-over-book (IOB). Note that PPE, CF, and CASH are normalized by total book

assets. For the construction of LSALES, we chain net sales revenue to 2000 dollars to adjust

for inflation before taking logarithms. We further remove all firms with non-positive book

values, common equity, capital, and sales, or negative dividends. Such firms have either

unreliable COMPUSTAT data or are typically distressed or severely unprofitable and are

likely to be constrained with respect to accessing financial markets. Within our framework

they can therefore be considered as potentially being “out of equilibrium.” Further,

we remove firms that were involved in substantial M&A activity, defined as acquisitions

amounting to over 15 percent of firm total assets. Finally, we remove outliers defined as

firm-year observations that are in the first and 99th percentile tails for (i) the area under the

marginal benefits curve (A), (ii) the normalized observed interest-over-book (IOB) expenses

(x∗), (iii) the book to market ratio (BTM), (iv) the financial distress measure (ZSCORE),

and (v) the four financial constraint measures (long term debt issuance or reduction, KZ,

CL, and WW). 6 We also windorsize the personal tax penalty at the first and 99th percentile

levels. This results in a sample of 102,896 firm-years. Table 1 provides an overview of

the sample construction. Table 2 provides the summary statistics and Table 3 shows the

correlations between the variables in each of the samples.

3.3 Financial constraints measures

One important assumption underlying our framework is that firms are able and willing to

adjust capital structure. If firms are not able to adjust their capital structures because they

are financially constrained or financially distressed (i.e., out of equilibrium) they should

be excluded from our estimation procedure. Moreover, recent research on dynamic capital

structure highlights that firms may not continuously adjust their leverage ratios due to non-

negligible adjustment costs (Leary and Roberts, 2005; Kurshev and Strebulaev, 2006; etc.).

Therefore, when a firm does not adjust capital structure it is not clear whether the firm is

acting optimally, is acting passively due to adjustment costs, or is acting suboptimally due

to being constrained in financial markets. Conversely, if a company makes a (substantial)

capital structure adjustment, we can deduce that transactions costs were not sufficient to

prevent capital structure adjustments.

6Removing the outliers of the other control variables (PPE, LSALES, DDIV, CF and CASH) does notchange the distribution of the sample much.

12

To address these issues, we also perform our analysis for those firm-year observations in

which a substantial capital structure adjustment takes place, i.e., those firm-year observations

in which there is substantial long term debt and/or an equity issuance or repurchase

(LTDEIR). In our sample the median level of long term debt issuance/reduction among all

firm-year observations is, respectively, 6.8/3.3 percent of book value. These numbers increase

to respectively 16.0/9.1 percent when we consider debt issuances/reductions above the 75th

percentile. For equity issuances/repurchases, the median is 0.7/0.9 percent, respectively

and 3.3/3.1 percent respectively at the 75th percentile. We define a firm to be financially

unconstrained if it has long term debt or equity adjustment (LTDEIR) above the median

issuance or reduction. Even when we tighten the definition by including only firms above

the 75th percentile, our main results do not change.

3.4 Data description of the subsamples

To explore industry effects, we estimate the marginal cost curve by industry. Additionally,

we perform our empirical analysis for four subsets of our primary sample. These subsets are

based on two criteria. The first criterion is based on the firms’ degree of financial distress,

which we measure by ZSCORE. The second criterion is based on the firms’ long term debt

and equity issuance and repurchases (LTDEIR), as explained in the previous section. The

amount of LTDEIR reflects the firm’s ability to adjust its capital structure and can be

interpreted as a measure of financial constraint. The four samples are then given by:

A : All firms

B : Financially non-distressed firms: ZSCORE above median

C : Financially unconstrained firms: LTDEIR above median

D : Financially unconstrained and non-distressed firms as defined in samples B and C

The analysis of unconstrained firms helps to highlight those firms that can “freely”

optimize their capital structure. Because we assume in our estimation procedure that, on

average, the cost curve of debt intersects the marginal benefit curve of debt at the actual level

of debt, our estimation is most accurate for those firms that have the flexibility to actively

optimize over the amount of debt. If the firm is constrained, it may be (temporarily) “out

of equilibrium.” Figure 3 plots the histograms of the area under the marginal benefits curve

for each subsample. The majority of the firms have areas under the marginal benefit curve

that are less than 0.5 percent of book value in a single year, which corresponds to around 5.1

percent of the firm’s book value in perpetuity. The low benefit for these firms arises either

13

0.0

5.1

.15

.2.2

5P

erce

nt

0 .01 .02 .03 .04 .05 .06 .07 .08 .09 .1 .11 .12Area Under Curve

Sample A

0.0

5.1

.15

.2.2

5P

erce

nt

0 .01 .02 .03 .04 .05 .06 .07 .08 .09 .1 .11 .12Area Under Curve

Sample B

0.0

5.1

.15

.2.2

5P

erce

nt

0 .01 .02 .03 .04 .05 .06 .07 .08 .09 .1 .11 .12Area Under Curve

Sample C

0.0

5.1

.15

.2.2

5P

erce

nt

0 .01 .02 .03 .04 .05 .06 .07 .08 .09 .1 .11 .12Area Under Curve

Sample D

Figure 3: Histogram of the area under the curve. This figure shows the relative frequency distribution ofthe area under the marginal benefits curve for samples: i) A, all firms, ii) B, non-distressed firms, iii) C,financially unconstrained firms, iv) D, non-distressed and unconstrained firms. Small bins indicate firms withdeclining marginal benefits or constantly low marginal benefits. Large bins indicate firms with consistentlyhigh marginal benefits.

from benefit functions that are relatively flat but at a low benefit level, or that are steeply

downward sloping. Firms in the 0.5 and above bin are typically firms that have constant

marginal tax rates at the highest tax rate. These are firms we would expect to load heavily

on debt in equilibrium unless their marginal cost curve is especially steep. As we move from

Sample A to Sample D, the distribution of the area under the marginal benefits curve shifts

to the right. This indicates that the firms in Sample D tend to be more profitable and thus

have higher total potential benefits to debt through interest deductions. Table 2 compares

the summary statistics of the four samples.

14

4 Results

4.1 Marginal Cost Curves with Fixed Intercept

We estimate the coefficients in equation 9 for the four subsamples A-D, which are reported

in the first four columns of Table 4. These estimates allow the slope of the marginal cost

curve to depend on the control variables, but assume that the intercept is the same for all

firms. We allow for intercept variation in the next section. The slope of the marginal cost

curve determines the convexity of the total cost curve and measures the increase in marginal

cost that results from a one-unit increase in the interest-over-book ratio. When the intercept

of the marginal cost curve is fixed, a larger slope on the marginal cost curve implies higher

marginal cost for all values of leverage.

Within our framework, the capital structure decision follows from a tradeoff between the

tax benefits of debt and the costs of debt. It is important to stress that, in our framework,

the marginal benefit curve only measures the tax benefits of debt. As a consequence, the

other benefits of debt, such as committing managers to operate efficiently (Jensen, 1984)

and engaging lenders to monitor the firm (Jensen and Meckling, 1976), are included in

our framework as negative costs, and therefore are reflected in our estimated marginal cost

curves. Our cost curves also include the traditional costs of debt, such as the cost of financial

distress (Scott, 1976), personal taxes (Miller, 1977), debt overhang (Myers, 1977), agency

conflicts between managers and investors or between different groups of investors, and any

other cost that firms consider in their optimal debt choice.

We interpret the cost coefficients embedded in the cost of debt functions, and compare

the implications from these coefficients to the capital structure regularities documented

elsewhere. For example, the literature has documented that large firms with tangible assets

and few growth options tend to use relatively large amounts of debt (Frank and Goyal, 2004).

As we will show, the effects of individual cost variables on the cost of debt function are

consistent with debt usage implications in the existing capital structure literature, which we

find very reassuring. There are a great many unanswered questions in the capital structure

literature in terms of interpreting individual coefficients, and by no means do we believe

that our procedure helps to solve every puzzle. Rather, our procedure quantifies just how

large the influence of individual variables must be on the cost of debt to explain capital

structure regularities. Table i summarizes the effect of the control variables on the cost of

debt function, and summarizes the standard capital structure result (as presented in Frank

and Goyal (1994) and elsewhere).

We begin by interpreting the effect of collateralizable assets on the cost of debt function.

15

Control Variable Cost of Debt LeveragePPE – +LSALES + +/-BTM – +DDIV + –CF + –CASH + –PTP + –

Table i: The influence of each of the control variables on (i) the cost of debt following from Table 4, and (ii)the leverage of the firm, as documented by the literature.

The -0.470 coefficient in Table 4, sample D shows that the cost of debt decreases in the

proportion of a firm’s assets made up by property, plant, and equipment (PPE, normalized

by the book value of assets). This is consistent with the common capital structure finding

that high PPE leads to increased use of debt (see Table i). Note that all control variables

are standardized (i.e., have mean zero and standard deviation of one) so that the results

have a one standard deviation interpretation. In terms of economic magnitude, the -0.470

coefficient indicates that a one standard deviation change in PPE would lead to a 0.470

reduction in the slope of the marginal cost function of debt.

The 0.405 coefficient on LSALES indicates that, on average, large firms face a higher

cost of debt. This result is somewhat surprising but is consistent with some recent literature

that indicates that large firms use less debt (Faulkender and Petersen, 2004; Kurshev and

Strebulaev, 2006). Other research (as summarized in Frank and Goyal, 2004) documents a

positive relation between size and debt usage. In Table 5 we show that the benefits of debt

also increase with firm size. Therefore, the net effect of size on the use of debt depends on

whether the cost or benefit effect dominates.7 The differing firm size implications documented

in various capital structure papers imply that the influences of size on the costs versus benefits

of debt dominate in different settings and samples.

Firms with growth opportunities (i.e., a low book-to-market (BTM)) on average face a

higher cost of debt (coefficient of -0.518). This is consistent with the common finding that

for growth firms the opportunity cost of debt is high because debt can restrict a firm’s ability

to exercise future growth opportunities due to debt overhang (Myers, 1977). The inflexibility

7For recent work on the relation between size and capital structure, see Kurshev and Strebulaev (2006).They argue that fixed costs of external financing lead to infrequent restructuring and create a wedge betweensmall and large firms. Small firms choose higher leverage at the moment of refinancing to compensate forless frequent rebalancing.

16

arising from debt covenants could also restrict a firm’s ability to optimally invest and exercise

growth options.

Dividend paying firms (DDIV) face a higher cost of debt, as indicated by the 1.066

coefficient in Sample D on Table 4. All else equal, committing to pay dividends reduces the

availability of cash flows to service debt, given the extreme stickiness of dividend payments

(e.g., Brav et al., 2005), thereby increasing debt costs. This result and interpretation

are consistent with the negative relation between dividend-paying status and debt ratios

documented in Frank and Goyal (2004) and elsewhere (see Table i).8

Large cash holdings (CASH) and cash flows (CF) indicate a higher cost of debt according

to the estimates of 1.837 and 1.504, respectively, in Sample D of Table 4. Interpreting these

results is somewhat ambiguous. It may just be that firms that face higher costs of debt, for

whatever reason, find it optimal to hold more cash, which shows up in the cost function.

Or that profitable (high CF) firms use less debt on average, which is consistent with the

implication of higher costs in our estimated cost of debt function.

Finally, using Graham’s (1999) method to quantify the personal tax penalty, we include

this measure in our analysis as one of the cost control variables.9 The 0.915 coefficient

indicates that the cost of debt increases as personal taxes on interest income increase. This

result is consistent with high personal taxes on interest income leading to less debt usage,

as documented by Graham (1999).

As summarized in Table i, in every instance the relation between the control variables

and the cost of debt function are consistent with the relations between these same variables

and debt ratios, as documented in the extant literature. This is reassuring and implies that

our methodology, though technical, is not producing spurious relations in the cost of debt

function. Moreover, our approach allows us to explicitly quantify the effect of each variable

on the implicit cost of debt function that we estimate. At the same time, our approach does

not resolve certain existing capital structure puzzles or debates, such as why profitable firms

use less debt.

It is important to note that we measure the cost of debt as perceived by the firm and

therefore reflected in its chosen debt policy. This cost of debt could, but does not have to,

coincide with the cost of debt perceived by debt markets.

8There is also a pecking-order interpretation of the positive relation between dividend-paying status andthe cost of debt. For companies that are on the margin between issuing either debt or equity (like Sample D,by construction), those that pay dividends may have lower information asymmetry costs and therefore userelatively more equity (and less debt). Therefore, the positive coefficient on dividends might indicate that,relative to equity, debt is less attractive in dividend-paying firms.

9Graham (1999) shows that when measuring the tax benefits of debt, a specification that adjusts forpersonal tax penalty statistically dominates specifications that do not.

17

4.2 Marginal Costs with Fixed Slope

In the previous section we assumed that only the slope of the marginal cost curve depends

on the control variables, but that these variables do not influence the intercept. As an

alternative experiment we assume that the slope of the marginal cost curve is fixed across

firms but that the intercept of the marginal cost curve depends on the control variables. The

results are presented in the last four columns of Table 4. We find that the qualitative results

are identical to the ones described in the previous section. That is, small, non-dividend

paying firms with high collateralizable assets, few growth options, and low cash face a lower

cost of debt.

4.3 Marginal Costs with Varying Slopes and Intercepts

In the previous two sections we assumed that either the intercept or the slope of the cost curve

is the same for all firms. In this section we explore the possibility that both the slope and

the intercept vary with the control variables. In equation 7 we include the control variables

and the interaction terms between the interest-over-book (IOB) and the control variables.

Econometrically this is challenging because the information contained in our instruments may

not be sufficient to identify the slope and intercept effects separately. Indeed our estimation

results indicate that when we include the interaction terms, the statistical significance of our

results decreases, specifically when we move from samples A through D with progressively

smaller sample sizes.

In addition, in this case, slope and intercept effects may offset each other. Only when

the coefficient signs of a certain control variable is the same for both the slope (interaction

term) and the intercept is the resulting effect of that control variable on the marginal cost

curve monotone. When the slope and intercept coefficients are of opposing signs, the effect is

ambiguous, leading to a higher cost curve in one region and potentially a lower cost curve in

another (e.g., higher intercept but flatter slope). This is not surprising and can be understood

in the following way. When we fix the slope across all firms and only allow the intercept

to vary with the control variables, as in section 4.2, changes in the control variables lead

to parallel shifts of the cost curve. When we fix the intercept, as in section 4.1, variation

in the control variables implies pivoting the marginal cost curve around the intercept point.

However if we let both the intercept and the slope vary, the procedure can pivot around

some center point of the curve.

18

4.4 Marginal Costs By Industry

The estimation results of the coefficients in equation 9 by industry are given in Table 6. These

estimates allow the slope of the marginal cost curve to vary with the control variables while

keeping the intercept fixed within an industry. We estimate the coefficients in equation 9

for (i) all firms, (ii) all firms excluding utilities, finance, and public administration10 (iii)

by industry as categorized by the two-digit Standard Industry Classification (SIC). The

classification of the industries is documented in Appendix B. We find steep, positive, and

highly significant slopes for the full sample as well as in each industry. The slope is

particularly steep for the utilities industry and flat for the finance, insurance and real estate

industry. At the same time, the intercept is particularly low for utilities and high for finance,

insurance, and real estate.

4.5 Interpreting Recent Capital Structure Theories

In this section we investigate recent research that proposes effects that may influence the cost

of debt function. In each case, these theories lead to control variables that greatly reduce

sample size, so they are not included in the main analysis discussed earlier.

4.5.1 Wages

Recent work by Berk, Stanton and Zechner (2006) suggests that ex post the true costs of

bankruptcy are born by the employees of the firm. As a consequence, the firm needs to

compensate employees ex ante if it makes riskier capital structure decisions. If a firm wants

to take on additional debt, part of the cost of debt will therefore be the higher wages that

employees require as compensation for bankruptcy risk. The authors argue that employees

self-select, implying that risk-averse employees will find employment with firms that take on

low leverage. As a consequence, changing leverage is more costly for firms with risk averse

employees. The theory suggests that we should include wages per employee as a control

variable in our analysis. We use log of labor expenses divided by number of employees

(LWAGE) as our COMPUSTAT proxy for wages.

Table 7 shows the estimation results when log wages per employee (LWAGE) is included

as a control variable. We do find that high wages are correlated with high costs of debt,

consistent with Berk, Stanton, and Zechner (2006). This result is significant and leaves the

results on other control variables largely unaffected. Note that our sample size drops from

25,977 to 2,211 in Sample D due to a lack of wage data in the COMPUSTAT database.

10Excluding these industries is common practice in the capital structure literature.

19

4.5.2 Macroeconomic Influences

Chen (2006) and Almeida and Philippon (2006) propose that bankruptcies are concentrated

in bad times, i.e., periods when consumers’ marginal utilities are high. This leads investors

to demand higher credit risk premia during bad times due to higher default rates and higher

default losses. This naturally suggests that credit spreads should play a significant role in

the time variation of the cost of debt from the viewpoint of financial markets. As noted

before, the cost of debt as perceived by the firm could, but does not have to, coincide with

the costs perceived by the markets.

Table 8 gives the results for the analysis when we include the Moody’s Baa-Aaa spread

as a control variable. When the spread is high, indicative of bad times with high volatility,

we expect the cost of debt to be high. Thus, we expect a positive sign on our credit spread

variable. We see that this is indeed true for the financially non-distressed samples (B and

D). However, we find the opposite for samples A and C. Financially distressed firms may be

forced to take on additional debt to continue operations in distressed times. As such, these

firms may be “out of equilibrium” and less suitable for our analysis. Therefore we rely more

heavily on the results in samples B and D.

The examples presented in the last two sections illustrate that our framework can be

used to analyze implications from various capital structure theories. After the tax benefits

of debt have been modeled, the control variables in the cost curve can be used to assess the

importance of other factors that affect capital structure decisions.

5 Applications Using Marginal Cost of Debt Functions

Using the fixed intercept results from the left side of Table 4, Sample D, the marginal cost

function for any particular firm can be determined by:

MCi,t = 0.190 + 4.192IOBi,t − 0.470PPEi,t ∗ IOBi,t + 0.405LSALESi,t ∗ IOBi,t

− 0.518BTMi,t ∗ IOBi,t + 1.066DDIVi,t ∗ IOBi,t + 1.837CFi,t ∗ IOBi,t

+ 1.504CASHi,t ∗ IOBi,t + 0.915PTPi,t ∗ IOBi,t

(14)

and, when using the fixed slope results from the right side of Table 4, Sample D, given by:

MCi,t = 0.120 + 5.086IOBi,t − 0.019PPEi,t + 0.015LSALESi,t − 0.029BTMi,t

+ 0.047DDIVi,t + 0.112CFi,t + 0.044CASHi,t + 0.035PTPi,t

(15)

20

where each of the control variables are standardized based on Sample A to have a mean

of zero and a standard deviation of one. The mean and standard deviation for each of the

control variables for Sample A is reported in Table 2.11

5.1 Case Studies of Optimal Debt Usage

Once the cost and benefit functions have been estimated, they can be used to draw inference

on optimal capital structure. We illustrate with three specific cases. The following three firms

are chosen based on name recognition and span of industries: i) Bed, Bath, and Beyond,

in the retail industry, ii) Columbia Broadcasting System (CBS), in the communications

industry, and iii) Morgan Stanley, in the financial services industry. Figure 4 displays the

marginal benefit and marginal cost curves for the three firms, respectively. We estimate the

marginal cost curve using the estimated coefficients from the analysis run on Sample D, as

reported in Table 4, where we assume that the intercept is fixed across firms and the slope

depends on the control variables (see equation (15).12 For these three cases we see that each

of these firms are indeed close to its ‘equilibrium’ because the marginal cost curve and the

marginal benefit curves intersect close to the actual amount of debt the firm uses.

Table ii compares the decile rankings of specific financial ratios for the three firms. All

three firms are large with relatively high total book assets and market equity. However, the

three firms differ substantially in terms of decile ranking for the other control variables.

In 2005, Bed, Bath, and Beyond does not pay dividends, has moderate asset

collaterizability, and relatively high growth opportunities, cash holdings, and sales. The

firm seems to minimize its payment obligations. Given the firm’s high growth opportunities,

cash holdings, sales, and income, the firms capital structure decision is consistent with our

cost/benefit analysis.

We examine CBS, Inc. just before it disappears from the Compustat database in 1994,

after merging with Westinghouse Electronic Corporation. In 1994, CBS has moderate

collateral, cash holdings, and debt to equity ratio, and high growth opportunities, dividend

payments, sales, and cash flow. The growth opportunities, dividend payments and cash

flow indicate a high marginal cost of debt. Our analysis suggests that CBS is about 20%

overlevered and should use less debt as defined by interest payment obligations. However,

their recommended or ‘optimal’ debt is still within 20% of the observed debt level.

11The control variables are not re-standardized by sample to prevent selection bias when analyzing acrosssamples.

12The qualitative results are similar for the case in which we fix the slope and let the intercept vary withthe control variables

21

0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.10

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0.5Bed, Bath, and Beyond, 2005

Interest−Over−Book−Value

MB

/ M

C R

ates

MB CurveMC CurveActual

0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.10

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0.5CBS, 1994

Interest−Over−Book−ValueM

B /

MC

Rat

es


0 0.01 0.02 0.03 0.04 0.05 0.060

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0.5Morgan Stanley, 2004

Interest−Over−Book−Value

MB

/ M

C R

ates


Figure 4: Marginal benefit and marginal cost curves for i) Bed, Bath, and Beyond in 2005, ii) CBS in 1994,and iii) Morgan Stanley in 2004. The vertical line reflects the actual debt in terms of interest of book valueobserved empirically.

Morgan Stanley in 2004 had low capital, dividend obligations, sales, and cash flow,

moderate growth opportunities, and high cash holdings and debt to equity ratio. The

fact that Morgan Stanley is a financial firm complicates what really should be classified

as debt and equity. Nonetheless, using our marginal cost curve derived from the sample D

population, we use the coefficients to calculate the marginal cost function for more general

firms. Morgan Stanley seems to be close to its optimal ratio of interest payments over book

value.

These cases illustrate that our analysis can be used not only to interpret capital structure

theories, but our analysis provides a benchmark against which companies can compare their

capital structure decisions. Our analysis identifies the intersection of each firm’s marginal

benefit and marginal cost curves, where the costs are backed out of debt choices made by

22

Table ii: Key financial characteristics of the three firms studied in our case analysis. TA is the total assetsexpressed in thousands of 2000 dollars, MCAP represents the market capitalization expressed in thousands of2000 dollars, BTM is the book equity to market equity ratio, PPE stands for plants, property, and equipmentover total book assets, DIVIDENDS is the total dividend payout in thousands of dollars over total bookassets, SALES is firm sales over total book assets, CASH is cash holdings over total book assets, CF isincome over total book value and D/E is the debt to equity ratio.

BBB, 2005 CBS, 1994 Morgan Stanley, 2004Ranking Value Ranking Value Ranking Value

TA ($thousand) 8 3000 9 2393 10 708624MCAP ($thousand) 9 9396 10 4420 10 50095PPE 6 0.218 5 0.241 1 0.008BTM 2 0.214 1 0.069 6 0.515DIVIDENDS 1 0 8 0.013 6 0.001SALES 9 1.718 9 1.718 1 0.051CASH 8 0.193 5 0.047 8 0.248CF 10 0.202 9 0.162 2 0.007D/E 1 0 5 0.183 10 8.940

other, similar firms.

6 Quantifying the Cost and Benefit of Debt

As just illustrated, using both the marginal benefit and marginal cost functions, we can

compute the ‘optimal’ or ‘equilibrium’ capital structure for a firm as the intersection between

the two curves. This allows us to infer whether a firm is over-, under-, or optimally leveraged.

We define over- and under-leverage as follows.

First, for each firm, we normalize the current interest-over-book value (IOB) to one;

this has the interpretation that a firm is at 100% of its observed IOB. The ‘optimal’ or

‘equilibrium’ level of debt is calculated as the intersection between the marginal benefit and

marginal cost curves. This ‘equilibrium’ level is then normalized by the observed level, i.e.,

‘equilibrium’ capital structure is expressed as a proportion of the observed. For example, a

firm that has an equilibrium factor of 0.8 should, according to our model, optimally have

80% of the observed IOB. Suppose the normalized equilibrium level of debt equals e, then

the percentage of underleverage/overleverage is given by:

Underleverage = 1− 1/e when e > 1 (16)

Overleverage = 1/e− 1 when e < 1 (17)

23

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

x = interest/assets

MB(x)MC(x)

x*

y*

xo

DWo = Cost of being overleveraged

y =

MC

,MB

.

Figure 5: The costs of being overleveraged. The figure shows the marginal benefit curve of debt [MB(x)],the marginal cost curve of debt [MC(x)], and the equilibrium level of debt [x∗] where marginal cost andmarginal benefit are equated. The marginal benefit level at x∗ (which equals the cost level at x∗) is denotedby y∗. The actual level of debt, denoted by [xo], exceeds the equilibrium level of debt [x∗]. The shaded areaindicates the deadweight loss of being over-leveraged.

Using our example of a firm with equilibrium factor of 0.8, this implies this firm is

overleveraged by 25%. We can also say this firm is 25% “out of equilibrium.” Equipped

with these definitions, we can use our benefit and cost functions to quantify the gross and

net benefits of debt and the costs of being “out of equilibrium.” Note that we use the

statement in a relative sense: the observed debt usage “relative to the optimum implied by

the coefficients of our empirical model.”

6.1 Gross and net benefits of debt

The observed (equilibrium) gross benefits of debt, GBDo (GBDe), is the area under the

marginal benefits curve up to the observed (equilibrium) level of interest over book value

(IOB). The observed (equilibrium) cost of debt, CDo (CDe), is the area under the marginal

cost curve up to the observed (equilibrium) level of IOB. The observed (equilibrium) net

benefits of debt, NBDo (NBDe), is the difference between the gross benefit of debt and the

cost of debt.

Tables 9 and 10 report the unconditional and conditional summary statistics for the gross

benefit of debt, cost of debt, and net benefit of debt. Values are reported for firms both at

the observed level of interest-over-book values (IOB) as well as at the equilibrium implied

24

by the results in Table 4. We see that although the gross benefits of debt are similar for

observed debt levels and equilibrium debt levels, the costs to debt are lower at equilibrium

levels. This results in a higher net benefit of debt for firms at the equilibrium implied by

our analysis, relative to their observed levels. On average, the net benefits of debt at the

implied equilibrium is 3.8% of firm value, and only 0.2% of firm value at observed debt

levels. This suggests that our analysis indeed is able to help firms choose a more optimal

capital structure. Th This result is consistent for both over- and under-leveraged firms. The

difference between the equilibrium and observed net benefits increases as firms move further

out of equilibrium and is on average 5.23% of firm value for firms that are further than 80%

or more from the equilibrium.

6.2 The cost of being underlevered or overlevered

Our analysis allows us to answer the question: how costly is it for firms to be out of

equilibrium? The cost of being ‘overlevered’ can provide insight on the potential cost of

financial distress, while the cost of being ‘underlevered’ can shed light on the cost of financial

constraints. Figure 5 depicts the cost of being overleveraged. The cost of being overleveraged,

DWo, is the loss due to additional costs from having interest over book value (IOB) above

the equilibrium. This is because overleveraged firms pay higher costs relative to the benefits

received. On the other hand, the cost of being underleveraged, DWu, is the deadweight loss

from lower benefits due to having IOB below the equilibrium; the deadweight loss comes

from leaving money on the table in the form of unused benefits relative to cost of debt. The

cost of being out of equilibrium, DWt, combines DWo and DWu into one measure.

Tables 9 and 10 also report the costs of being “out of equilibrium” as value of the

deadweight losses mentioned above. Table 9 show that on average the cost of being out of

equilibrium is 2.2-2.6% of firm value. However, we see that the costs for overleveraged firms

(on average, 3.3-4.0% of firm value) are much higher than those for underleveraged firms (on

average, 1.2-1.3% of firm value). This suggests that it is much more costly for firms to be

overleveraged than underleveraged. This gives insight as to why firms might be conservative

in their debt usage (Graham, 2000). If a firm is incorrect in the analysis of its optimal capital

structure, it is less costly to err on the side of having too little debt than too much debt.

Table 10, panel A show that, sensibly, costs of being out of equilibrium increases as firms

are further away from the equilibrium. We see that a vast majority of overleveraged firm

are more than 80% away from the equilibrium. This suggests that deadweight costs of being

out of equilibrium has a heavy impact.

25

7 Robustness checks

7.1 Different financial constraint measures

As discussed previously, our estimation procedure depends on the assumption that, on

average, firms optimize over their capital structure decisions. In this section we explore

how excluding firms based on a variety of financial constraint or financial distress measures

affects our results. Previously, we used long-term debt issuance or reduction as a measure

of financial constraint. As an additional robustness check, we also identify unconstrained

firms based on (i) the financial constraints measure derived by Kaplan and Zingales (1997)

as estimated by Lamont, Polk, and Saa-Requejo (2001), hereafter referred to as KZ, (ii)

the Cleary (1999) index, hereafter called CL, and (iii) the Whited and Wu (2005) index,

hereafter called WW.

Kaplan and Zingales (1997) categorize the 49 low dividend paying firms of Fazzari,

Hubbard, and Petersen (1988) into five degrees of financial constraint and estimate an

ordered logit model to obtain the probability of a firm falling into any one of the five financial

constraint categories, with financial slack being the lowest state and financial constraint the

highest state. Lamont, Polk, and Saa-Requejo (2001) report the coefficients for the KZ index

that uses only variables available on COMPUSTAT. We use these coefficients to construct our

KZ index. Following the approach of Kaplan and Zingales (1997), Cleary (1999) calculates a

more general financial constraint measure by grouping firms into categories based on whether

they increase or decrease dividend payments. Using this classification procedure, Cleary

(1999) performs discriminant analysis to obtain a measure for financial constraint. We

reproduce this procedure over Cleary’s (1999) sample period of 1987 to 1994 to obtain the

coefficients for our CL index. Finally, in a recent paper, Whited and Wu (2005) argue

that the KZ measure is inconsistent with the intuitive behavior of financially constrained

firms. They derive an alternative measure of financial constraint by formulating the dynamic

optimization problem of a firm that faces the constraint that the distributions of the firm

(e.g., dividends) need to exceed a certain lower bound. They parameterize the Lagrange

multiplier on this constraint and estimate its coefficients with GMM. Effectively, the WW

index indicates that a firm is financially constrained if its sales growth is considerably lower

than its industry’s sales growth. In other words, a highly constrained firm is a slow growing

firm in a fast growing industry. A fully unconstrained firm is a fast growing firm in a slow

growing industry.

26

The formula for the KZ index is given by

KZi,t = −1.002CFi,t + 3.139LTDi,t − 39.367TDIVi,t − 1.314CASHi,t + 0.282Qi,t, (18)

the CL index is given by

CLi,t = 0.176CURi,t−1 − 0.0003FCi,t−1 + 0.008SLi,t−1 − 2.802NI%i,t−1

+0.018SGi,t−1 + 4.372DEBTi,t−1,(19)

and the WW index is given by

WWi,t = −0.091CFi,t − 0.062DDIVi,t + 0.021LTDi,t − 0.044LTAi,t

+0.102ISGi,t − 0.035SGi,t,(20)

where CF is cash flows, LTD is long-term debt, TDIV is total dividends over assets, Q is

Tobin’s Q, DDIV is a dummy variable that equals 1 when a firm pays dividends and 0

otherwise, ISG is industry sales growth, SG is the firm’s sales growth, CUR is the firm’s

current ratio, FC is the fixed charge coverage, NI% is the firms net income margin, DEBT

is the firm’s debt ratio, and SL is the ratio of slack over net fixed assets.

In this section we explore these other financial constraint measures. In particular, in

addition to the four samples described above, we also perform our analysis using the following

samples:

E : KZ below median and ZSCORE above median

F : CL below median and ZSCORE above median

G : WW below median and ZSCORE above median

H : LTDEIR above 3rd quartile and ZSCORE above median

I : KZ below 1st quartile and ZSCORE above median

J : CL below first quartile and ZSCORE above median

K : WW below first quartile and ZSCORE above median

Table 1 summarizes the sample size for each of the 11 samples. The estimation results

given in Tables 11 and 12 indicate that our findings are robust to the different samples and

our main qualitative results are broadly unaffected.

27

7.2 Alternative Instrumental Variables

Thus far, we have used the area measure Ai,t to summarize the variation in the marginal

benefit curve of debt. If the marginal benefit curve was a simple linear function with a

constant slope, summarizing the variation would be an easy task because the y-intercept, or

the x-intercept, of each function would suffice. However, not only is the functional form of

the benefit curves non-linear, the shape is potentially different for each firm for each year. In

other words, the variation of the marginal benefit curves is caused by more than just parallel

shifts. As a consequence, the challenge is to summarize this changing functional form of the

benefit curve, which, empirically, we achieve through the area measure.

To more fully represent the various shapes of the marginal benefit function, we

could include the y-intercept of the marginal benefit curve as an additional instrument.

Alternatively, we could include several variables that represent the area for different

partitions of benefit function, to more explicitly capture its shape over different ranges.

Yet another alternative is to approximate the marginal benefit curve in each year and for

each firm by a polynomial expansion (Taylor or Chebyshev polynomials) and include the

coefficients on these polynomials as the instruments. Our main conclusions are robust to

using these different instruments.

7.3 Time Series Analysis - Tax Regime Shifts

So far we have used both cross-sectional and time series variation of the marginal benefit

curves to estimate the marginal cost curve of debt. To purge potential variation of cost

effects from the benefit functions, we included a set of control variables that are theorized

to be correlated with the marginal cost curve. After purging the cost effects, we are left

with ”exogenous” variation of the marginal benefit curve, which we use as the identifying

instrument to recover the marginal cost curve. Even though we believe that we have

controlled for the most important cost effects in our analysis (i.e., those used in the capital

structure literature), there is no way to ensure that all relevant cost control variables are

included. Failing to include cost control variables could lead to an omitted variables bias.

To rigorously address this issue, we repeat our analysis in this section, using as the sole

identifying instrument the corporate tax regime shifts over the period 1980-2005 (the bottom-

up approach mentioned in the introduction). The tax regime shifts should be correlated with

the benefit function but not the cost function, thereby permitting identification of the cost

curve.

Tax regime changes lead to changes both in the level of the statutory corporate tax rates

28

.1.2

.3.4

.5T

ax R

ate

1980 1985 1990 1995 2000 2005Year

0 to 25 25 to 50 50 to 75 75 to 100

100 to 335 335 to 1000 1000 to 1405 1405 to 10000

10000 to 15000 15000 to 18333 18333+

Eleven Income Brackets in Thousands $Corporate Tax Rates Across Years

Figure 6: Corporate federal income tax rates for the eleven statutory tax brackets.

as well as the income level applicable to each tax bracket. To incorporate both shifts, we

create eleven non-overlapping brackets across all years in the sample period.13 For example

in 1980 the lowest tax bracket is from $0 to $25,000, but changes to $0 to $50,000 in 1988.

This results in two brackets in every year: one from $0 to $25,000 and one from $25,000 to

$50,000. Figure 6 plots the marginal tax for the eleven income brackets. Appendix C lists

the corporate tax rates for all eleven brackets in detail.

Time variation in the tax regimes can be considered a ‘cleaner’ instrument that is unlikely

to be correlated with cost effects. The disadvantage is that this approach uses much less

information to uncover the marginal cost curve of debt: it does not use the cross-sectional

variation in marginal benefit curves and therefore does not use the fact that we “observe”

the whole marginal benefit curve of debt, as opposed to just the equilibrium points where

the marginal cost and marginal benefit curves intersect.

Table 13 reports the coefficient estimates based on using tax regime shifts as an

instrument. The bottom-up analysis, which only uses the exogenous tax regime shifts as

13To address the possibility that the results from the time series analysis are largely driven by the rateshifts in the lower tax brackets, the entire analysis is repeated using only the top tax bracket. The resultsare similar to the results using all eleven brackets.

29

the identifying instrument, leads to similar regression results, for both the slope of the

marginal cost curve and the dependence of the intercept on the cost control variables. This

corroborates our earlier approach and results. The marginal cost curve is still steep with

a slope of the same order of magnitude as in our earlier analysis. Further, the signs and

magnitudes of the coefficients of the cost control variables, which influence the intercept and

the slope of the cost curve, respectively, are the same as well.

Finally, we also perform our main analysis based on data between 1998 and 2005. During

this period there were no exogenous tax regime shifts, meaning that for this sample, the time

series approach described above is not possible. The identification of the cost curve during

this period is mainly based on cross-sectional information, and it also leads to highly similar

results.14 This indicates that it does not seem to matter whether we identify the cost curve

based on (i) cross-sectional information, (ii) time series information, or (iii) a combination

of the two. We consider this to be a very reassuring corroboration of our findings.

8 Conclusion

We use panel data from 1980 to 2005 to estimate cost curves for corporate debt. We simulate

debt tax benefit curves and assume that for financially unconstrained firms, the benefit curve

intersects the cost curve at the actual level of debt, on average. Using this equilibrium

condition, exogenous shifts to the benefit curves enable us to identify the marginal cost

function. We recover marginal cost curves that are steeply positively sloped. Both the

slope and the intercept of these curves depend on firm characteristics such as collateral,

sales, book-to-market, cash flows, cash holdings, personal tax penalty, and whether the firm

pays dividends. We perform a multitude of robustness checks and find that our findings

are robust across industries and years. Our results are also robust to accounting for fixed

adjustment costs of debt. As such, our framework provides a new parsimonious environment

to estimate and evaluate competing capital structure theories. We also provide firm specific

recommendations of optimal debt policy against which firms’ actual debt choices can be

benchmarked.

14This point was further confirmed by including year dummies in our analysis.

30

References

Almeida, H., and T. Philippon, ”The Risk-Adjusted Cost of Financial Distress”, Journal ofFinance, forthcoming.

Altman, E., 1968, Financial Ratios, Discriminant Analysis and the Prediction of CorporateBankruptcy, Journal of Finance, 23, 4, 589-609.

Berk, J., Stanton, R., and J. Zechner, 2006, Human Capital, Bankruptcy and CapitalStructure, working paper UC Berkeley.

Brav, A., Graham, J., Harvey, C., and Michaely, R., 2005, Payout Policy in the 21st Century,Journal of Financial Economics, 77, 483-527.

Carlson, M., and A. Lazrak, 2006, Leverage Choice and Credit Spread Dynamics whenManagers Risk Shift, working paper, Sauder School of Business, UBC.

Chen, H., 2006, Macroeconomic Conditions and the Puzzles of Credit Spreads and CapitalStructure, working paper, University of Chicago GSB.

Chevalier, J.A., 1995, Capital Structure and Product Market Competition: EmpiricalEvidence from the Supermarket Industry, American Economic Review, 84, 3, 415-435.

Cleary, S., 1999, The Relationship between Firm Investment and Financial Status, Journalof Finance, 54, 2, 673-692.

Erickson, T., and T.M. Whited, 2006, On the accuracy of different measures of Q FinancialManagement, 35, 5-33.

Fazzari, S.M., R.G. Hubbard, B.C. Petersen, 1988, Financing constraints and coporateinvestment, Brookings Papers on Economic Activity, 1988, 1, 141-206.

Fischer, E., Heinkel, R. and J. Zechner, 1989, Optimal Dynamic Capital Structure Choice:Theory and Tests, Journal of Finance, 44, 19-40.

Frank, M.Z., and V.K. Goyal, 2004, Capital structure decisions: which factors are reliablyimportant?

Graham, J.R., 1999, Do personal taxes affect corporate financing decisions? Journal ofPublic Economics, 73, 147-185.

Graham, J.R., 2000, How big are the tax benefits of debt? Journal of Finance, 55, 1901-1941.

Graham, J.R., Lemmon, M. and J. Schallheim, 1998, Debt, Leases, Taxes, and theEndogeneity of Corporate Tax Status, Journal of Finance 53, 131-161.

Graham, J.R., and C. Smith, 1999, Tax Incentives to Hedge, Journal of Finance 54, 2241-2262.

31

Jensen, M.C., and W.H. Meckling, 1976, Theory of the flrm: Managerial behaviour, agencycosts and ownership structure, Journal of Financial Economics, 3, 305-360.

Kraus, A., and R. Litzenberg, 1973, A State-Preference Model of Optimal Financial Leverage,Journal of Finance, September, 911-922.

Kurshev, A. and I.A. Strebulaev, 2006, Firm Size and Capital Structure, working paperStanford GSB.

Lamont, O., C. Polk, and J. Saa-Requejo, 2001, Financial constraints and stock returns,Review of Financial Studies 14, 529-554.

Leary, M., and M. Roberts, 2005, Do Firms Rebalance their Capital Structures?, Journal ofFinance, 2575-2619.

MacKie-Mason, J., 1990, Do taxes affect corporate financing decisions?, Journal of Finance,45, 1471-1493.

Miller, M.H., 1977, Debt and Taxes, Journal of Finance, 32, 2, 261-275.

Myers, S.C., 1977, Determinants of corporate borrowing, Journal of Financial Economics 5,147-175.

Polk, C., R. Cohen, and T. Vuolteenaho, 2003, The value spread, Journal of Finance, 58,609-641.

Roberts, M., R. Michaely, 2006, Dividend Smoothing, Agency Costs, and InformationAsymmetry: Lessons from the Dividend Policies of Private Firms, working paper,Wharton, UPenn.

Scott, J.H. 1976, A theory of optimal capital structure, Bell Journal of Economics, 7, 33-54.

Strebulaev, I.A. and B. Yang, 2006, The Mystery of Zero-Leverage Firms, working paperStanford GSB.

Strebulaev, I.A., 2006, Do Tests of Capital Structure Theory Mean What They Say?,forthcoming, Journal of Finance.

Warner, J.B., 1977, Bankruptcy Costs: Some Evidence, Journal of Finance, 32(2), 337-347.

Whited, T., and G. Wu, 2005, Financial constraints risk, Review of Financial Studies, 19,531-559.

32

Appendix A

Detailed description on the construction of the control variables using in the analysis andvariables included in the summary statistics reported in Table 2. Numbers in parenthesesindicate the corresponding COMPUSTAT annual industrial data items.

PPE = Net Plants, Property, and Equipment (8)Total Book Assets (6)

LSALES = log (Net Sales (12))

BTM = Total Common Equity (60)Fiscal Year Close Price (199) *Common Shares Outstanding (54)

DDIV =

1 if Common Dividends (21) > 0

0 if Common Dividends (21) = 0

CF = EBIT (18) + Depreciation (14)Total Book Assets (6)

CASH = Cash and Short Term Investments (1)Total Book Assets (6)

HHI = H− 1N

1− 1N

where H =N∑i

s2i and si = market share of firm i

LWAGE = log (WAGE) = log(

Labor and Related Expenses (42)Number of Employees (29)

)

CS = Moody’s Baa Rate−Moody’s Aaa Rate (Source : Economagic)

ZSCORE = 3.3*Pretax Income (170) + 1.0*Net Sales (12) + 1.4*Retained Earnings (36) + 1.2*Working Capital (179)Total Book Assets (6)

TLC = Net Carryloss Forward (52)Net Sales (12)

PTP = τp − (1− τ c)τe for τ c = observed marginal tax rate and τe = [d + (1− d)gα]τp

where d is the dividend payout ratio, g is 0.4 before 1987 and 1.0 after (although gτp is never greater than

0.28), α is 0.25, and τp is 47.4% for 1980-1981, 40.7% for 1982-1986, 33.1% for 1987, 28.7% for 1988-1992,

and 29.6% for 1993 and onwards.

33

Appendix B

Firms are classified by their two-digit SIC codes into industries. The following table describes

the industries and the corresponding two-digit SIC codes we use in our analysis.

Agriculture, Forestry, and Fishing 01-09

Mining 10-14

Construction 15-17

Manufacturing 20-39

Transportation 40-47

Communication 48

Utilities 49

Wholesale Trade 50-51

Retail Trade 52-59

Finance, Insurance, and Real Estate 60-67

Services 70-89

Public Administration 91-99

34

Appendix C

Corporate tax rates over the period 1980 to 2005. Both the corporate tax rates as well as

their corresponding income tax brackets change during this period. To resolve this issue,

eleven non-overlapping income tax brackets are created that exist for all years.

Year Income Tax Bracket in Thousands $$

0 25 50 75 100 335 1000 1405 10000 15000 18333+to to to to to to to to to to25 50 75 100 335 1000 1405 10000 15000 18333

1980 0.170 0.200 0.300 0.400 0.460 0.460 0.510 0.460 0.460 0.460 0.460

1981 0.170 0.200 0.300 0.400 0.460 0.460 0.510 0.460 0.460 0.460 0.460

1982 0.160 0.190 0.300 0.400 0.460 0.460 0.510 0.460 0.460 0.460 0.460

1983 0.150 0.180 0.300 0.400 0.460 0.460 0.510 0.460 0.460 0.460 0.460

1984 0.150 0.180 0.300 0.400 0.460 0.460 0.510 0.460 0.460 0.460 0.460

1985 0.150 0.180 0.300 0.400 0.460 0.460 0.510 0.460 0.460 0.460 0.460

1986 0.150 0.180 0.300 0.400 0.460 0.460 0.510 0.460 0.460 0.460 0.460

1987 0.150 0.165 0.275 0.370 0.425 0.400 0.425 0.400 0.400 0.400 0.400

1988 0.150 0.150 0.250 0.340 0.390 0.340 0.340 0.340 0.340 0.340 0.340

1989 0.150 0.150 0.250 0.340 0.390 0.340 0.340 0.340 0.340 0.340 0.340

1990 0.150 0.150 0.250 0.340 0.390 0.340 0.340 0.340 0.340 0.340 0.340

1991 0.150 0.150 0.250 0.340 0.390 0.340 0.340 0.340 0.340 0.340 0.340

1992 0.150 0.150 0.250 0.340 0.390 0.340 0.340 0.340 0.340 0.340 0.340

1993 0.150 0.150 0.250 0.340 0.390 0.340 0.340 0.340 0.350 0.380 0.350

1994 0.150 0.150 0.250 0.340 0.390 0.340 0.340 0.340 0.350 0.380 0.350

1995 0.150 0.150 0.250 0.340 0.390 0.340 0.340 0.340 0.350 0.380 0.350

1996 0.150 0.150 0.250 0.340 0.390 0.340 0.340 0.340 0.350 0.380 0.350

1997 0.150 0.150 0.250 0.340 0.390 0.340 0.340 0.340 0.350 0.380 0.350

1998 0.150 0.150 0.250 0.340 0.390 0.340 0.340 0.340 0.350 0.380 0.350

1999 0.150 0.150 0.250 0.340 0.390 0.340 0.340 0.340 0.350 0.380 0.350

2000 0.150 0.150 0.250 0.340 0.390 0.340 0.340 0.340 0.350 0.380 0.350

2001 0.150 0.150 0.250 0.340 0.390 0.340 0.340 0.340 0.350 0.380 0.350

2002 0.150 0.150 0.250 0.340 0.390 0.340 0.340 0.340 0.350 0.380 0.350

2003 0.150 0.150 0.250 0.340 0.390 0.340 0.340 0.340 0.350 0.380 0.350

2004 0.150 0.150 0.250 0.340 0.390 0.340 0.340 0.340 0.350 0.380 0.350

2005 0.150 0.150 0.250 0.340 0.390 0.340 0.340 0.340 0.350 0.380 0.350

35

Table 1: Sample construction. y∗i,t is the ‘equilibrium’ marginal benefit level, x∗i,t is the observed or‘equilibrium’ interest payments over book value (IOB), A is the area under the marginal benefit curve, andC is the set of control (cost) variables. C ≡ {PPE, LSALES, BTM,DDIV, CF,CASH, PTP}. ZSCOREis a measure of financial distress. LTDEIR is the long term debt and/or equity issuance or repurchase usedto measure for financial constraint. KZ, CL, and WW are financial constraint measures as defined by Kaplanand Zingales (1997), Cleary (1999), and Whited and Wu (2005) indices respectively.

Sample No. Obs

All firm-year obs. with marginal benefit (MB) curves and COMPUSTAT data in 1980-2005 119,199

Non M&A firm-years with positive book value, common equity, capital, and sales 108,789

Sample without outliers 102,896

Sample with non-missing (y∗i,t, x∗i,t, Ai,t, Ci,t) variables: Sample A 80,039

Sample of firm-years without financial distress: Sample B 39,112ZSCORE above median

Sample of financially unconstrained firm-years: Sample C 57,310LTDEIR above median

Sample of financially unconstrained and non-distressed firm-years: Sample D 26,725LTDEIR above median and ZSCORE above median

For robustness checks:

Sample of financially unconstrained and non-distressed firm-years: Sample E 23,147KZ below median and ZSCORE above median

Sample of financially unconstrained and non-distressed firms: Sample F 20,542CL below median and ZSCORE above median

Sample of financially unconstrained and non-distressed firms: Sample G 20,668WW below median and ZSCORE above median

Sample of financially unconstrained and non-distressed firms: Sample H 13,843LTDEIR above third quartile and ZSCORE above median

Sample of financially unconstrained and non-distressed firms: Sample I 11,077KZ below first quartile and ZSCORE above median

Sample of financially unconstrained and non-distressed firms: Sample J 11,183CL below first quartile and ZSCORE above median

Sample of financially unconstrained and non-distressed firms: Sample K 9,176WW below first quartile and ZSCORE above median

36

Table 2: Summary statistics for the samples used in the analysis. HCR is the historical credit rankings basedon the S&P long term domestic issuer credit ratings. PPE is property, plant, and equipment over total bookvalues, LSALES is log of net sales expressed in 2000 dollars, BTM is book equity to market equity, DDIVis an indicator for dividend paying firms, CF is net cashflow over total book values, CASH is cash holdingsover total book values, PTP is the personal tax penalty on interest as measured in Graham (1999), and HHIis the normalized Herfindahl-Hirschman index for industry concentration based on the 3-digit SIC. ZSCOREis a measure of financial distress. UNFC is an indicator for financially unconstrained firms, defined as firmsthat adjust capital structure either through debt or equity issuance or repurchase. KZ, CL, and WW arefinancial constraint measures as defined by the Kaplan and Zingales (1997), Cleary (1999), and Whited andWu (2005) indices, respectively. TLC is total tax loss carryforwards normalized by total sales. WAGE istotal labor compensation normalized by total employees.

Sample A: All Firms

No. Obs Mean Std. Dev Min Max

Area Under Curve 80039 0.032 0.026 0.000 0.137Interest Over Book Value 80039 0.032 0.024 0.000 0.133HCR 15537 4.063 1.289 1 10PPE 80039 0.327 0.244 0.000 1.000LSALES 80039 5.014 2.367 -7.028 12.582BTM 80039 0.789 0.643 0.034 4.674DDIV 80039 0.430 0.495 0 1CF 80039 0.051 0.161 -4.847 3.121CASH 80039 0.133 0.172 -0.010 1.000PTP 80039 0.249 0.091 -0.171 0.451HHI 79711 0.188 0.153 0.000 1.000ZSCORE 73435 1.602 2.053 -13.7 5.595UNFC 80039 0.964 0.186 0 1CL 59176 1.935 86.078 -627.915 13843KZ 72529 -13.069 1280.381 -335562 1198WW 70372 -0.222 0.367 -21.715 50.016TLC 58852 2.686 106.326 0.000 15571WAGE 9102 53.442 398.182 0.000 20358

Sample B: Financially Non-distressed FirmsZSCORE above medianNo. Obs Mean Std. Dev Min Max

Area Under Curve 39112 0.041 0.027 0.000 0.137Interest Over Book Value 39112 0.028 0.022 0.000 0.133HCR 5838 3.771 1.250 1 10PPE 39112 0.281 0.173 0.000 0.953LSALES 39112 5.544 1.916 -5.606 12.582BTM 39112 0.768 0.595 0.035 4.639DDIV 39112 0.518 0.500 0 1CF 39112 0.109 0.071 -0.902 1.219CASH 39112 0.131 0.151 -0.002 0.992PTP 39112 0.259 0.088 -0.171 0.451HHI 38939 0.200 0.154 0.000 1.000ZSCORE 39112 2.850 0.772 1.801 5.60UNFC 39112 0.970 0.171 0 1CL 31465 -0.130 1.571 -207.099 17.033KZ 36499 -17.489 1800.499 -335562 25.312WW 37791 -0.233 0.202 -20.285 14.331TLC 32576 0.024 0.340 0.000 50.598WAGE 3665 38.786 285.333 0.003 14500

37

Sample C: Financially Unconstrained FirmsLTDEIR above medianNo. Obs Mean Std. Dev Min Max

Area Under Curve 57310 0.033 0.026 0.000 0.137Interest Over Book Value 57310 0.034 0.024 0.000 0.133HCR 12246 4.123 1.262 1 10PPE 57310 0.338 0.246 0.000 1.000LSALES 57310 5.104 2.333 -7.028 12.582BTM 57310 0.728 0.608 0.034 4.674DDIV 57310 0.403 0.490 0 1CF 57310 0.050 0.165 -4.847 1.898CASH 57310 0.128 0.170 -0.010 0.999PTP 57310 0.249 0.088 -0.171 0.451HHI 57063 0.186 0.152 0.000 1.000ZSCORE 53009 1.499 2.08 -13.69 5.59UNFC 57310 1.000 0.00 1 1CL 43179 2.285 98.65 -628 13843KZ 51522 -6.525 106.72 -14708 1198WW 50302 -0.226 0.389 -21.7 50.0TLC 41161 2.930 118.618 0.000 15571WAGE 6256 52.888 414.916 0.000 20357

Sample D: Financially Unconstrained and Non-distressedLTDEIR above median and ZSCORE above median

No. Obs Mean Std. Dev Min Max

Area Under Curve 26725 0.042 0.027 0.000 0.137Interest Over Book Value 26725 0.030 0.022 0.000 0.132HCR 4592 3.799 1.236 1 10PPE 26725 0.285 0.174 0.000 0.953LSALES 26725 5.668 1.873 -2.577 12.582BTM 26725 0.703 0.558 0.035 4.598DDIV 26725 0.490 0.500 0 1CF 26725 0.113 0.070 -0.902 1.219CASH 26725 0.125 0.149 -0.002 0.985PTP 26725 0.258 0.084 -0.171 0.451HHI 26598 0.198 0.153 0.000 0.993ZSCORE 26725 2.817 0.763 1.801 5.59UNFC 26725 1.000 0.000 1 1CL 21870 -0.103 1.648 -207.1 17.0KZ 24678 -5.574 62.77 -7213 21.8WW 25656 -0.238 0.222 -20.3 14.3TLC 21870 0.023 0.368 0.000 50.6WAGE 2355 37.006 194 0.003 9393

38

Table 3: Correlation between control variables used in the analysis. PPE is plant, property, and equipmentover total book values, LSALES is log of net sales expressed in 2000 dollars, BTM is book equity to marketequity, DDIV is an indicator for dividend paying firms, CF is net cashflow over total book values, CASH iscash holdings over total book values, and PTP is the personal tax penalty as measured in Graham (1999).KZ, CL, and WW are financial constraint measures as defined by the Kaplan Zingales (1997), Cleary (1999),and Whited Wu (2005) indices, respectively. ZSCORE is a measure of financial distress.

Sample A: All Firms

PPE LSALES BTM DDIV CF CASH

LSALES 0.1048BTM 0.0482 -0.1166DDIV 0.1560 0.5074 -0.0707CF 0.1413 0.3497 -0.0654 0.2579CASH -0.3442 -0.3085 -0.1504 -0.1707 -0.1638PTP -0.0266 -0.1698 0.0902 -0.1891 0.0011 -0.0029

Sample B: Financially Non-distressed FirmsZSCORE above median


LSALES 0.2405BTM -0.0196 -0.2072DDIV 0.2478 0.4157 -0.0949CF 0.2059 0.0397 -0.4281 0.1138CASH -0.3067 -0.2507 -0.1579 -0.1158 0.2176PTP -0.0260 -0.1835 0.1359 -0.1745 -0.0339 0.0009

Sample C: Financially Unconstrained FirmsLTDEIR above median


LSALES 0.0795BTM 0.0854 -0.0827DDIV 0.1451 0.5098 -0.0535CF 0.1388 0.3717 -0.0391 0.2436CASH -0.3537 -0.2948 -0.1922 -0.1934 -0.1968PTP -0.0077 -0.1845 0.0869 -0.2029 -0.0013 -0.0027

Sample D: Financially Unconstrained and Non-distressedLTDEIR above median and ZSCORE above median


LSALES 0.2095BTM 0.0172 -0.1972DDIV 0.2310 0.4207 -0.0798CF 0.1780 0.0261 -0.4213 0.0760CASH -0.3217 -0.2108 -0.1899 -0.1502 0.2335PTP -0.0194 -0.2134 0.1326 -0.1945 -0.0511 0.0148

39

Tab

le4:

Mar

gina

lcos

tof

debt

acro

ssm

ain

sam

ples

.G

MM

esti

mat

ion

ofth

eco

effici

ents

ineq

uati

ons

5an

d9

for

alli

ndus

trie

sex

cept

utili

ties

,fina

nce,

and

publ

icad

min

istr

atio

nfo

rsa

mpl

esA

thro

ugh

D.T

heer

ror

func

tion

sar

ede

fined

acco

rdin

gto

equa

tion

s6

and

10w

here

y∗ i,

tis

the

‘equ

ilibr

ium

’m

argi

nalbe

nefit

/cos

tle

vel,

x∗ i,

tis

the

obse

rved

or‘e

quili

briu

m’in

tere

stex

pens

esov

erbo

okva

lue

(IO

B)

and

Cis

the

set

ofco

stco

ntro

lva

riab

les.

GM

Mm

omen

tsar

eob

tain

edby

inte

ract

ing

the

erro

rfu

ncti

onw

ith

the

follo

win

gin

stru

men

ts:

the

cons

tant

term

,the

vari

atio

nof

the

mar

gina

lben

efit

curv

eA

i,t,a

ndea

chof

the

cont

rolv

aria

bles

.T

hese

tof

cont

rolv

aria

bles

,C,i

nclu

des{P

PE

,LS

AL

ES,B

TM

,DD

IV,C

F,C

AS

H,P

TP}w

here

PP

Eis

plan

t,pr

oper

ty,a

ndeq

uipm

ent

over

tota

lboo

kva

lues

,LSA

LE

Sis

log

ofne

tsa

les,

BT

Mis

book

equi

tyto

mar

ket

equi

ty,D

DIV

isan

indi

cato

rfo

rdi

vide

ndpa

ying

firm

,C

Fis

net

cash

flow

over

tota

lbo

okva

lues

,C

ASH

isca

shho

ldin

gsov

erto

talbo

okva

lues

,an

dP

TP

isth

epe

rson

alta

xpe

nalty

asm

easu

red

inG

raha

m(1

999)

.T

heco

ntro

lva

riab

les

are

stan

dard

ized

toha

vem

ean

zero

and

stan

dard

devi

atio

non

eba

sed

onSa

mpl

eA

(and

are

not

re-s

tand

ardi

zed

acro

sssa

mpl

es).

Rob

ust

GM

Mst

anda

rder

rors

are

repo

rted

inth

epa

rent

hese

s.Si

gnifi

canc

eat

the

10%

leve

lis

indi

cate

dby

*,5%

leve

lby

**,an

d1%

leve

lby

***.

(9)

MC

i,t

=a

+bx

i,t+P c∈C

θcc i

,tx

i,t+

ξi,

t(5

)M

Ci,

t=

a+

bxi,

t+P c∈C

δcc i

,t+

ξi,

t

Sam

ple

ASam

ple

BSam

ple

CSam

ple

DSam

ple

ASam

ple

BSam

ple

CSam

ple

D

Const

ant

0.0

17

***

0.2

02

***

-0.0

20

***

0.1

90

***

-0.0

12

***

0.1

34

***

-0.0

55

***

0.1

20

***

(0.0

05)

(0.0

03)

(0.0

06)

(0.0

04)

(0.0

05)

(0.0

04)

(0.0

07)

(0.0

05)

IOB

8.8

14

***

3.8

71

***

9.5

73

***

4.1

92

***

8.6

10

***

4.8

52

***

9.4

75

***

5.0

86

***

(0.1

56)

(0.1

25)

(0.2

02)

(0.1

57)

(0.1

41)

(0.1

00)

(0.1

93)

(0.1

28)

PP

E*IO

B-0

.731

***

-0.3

71

***

-0.8

52

***

-0.4

70

***

(0.0

31)

(0.0

40)

(0.0

37)

(0.0

46)

LSA

LE

S*IO

B0.6

61

***

0.3

29

***

0.8

60

***

0.4

05

***

(0.0

45)

(0.0

38)

(0.0

55)

(0.0

45)

BT

M*IO

B-0

.544

***

-0.4

53

***

-0.6

28

***

-0.5

18

***

(0.0

29)

(0.0

32)

(0.0

35)

(0.0

39)

DD

IV*IO

B1.8

30

***

1.2

11

***

1.6

58

***

1.0

66

***

(0.0

41)

(0.0

30)

(0.0

48)

(0.0

34)

CF*IO

B2.2

97

***

2.0

88

***

2.3

94

***

1.8

37

***

(0.0

78)

(0.1

81)

(0.0

86)

(0.2

06)

CA

SH

*IO

B2.4

98

***

1.4

16

***

2.6

99

***

1.5

04

***

(0.0

91)

(0.0

71)

(0.1

15)

(0.0

89)

PT

P*IO

B0.6

85

***

1.0

67

***

0.4

86

***

0.9

15

***

(0.0

31)

(0.0

27)

(0.0

39)

(0.0

31)

PP

E-0

.024

***

-0.0

13

***

-0.0

29

***

-0.0

19

***

(0.0

01)

(0.0

01)

(0.0

01)

(0.0

02)

LSA

LE

S0.0

18

***

0.0

11

***

0.0

26

***

0.0

15

***

(0.0

01)

(0.0

01)

(0.0

02)

(0.0

01)

BT

M-0

.020

***

-0.0

24

***

-0.0

26

***

-0.0

29

***

(0.0

01)

(0.0

01)

(0.0

01)

(0.0

02)

DD

IV0.0

81

***

0.0

51

***

0.0

78

***

0.0

47

***

(0.0

01)

(0.0

01)

(0.0

01)

(0.0

01)

CF

0.1

03

***

0.1

15

***

0.1

04

***

0.1

12

***

(0.0

02)

(0.0

05)

(0.0

03)

(0.0

05)

CA

SH

0.0

79

***

0.0

43

***

0.0

87

***

0.0

44

***

(0.0

02)

(0.0

01)

(0.0

02)

(0.0

02)

PT

P0.0

24

***

0.0

38

***

0.0

16

***

0.0

35

***

(0.0

01)

(0.0

01)

(0.0

01)

(0.0

01)

No.

Obs.

69569

37978

50345

25977

69569

37978

50345

25977

40

Table 5: Regressing area under the marginal benefits curve on the set of cost control variables. The setof control variables, C, includes {PPE,LSALES,BTM, DDIV, CASHFLOW,CASH,PTP} where PPEis plant, property, and equipment over total book values, LSALES is log of sales expressed in 2000 dollars,BTM is book equity to market equity, DDIV is an indicator for dividend paying firms, CF is net cashflowover total book values, CASH is cash holdings over total book values, and PTP is the personal tax penalty.Standard errors are reported in the parentheses. Significance at the 10% level is indicated by *, 5% level by**, and 1% level by ***.

Sample A Sample B Sample C Sample D

Panel A: Ai,t = α +P

c∈Cβcci,t + νi,t

PPE -0.0009 *** -0.0003 ** -0.0014 *** -0.0003 *(0.0001) (0.0001) (0.0001) (0.0002)

LSALES 0.0031 *** 0.0031 *** 0.0025 *** 0.0022 ***(0.0001) (0.0001) (0.0001) (0.0002)

BTM -0.0025 *** -0.0034 *** -0.0024 *** -0.0035 ***(0.0001) (0.0001) (0.0001) (0.0002)

DDIV 0.0049 *** 0.0039 *** 0.0053 *** 0.0043 ***(0.0001) (0.0001) (0.0001) (0.0002)

CF 0.0048 *** 0.0050 *** 0.0054 *** 0.0051 ***(0.0001) (0.0002) (0.0002) (0.0002)

CASH -0.0032 *** -0.0050 *** -0.0032 *** -0.0051 ***(0.0001) (0.0001) (0.0001) (0.0002)

PTP 0.0089 *** 0.0111 *** 0.0088 *** 0.0111 ***(0.0001) (0.0002) (0.0001) (0.0002)

Constant 0.0324 *** 0.0406 *** 0.0332 *** 0.0420 ***(0.0001) (0.0001) (0.0001) (0.0001)

No. Obs. 80039 39112 57310 26725R2 0.2446 0.2632 0.2468 0.2548

41

Tab

le6:

Mar

gina

lco

stof

debt

byin

dust

ry.

GM

Mes

tim

atio

nof

the

coeffi

cien

tsin

equa

tion

9(s

eebe

low

)by

indu

stry

.T

heer

ror

func

tion

isde

fined

ineq

uati

on10

whe

rey∗ i,

tis

the

‘equ

ilibr

ium

’m

argi

nal

bene

fit/c

ost

leve

l,x∗ i,

tis

the

obse

rved

or‘e

quili

briu

m’

inte

rest

expe

nses

over

book

valu

e(I

OB

)an

dC

isth

ese

tof

cost

cont

rol

vari

able

s.G

MM

mom

ents

are

obta

ined

byin

tera

ctin

gth

eer

ror

func

tion

wit

hth

efo

llow

ing

inst

rum

ents

:th

eco

nsta

ntte

rm,

the

vari

atio

nof

the

mar

gina

lbe

nefit

curv

eA

i,t,

and

each

ofth

eco

ntro

lva

riab

les.

The

set

ofco

ntro

lva

riab

les,

C,

incl

udes

{PP

E,L

SA

LE

S,B

TM

,DD

IV,C

F,C

AS

H,P

TP}w

here

PP

Eis

plan

t,pr

oper

ty,a

ndeq

uipm

ent

over

tota

lboo

kva

lues

,LSA

LE

Sis

log

ofne

tsa

les,

BT

Mis

book

equi

tyto

mar

keteq

uity

,DD

IVis

anin

dica

tor

for

divi

dend

payi

ngfir

m,C

Fis

netca

shflo

wov

erto

talb

ook

valu

es,C

ASH

isca

shho

ldin

gsov

erto

talbo

okva

lues

,an

dP

TP

ispe

rson

aliz

edta

xpe

nalty

asm

easu

red

inG

raha

m(1

999)

.T

heco

ntro

lva

riab

les

are

stan

dard

ized

toha

vem

ean

zero

and

stan

dard

devi

atio

non

efo

rea

chin

dust

ry.

Rob

ust

GM

Mst

anda

rder

rors

are

repo

rted

inth

epa

rent

hese

s.Si

gnifi

canc

eat

the

10%

leve

lis

indi

cate

dby

*,5%

leve

lby

**,an

d1%

leve

lby

***.

(9)

MC

i,t

=a

+bx

i,t+P c∈C

θcc i

,tx

i,t+

ξi,

t

Const

ant

IOB

PP

ELSA

LE

SB

TM

DD

IVC

FC

ASH

PT

PN

o.

Obs

*IO

B*IO

B*IO

B*IO

B*IO

B*IO

B*IO

BSam

ple

A:A

llFir

ms

All

Indust

ries

0.0

32

***

8.3

73

***

-0.5

07

***

0.6

97

***

-0.5

17

***

1.8

12

***

2.1

78

***

2.2

57

***

0.7

11

***

80039

(0.0

04)

(0.1

32)

(0.0

28)

(0.0

40)

(0.0

27)

(0.0

36)

(0.0

73)

(0.0

83)

(0.0

29)

All

Indust

ries

Exce

pt

Uti

liti

es,

0.0

17

***

8.8

14

***

-0.7

31

***

0.6

61

***

-0.5

44

***

1.8

30

***

2.2

97

***

2.4

98

***

0.6

85

***

69569

Fin

ance

,&

Public

Adm

in.

(0.0

05)

(0.1

56)

(0.0

31)

(0.0

45)

(0.0

29)

(0.0

41)

(0.0

78)

(0.0

91)

(0.0

31)

Min

ing

&C

onst

ruct

ion

-0.0

09

8.4

57

***

0.3

26

***

1.2

31

***

-0.5

00

***

1.2

73

***

2.4

66

***

2.3

98

***

0.4

98

***

6003

(0.0

11)

(0.4

41)

(0.1

26)

(0.2

05)

(0.1

08)

(0.1

40)

(0.2

79)

(0.2

96)

(0.1

20)

Uti

liti

es-0

.097

***

13.2

37***

-0.0

17

-0.1

88

-0.5

45

***

2.2

61

***

1.4

24

***

1.3

76

***

0.9

55

***

4013

(0.0

32)

(0.9

76)

(0.1

51)

(0.1

22)

(0.1

37)

(0.1

71)

(0.2

62)

(0.2

73)

(0.1

27)

Manufa

cturi

ng

0.0

14

11.1

29***

-0.0

30

0.6

54

***

-0.5

69

***

2.0

18

***

3.4

71

***

4.1

45

***

0.6

33

***

39888

(0.0

09)

(0.3

59)

(0.0

48)

(0.0

70)

(0.0

47)

(0.0

68)

(0.1

71)

(0.2

01)

(0.0

49)

Whole

sale

&R

etail

Tra

de

-0.0

45

***

7.7

52

***

-0.7

28

***

0.9

48

***

-0.6

30

***

1.1

51

***

1.1

05

***

0.3

93

***

0.8

12

***

9364

(0.0

17)

(0.3

85)

(0.0

63)

(0.0

85)

(0.0

62)

(0.0

73)

(0.1

28)

(0.1

09)

(0.0

61)

Tra

nsp

ort

ati

on,W

are

housi

ng,

-0.0

48

*8.8

90

***

-0.5

02

***

0.1

75

-0.4

52

***

2.1

65

***

2.1

43

***

0.5

36

***

0.5

87

***

3692

&C

om

munic

ati

on

(0.0

27)

(0.6

95)

(0.1

55)

(0.1

62)

(0.1

21)

(0.1

84)

(0.2

11)

(0.2

07)

(0.1

39)

Fin

ance

,In

sura

nce

,&

Rea

lE

state

0.1

73

***

4.0

67

***

-0.4

12

***

0.3

83

***

-0.2

84

***

0.8

76

***

1.4

82

***

0.7

97

***

0.5

30

***

5928

(0.0

06)

(0.1

66)

(0.0

55)

(0.1

01)

(0.0

57)

(0.0

77)

(0.1

98)

(0.1

41)

(0.0

72)

Ser

vic

es&

Lei

sure

-0.0

64

***

10.0

54***

-1.1

36

***

1.0

97

***

-0.3

98

***

1.0

43

***

1.7

03

***

3.7

11

***

0.5

50

***

10287

(0.0

13)

(0.4

06)

(0.0

79)

(0.1

06)

(0.0

72)

(0.0

81)

(0.1

35)

(0.2

35)

(0.0

72)

Sam

ple

D:Fin

anci

ally

Unco

nst

rain

edand

Non-d

istr

esse

d:

LT

DE

IRabove

med

ian

and

ZSC

OR

Eabove

med

ian

All

Indust

ries

0.1

91

***

4.1

16

***

-0.4

71

***

0.4

46

***

-0.5

14

***

1.0

83

***

1.7

11

***

1.4

82

***

1.0

16

***

26725

(0.0

04)

(0.1

50)

(0.0

47)

(0.0

45)

(0.0

38)

(0.0

34)

(0.1

92)

(0.0

84)

(0.0

34)

All

Indust

ries

Exce

pt

Uti

liti

es,

0.1

90

***

4.1

92

***

-0.4

70

***

0.4

05

***

-0.5

18

***

1.0

66

***

1.8

37

***

1.5

04

***

0.9

15

***

25977

Fin

ance

,&

Public

Adm

in.

(0.0

04)

(0.1

57)

(0.0

46)

(0.0

45)

(0.0

39)

(0.0

34)

(0.2

06)

(0.0

89)

(0.0

31)

Min

ing

&C

onst

ruct

ion

0.1

38

***

6.8

72

***

1.2

28

**

0.3

97

-0.7

07

**

0.6

14

**

1.8

83

**

1.3

74

**

0.5

34

*648

(0.0

23)

(1.3

13)

(0.5

76)

(0.5

45)

(0.2

89)

(0.2

97)

(0.8

71)

(0.6

31)

(0.2

89)

Uti

liti

es0.1

06

**

7.3

26

***

0.4

49

0.0

46

-1.0

31

**

1.4

48

***

-0.9

37

1.6

77

***

2.8

68

***

171

(0.0

53)

(1.6

20)

(0.4

91)

(0.5

45)

(0.5

21)

(0.4

82)

(0.8

41)

(0.5

03)

(0.6

02)

Manufa

cturi

ng

0.1

97

***

5.2

85

***

0.1

49

**

0.4

05

***

-0.4

77

***

1.0

62

***

2.8

65

***

2.7

86

***

0.9

32

***

16489

(0.0

05)

(0.2

97)

(0.0

60)

(0.0

61)

(0.0

65)

(0.0

47)

(0.4

41)

(0.1

59)

(0.0

39)

Whole

sale

&R

etail

Tra

de

0.0

75

***

5.2

08

***

-0.6

01

***

0.7

78

***

-0.6

01

***

0.7

77

***

1.0

13

***

0.1

40

0.7

82

***

4981

(0.0

14)

(0.3

09)

(0.0

71)

(0.0

89)

(0.0

68)

(0.0

65)

(0.1

57)

(0.0

95)

(0.0

63)

Tra

nsp

ort

ati

on,W

are

housi

ng,

0.1

66

***

3.0

70

***

-0.8

00

***

0.7

98

**

-0.5

90

***

0.5

70

***

1.8

08

***

0.1

25

1.0

14

***

712

&C

om

munic

ati

on

(0.0

21)

(0.5

25)

(0.2

50)

(0.3

33)

(0.1

85)

(0.1

96)

(0.5

27)

(0.2

01)

(0.1

73)

Fin

ance

,In

sura

nce

,&

Rea

lE

state

0.2

55

***

1.7

47

***

-0.7

25

***

0.5

27

*-0

.236

0.7

99

***

0.2

11

0.9

65

***

2.0

36

***

482

(0.0

15)

(0.6

11)

(0.2

63)

(0.2

78)

(0.2

18)

(0.2

17)

(0.4

71)

(0.2

55)

(0.4

09)

Ser

vic

es&

Lei

sure

0.1

56

***

5.1

01

***

-0.4

12

***

0.8

61

***

-0.5

14

***

0.5

02

***

0.5

21

*2.4

51

***

0.8

29

***

3068

(0.0

13)

(0.4

73)

(0.1

25)

(0.1

22)

(0.1

08)

(0.0

77)

(0.2

78)

(0.2

43)

(0.0

98)

42

Tab

le7:

Wag

esan

dla

bor

com

pens

atio

nfo

rco

rpor

ate

capi

tal

stru

ctur

ech

oice

s.G

MM

esti

mat

ion

ofth

eco

effici

ents

ineq

uati

ons

5an

d9

for

all

indu

stri

esfo

rsa

mpl

esA

thro

ugh

D.

The

erro

rfu

ncti

ons

are

defin

edac

cord

ing

toeq

uati

ons

6an

d10

whe

rey∗ i,

tis

the

‘equ

ilibr

ium

’m

argi

nal

bene

fit/c

ost

leve

l,x∗ i,

tis

the

obse

rved

or‘e

quili

briu

m’

inte

rest

expe

nses

over

book

valu

e(I

OB

)an

dC

isth

ese

tof

cost

cont

rol

vari

able

s.G

MM

mom

ents

are

obta

ined

byin

tera

ctin

gth

eer

ror

func

tion

wit

hth

efo

llow

ing

inst

rum

ents

:th

eco

nsta

ntte

rm,t

heva

riat

ion

ofth

em

argi

nalb

enefi

tcu

rve

Ai,

t,

and

each

ofth

eco

ntro

lva

riab

les.

The

set

ofco

ntro

lva

riab

les,

C,

incl

udes{P

PE

,LS

AL

ES,B

TM

,DD

IV,C

F,C

AS

H,P

TP}

whe

reP

PE

ispl

ant,

prop

erty

,an

deq

uipm

ent

over

tota

lbo

okva

lues

,LSA

LE

Sis

log

ofne

tsa

les,

BT

Mis

book

equi

tyto

mar

ket

equi

ty,D

DIV

isan

indi

cato

rfo

rdi

vide

ndpa

ying

firm

,C

Fis

net

cash

flow

over

tota

lbo

okva

lues

,C

ASH

isca

shho

ldin

gsov

erto

talbo

okva

lues

,an

dP

TP

isth

epe

rson

alta

xpe

nalty

asm

easu

red

inG

raha

m(1

999)

.T

heco

ntro

lva

riab

les

are

stan

dard

ized

toha

vem

ean

zero

and

stan

dard

devi

atio

non

eba

sed

onSa

mpl

eA

(and

are

not

re-s

tand

ardi

zed

acro

sssa

mpl

es).

Rob

ust

GM

Mst

anda

rder

rors

are

repo

rted

inth

epa

rent

hese

s.Si

gnifi

canc

eat

the

10%

leve

lis

indi

cate

dby

*,5%

leve

lby

**,an

d1%

leve

lby

***.

(9)

MC

i,t

=a

+bx

i,t+P c∈C

θcc i

,tx

i,t+

ξi,

t(5

)M

Ci,

t=

a+

bxi,

t+P c∈C

δcc i

,t+

ξi,

t

Sam

ple

ASam

ple

BSam

ple

CSam

ple

DSam

ple

ASam

ple

BSam

ple

CSam

ple

D

Const

ant

0.0

49

***

0.2

47

***

-0.0

79

***

0.1

99

***

0.0

16

0.1

88

***

-0.0

60

**

0.1

63

***

(0.0

15)

(0.0

09)

(0.0

30)

(0.0

15)

(0.0

15)

(0.0

10)

(0.0

24)

(0.0

15)

IOB

7.3

84

***

2.6

05

***

10.2

07***

3.7

74

***

7.1

51

***

3.2

22

***

8.6

47

***

3.7

36

***

(0.4

68)

(0.2

71)

(0.8

35)

(0.4

25)

(0.4

10)

(0.2

45)

(0.6

35)

(0.3

43)

PP

E*IO

B-0

.725

***

-0.2

17

***

-1.0

39

***

-0.3

91

***

(0.0

94)

(0.0

84)

(0.1

39)

(0.1

10)

LSA

LE

S*IO

B0.8

44

***

0.7

38

***

1.0

20

***

0.9

10

***

(0.1

29)

(0.1

02)

(0.1

88)

(0.1

41)

BT

M*IO

B-0

.846

***

-0.4

63

***

-1.1

76

***

-0.5

93

***

(0.1

02)

(0.0

93)

(0.1

54)

(0.1

31)

DD

IV*IO

B2.4

34

***

1.3

75

***

2.6

77

***

1.3

61

***

(0.1

36)

(0.0

92)

(0.2

05)

(0.1

26)

CF*IO

B1.4

02

***

0.7

11

***

1.4

33

***

0.6

09

**

(0.2

28)

(0.2

03)

(0.2

85)

(0.2

46)

CA

SH

*IO

B0.7

81

***

0.4

10

***

0.3

25

0.0

91

(0.2

08)

(0.1

30)

(0.3

01)

(0.1

86)

PT

P*IO

B1.3

22

***

1.5

50

***

0.8

88

***

1.2

19

***

(0.0

93)

(0.0

90)

(0.1

36)

(0.1

13)

LW

AG

E*IO

B0.6

88

***

0.3

50

***

0.9

82

***

0.5

76

***

(0.1

04)

(0.0

92)

(0.1

60)

(0.1

35)

PP

E-0

.021

***

-0.0

05

**

-0.0

30

***

-0.0

11

***

(0.0

03)

(0.0

03)

(0.0

04)

(0.0

03)

LSA

LE

S-0

.001

0.0

15

***

0.0

03

0.0

20

***

(0.0

04)

(0.0

02)

(0.0

05)

(0.0

04)

BT

M-0

.027

***

-0.0

21

***

-0.0

40

***

-0.0

31

***

(0.0

04)

(0.0

04)

(0.0

05)

(0.0

06)

DD

IV0.1

11

***

0.0

61

***

0.1

13

***

0.0

58

***

(0.0

05)

(0.0

03)

(0.0

07)

(0.0

04)

CF

0.1

03

***

0.0

61

***

0.1

06

***

0.0

56

***

(0.0

07)

(0.0

10)

(0.0

10)

(0.0

14)

CA

SH

0.0

45

***

0.0

22

***

0.0

35

***

0.0

17

***

(0.0

05)

(0.0

04)

(0.0

07)

(0.0

05)

PT

P0.0

38

***

0.0

47

***

0.0

30

***

0.0

42

***

(0.0

03)

(0.0

03)

(0.0

04)

(0.0

04)

LW

AG

E0.0

21

***

0.0

00

0.0

22

***

0.0

03

(0.0

04)

(0.0

02)

(0.0

05)

(0.0

03)

No.

Obs.

6413

3462

4447

2211

6413

3462

4447

2211

43

Tab

le8:

Mac

roec

onom

icin

fluen

ce.

GM

Mes

tim

atio

nof

the

coeffi

cien

tsin

equa

tion

s5

and

9fo

ral

lin

dust

ries

for

sam

ples

Ath

roug

hD

.T

heer

ror

func

tion

sar

ede

fined

acco

rdin

gto

equa

tion

s6

and

10,w

here

y∗ i,

tis

the

‘equ

ilibr

ium

’m

argi

nalbe

nefit

/cos

tle

vel,

x∗ i,

tis

the

obse

rved

or‘e

quili

briu

m’

inte

rest

expe

nses

over

book

valu

e(I

OB

)an

dC

isth

ese

tof

cost

cont

rolv

aria

bles

.G

MM

mom

ents

are

obta

ined

byin

tera

ctin

gth

eer

ror

func

tion

wit

hth

efo

llow

ing

inst

rum

ents

:th

eco

nsta

ntte

rm,th

eva

riat

ion

ofth

em

argi

nalbe

nefit

curv

eA

i,t,an

dea

chof

the

cont

rolva

riab

les.

The

set

ofco

ntro

lva

riab

les,

C,in

clud

es{P

PE

,LS

AL

ES,B

TM

,DD

IV,C

F,C

AS

H,P

TP

,CS}

whe

reP

PE

ispl

ant,

prop

erty

,an

deq

uipm

ent

over

tota

lbo

okva

lues

,LSA

LE

Sis

log

ofne

tsa

les,

BT

Mis

book

equi

tyto

mar

ket

equi

ty,D

DIV

isan

indi

cato

rfo

rdi

vide

ndpa

ying

firm

,CF

isne

tca

shflo

wov

erto

talb

ook

valu

es,C

ASH

isca

shho

ldin

gsov

erto

talbo

okva

lues

,P

TP

isth

epe

rson

alta

xpe

nalty

asm

easu

red

inG

raha

m(1

999)

,an

dC

Sis

Moo

dy’s

Baa

rate

min

usM

oody

’sA

aara

te.

The

cont

rol

vari

able

sar

est

anda

rdiz

edto

have

mea

nze

roan

dst

anda

rdde

viat

ion

one

base

don

Sam

ple

A(a

ndar

eno

tre

-sta

ndar

dize

dac

ross

sam

ples

).R

obus

tG

MM

stan

dard

erro

rsar

ere

port

edin

the

pare

nthe

ses.

Sign

ifica

nce

atth

e10

%le

velis

indi

cate

dby

*,5%

leve

lby

**,an

d1%

leve

lby

***.

(9)

MC

i,t

=a

+bx

i,t+P c∈C

θcc i

,tx

i,t+

ξi,

t(5

)M

Ci,

t=

a+

bxi,

t+P c∈C

δcc i

,t+

ξi,

t

Sam

ple

ASam

ple

BSam

ple

CSam

ple

DSam

ple

ASam

ple

BSam

ple

CSam

ple

D

Const

ant

0.0

13

***

0.2

06

***

-0.0

28

***

0.1

92

***

-0.0

30

***

0.1

38

***

-0.0

89

***

0.1

19

***

(0.0

05)

(0.0

03)

(0.0

06)

(0.0

04)

(0.0

05)

(0.0

04)

(0.0

08)

(0.0

05)

IOB

8.9

51

***

3.7

20

***

9.8

59

***

4.1

09

***

9.1

59

***

4.7

35

***

10.4

34***

5.1

25

***

(0.1

64)

(0.1

28)

(0.2

15)

(0.1

64)

(0.1

66)

(0.1

08)

(0.2

38)

(0.1

44)

PP

E*IO

B-0

.731

***

-0.3

76

***

-0.8

49

***

-0.4

72

***

(0.0

32)

(0.0

39)

(0.0

38)

(0.0

45)

LSA

LE

S*IO

B0.6

32

***

0.3

59

***

0.8

05

***

0.4

19

***

(0.0

45)

(0.0

37)

(0.0

56)

(0.0

44)

BT

M*IO

B-0

.514

***

-0.4

92

***

-0.5

68

***

-0.5

38

***

(0.0

29)

(0.0

32)

(0.0

36)

(0.0

39)

DD

IV*IO

B1.9

44

***

1.0

94

***

1.8

68

***

1.0

11

***

(0.0

46)

(0.0

33)

(0.0

55)

(0.0

39)

CF*IO

B2.3

11

***

2.0

94

***

2.4

21

***

1.8

40

***

(0.0

78)

(0.1

81)

(0.0

87)

(0.2

06)

CA

SH

*IO

B2.5

17

***

1.3

92

***

2.7

50

***

1.4

87

***

(0.0

92)

(0.0

70)

(0.1

19)

(0.0

89)

PT

P*IO

B0.8

79

***

0.8

59

***

0.8

64

***

0.8

13

***

(0.0

39)

(0.0

34)

(0.0

47)

(0.0

40)

CS*IO

B-0

.287

***

0.2

96

***

-0.5

65

***

0.1

48

***

(0.0

41)

(0.0

33)

(0.0

50)

(0.0

40)

PP

E-0

.025

***

-0.0

14

***

-0.0

30

***

-0.0

19

***

(0.0

01)

(0.0

01)

(0.0

01)

(0.0

02)

LSA

LE

S0.0

15

***

0.0

11

***

0.0

22

***

0.0

15

***

(0.0

01)

(0.0

01)

(0.0

02)

(0.0

01)

BT

M-0

.018

***

-0.0

25

***

-0.0

23

***

-0.0

29

***

(0.0

01)

(0.0

01)

(0.0

02)

(0.0

02)

DD

IV0.0

91

***

0.0

49

***

0.0

92

***

0.0

47

***

(0.0

02)

(0.0

01)

(0.0

02)

(0.0

01)

CF

0.1

05

***

0.1

15

***

0.1

09

***

0.1

13

***

(0.0

02)

(0.0

05)

(0.0

03)

(0.0

05)

CA

SH

0.0

83

***

0.0

42

***

0.0

94

***

0.0

45

***

(0.0

02)

(0.0

01)

(0.0

03)

(0.0

02)

PT

P0.0

36

***

0.0

35

***

0.0

37

***

0.0

36

***

(0.0

01)

(0.0

01)

(0.0

02)

(0.0

02)

CS

-0.0

23

***

0.0

05

***

-0.0

40

***

-0.0

02

(0.0

02)

(0.0

01)

(0.0

02)

(0.0

02)

No.

Obs.

69569

37978

50345

25977

69569

37978

50345

25977

44

Tab

le9:

Sum

mar

yst

atis

tics

for

bene

fitan

dco

stof

debt

.M

easu

res

are

base

don

the

mar

gina

lco

stcu

rves

esti

mat

edfr

omeq

uati

ons

(9)

and

(5)

for

sam

ple

Dfo

ral

lin

dust

ries

exce

ptut

iliti

es,

finan

ce,

and

publ

icad

min

stra

tion

.T

heob

serv

edgr

oss

bene

fits

ofde

bt,

GB

Do,

isth

ear

eaun

der

the

mar

gina

lben

efits

curv

eup

toth

eob

serv

edle

velo

fint

eres

tov

erbo

okva

lue

(IO

B).

The

obse

rved

cost

ofde

bt,C

Do

isth

ear

eaun

der

the

mar

gina

lcos

tcu

rve

upto

the

obse

rved

leve

lof

IOB

.T

heob

serv

edne

tbe

nefit

sof

debt

,N

BD

o,is

the

area

unde

rth

em

argi

nalbe

nefit

scu

rve

min

usth

ear

eaun

der

the

mar

gina

lco

stcu

rve

upto

the

obse

rved

IOB

.E

quili

briu

mis

defin

edas

the

inte

rsec

tion

ofth

em

argi

nal

bene

fitan

dco

stcu

rves

.T

heob

serv

edgr

oss

bene

fits

ofde

bt,G

BD

e,is

the

area

unde

rth

em

argi

nalbe

nefit

scu

rve

upto

the

equi

libri

umle

velof

IOB

.T

heeq

uilib

rium

cost

ofde

bt,C

De

isth

ear

eaun

der

the

mar

gina

lcos

tcu

rve

upto

the

equi

libri

umle

velo

fIO

B.T

heeq

uilib

rium

net

bene

fits

ofde

bt,N

BD

e,i

sth

ear

eaun

der

the

mar

gina

lbe

nefit

scu

rve

min

usth

ear

eaun

der

the

mar

gina

lcos

tcu

rve

upto

the

equi

libri

umIO

B.T

heco

stof

bein

gov

erle

vera

ged,

DW

o,i

sth

ede

adw

eigh

tlo

ssfr

omad

diti

onal

cost

sdu

eto

havi

ngIO

Bab

ove

the

equi

libri

um.

The

cost

ofbe

ing

unde

rlev

erag

ed,D

Wu,is

the

dead

wei

ght

loss

from

low

erbe

nefit

sdu

eto

havi

ngIO

Bbe

low

the

equi

libri

um.

The

cost

ofbe

ing

out

ofeq

uilib

rium

,D

Wt,co

mbi

nes

DW

oan

dD

Wu

into

one

mea

sure

.

Analy

sis

Base

don

(9)

MC

i,t

=a

+bx

i,t+P c∈C

θcc i

,tx

i,t+

ξi,

t

NM

ean

Std

.D

ev.

1%

10%

25%

Med

ian

75%

90%

99%

Obse

rved

gro

ssben

efits

ofdeb

t(G

BD

o)

85356

0.0

895

0.0

860

0.0

000

0.0

028

0.0

231

0.0

680

0.1

294

0.2

050

0.3

807

Obse

rved

cost

sofdeb

t(C

Do)

69569

0.0

874

0.0

838

0.0

000

0.0

125

0.0

312

0.0

650

0.1

170

0.1

888

0.3

986

Obse

rved

net

ben

efits

ofdeb

t(N

BD

o)

69569

0.0

024

0.0

572

-0.2

179

-0.0

593

-0.0

099

0.0

142

0.0

295

0.0

510

0.0

966

Equilib

rium

gro

ssben

efits

ofdeb

t(G

BD

e)

69569

0.1

014

0.0

919

0.0

000

0.0

000

0.0

126

0.0

939

0.1

548

0.2

126

0.3

707

Equilib

rium

cost

sofdeb

t(C

De)

69569

0.0

728

0.0

622

0.0

000

0.0

000

0.0

105

0.0

705

0.1

124

0.1

505

0.2

465

Equilib

rium

net

ben

efits

ofdeb

t(N

BD

o)

69569

0.0

285

0.0

384

0.0

000

0.0

000

0.0

017

0.0

224

0.0

411

0.0

623

0.1

316

Cost

ofbei

ng

out

ofeq

uilib

rium

(DW

t)

69569

0.0

262

0.0

506

0.0

000

0.0

003

0.0

023

0.0

093

0.0

270

0.0

713

0.2

451

Cost

ofover

lever

agin

g(D

Wo)

35521

0.0

401

0.0

612

0.0

000

0.0

004

0.0

036

0.0

178

0.0

524

0.1

086

0.2

847

Cost

ofunder

lever

agin

g(D

Wu)

34048

0.0

117

0.0

304

0.0

000

0.0

002

0.0

017

0.0

061

0.0

134

0.0

247

0.0

999

Analy

sis

Base

don

(5)

MC

i,t

=a

+bx

i,t+P c∈C

δcc i

,t+

ξi,

t

NM

ean

Std

.D

ev.

1%

10%

25%

Med

ian

75%

90%

99%

Obse

rved

gro

ssben

efits

ofdeb

t(G

BD

o)

85356

0.0

895

0.0

860

0.0

000

0.0

028

0.0

231

0.0

680

0.1

294

0.2

050

0.3

807

Obse

rved

cost

sofdeb

t(C

Do)

69569

0.0

729

0.0

854

-0.0

535

0.0

036

0.0

196

0.0

509

0.1

002

0.1

733

0.3

923

Obse

rved

net

ben

efits

ofdeb

t(N

BD

o)

69569

0.0

169

0.0

581

-0.1

993

-0.0

303

0.0

007

0.0

179

0.0

420

0.0

698

0.1

484

Equilib

rium

gro

ssben

efits

ofdeb

t(G

BD

e)

69569

0.1

022

0.0

831

0.0

000

0.0

000

0.0

259

0.0

954

0.1

574

0.2

162

0.3

223

Equilib

rium

cost

sofdeb

t(C

De)

69171

0.0

636

0.0

643

-0.0

676

0.0

000

0.0

113

0.0

652

0.1

025

0.1

501

0.1

949

Equilib

rium

net

ben

efits

ofdeb

t(N

BD

o)

69171

0.0

392

0.0

465

0.0

000

0.0

000

0.0

087

0.0

288

0.0

562

0.0

868

0.1

839

Cost

ofbei

ng

out

ofeq

uilib

rium

(DW

t)

69171

0.0

224

0.0

448

0.0

000

0.0

002

0.0

017

0.0

076

0.0

229

0.0

561

0.2

247

Cost

ofover

lever

agin

g(D

Wo)

32556

0.0

327

0.0

563

0.0

000

0.0

003

0.0

018

0.0

103

0.0

378

0.0

916

0.2

761

Cost

ofunder

lever

agin

g(D

Wu)

36615

0.0

133

0.0

281

0.0

000

0.0

002

0.0

015

0.0

063

0.0

161

0.0

313

0.0

964

45

Tab

le10

:C

ondi

tion

alsu

mm

ary

stat

isti

csfo

rbe

nefit

and

cost

mea

sure

sba

sed

onth

em

argi

nalco

stcu

rves

esti

mat

edfr

omeq

uati

ons

(9)

and

(5)

for

sam

ple

Dfo

ral

lin

dust

ries

exce

ptut

iliti

es,fi

nanc

e,an

dpu

blic

adm

inst

rati

on.

Pan

elA

grou

psob

serv

atio

nsba

sed

onho

wfa

ra

firm

-yea

rob

serv

atio

nac

tual

lyob

serv

edis

from

the

equi

libri

umfo

ri)

all

obse

rvat

ions

,ii)

over

leve

rage

dfir

m-y

ears

,an

diii

)un

derl

ever

aged

firm

-yea

rs.

Pan

elB

grou

psob

serv

atio

nsba

sed

onho

wfa

rth

eeq

uilib

rium

IOB

ofa

firm

-yea

rob

serv

atio

nis

from

the

obse

rved

actu

alIO

B.

The

cuto

ffsfo

llow

thos

eof

the

mar

gina

lben

efit

curv

es.

The

obse

rved

gros

sbe

nefit

sof

debt

,GB

Do,i

sth

ear

eaun

der

the

mar

gina

lben

efits

curv

eup

toth

eob

serv

edle

velo

fin

tere

stov

erbo

okva

lue

(IO

B).

The

obse

rved

cost

ofde

bt,

CD

ois

the

area

unde

rth

em

argi

nal

cost

curv

eup

toth

eob

serv

edle

vel

ofIO

B.

The

obse

rved

net

bene

fits

ofde

bt,

NB

Do,

isth

ear

eaun

der

the

mar

gina

lbe

nefit

scu

rve

min

usth

ear

eaun

der

the

mar

gina

lco

stcu

rve

upto

the

obse

rved

IOB

.E

quili

briu

mis

defin

edas

the

inte

rsec

tion

ofth

em

argi

nalbe

nefit

and

cost

curv

es.

The

obse

rved

gros

sbe

nefit

sof

debt

,G

BD

e,is

the

area

unde

rth

em

argi

nalbe

nefit

scu

rve

upto

the

equi

libri

umle

velof

IOB

.T

heeq

uilib

rium

cost

ofde

bt,C

De

isth

ear

eaun

der

the

mar

gina

lco

stcu

rve

upto

the

equi

libri

umle

velo

fIO

B.T

heeq

uilib

rium

net

bene

fits

ofde

bt,N

BD

e,i

sth

ear

eaun

der

the

mar

gina

lben

efits

curv

em

inus

the

area

unde

rth

em

argi

nal

cost

curv

eup

toth

eeq

uilib

rium

IOB

.T

heco

stof

bein

gov

erle

vera

ged,

DW

o,is

the

dead

wei

ght

loss

from

addi

tion

alco

sts

due

toha

ving

IOB

abov

eth

eeq

uilib

rium

.T

heco

stof

bein

gun

derl

ever

aged

,D

Wu,is

the

dead

wei

ght

loss

from

low

erbe

nefit

sdu

eto

havi

ngIO

Bbe

low

the

equi

libri

um.

The

cost

ofbe

ing

out

ofeq

uilib

rium

,D

Wt,co

mbi

nes

DW

oan

dD

Wu

into

one

mea

sure

.

Panel

A:G

roupin

gby

per

centa

ge

ofobse

rved

IOB

from

equilib

rium

IOB

Analy

sis

Base

don

(9)

MC

i,t

=a

+bx

i,t+P c∈C

θcc i

,tx

i,t+

ξi,

tA

naly

sis

Base

don

(5)

MC

i,t

=a

+bx

i,t+P c∈C

δcc i

,t+

ξi,

t

NG

BD

oC

Do

NB

Do

GB

De

CD

eN

BD

eD

Wt

NG

BD

oC

Do

NB

Do

GB

De

CD

eN

BD

eD

Wt

All

Obse

rvati

ons

Wit

hin

5%

ofeq

uilib

rium

2897

0.1

453

0.1

065

0.0

388

0.1

471

0.1

070

0.0

400

0.0

012

3281

0.1

354

0.0

856

0.0

498

0.1

358

0.0

859

0.0

499

0.0

001

Bet

wee

n5%

-10%

from

equilib

rium

3035

0.1

396

0.1

010

0.0

386

0.1

428

0.1

023

0.0

405

0.0

019

3346

0.1

323

0.0

829

0.0

494

0.1

338

0.0

840

0.0

498

0.0

004

Bet

wee

n10%

-20%

from

equilib

rium

6052

0.1

388

0.1

009

0.0

379

0.1

466

0.1

054

0.0

412

0.0

033

6744

0.1

315

0.0

829

0.0

486

0.1

364

0.0

862

0.0

502

0.0

016

Bet

wee

n20%

-40%

from

equilib

rium

11551

0.1

255

0.0

906

0.0

348

0.1

477

0.1

067

0.0

411

0.0

062

12977

0.1

177

0.0

728

0.0

448

0.1

369

0.0

867

0.0

502

0.0

054

Bet

wee

n40%

-60%

from

equilib

rium

10884

0.1

001

0.0

728

0.0

273

0.1

445

0.1

037

0.0

408

0.0

134

11967

0.0

930

0.0

574

0.0

357

0.1

323

0.0

828

0.0

495

0.0

139

Bet

wee

n60%

-80%

from

equilib

rium

8231

0.0

701

0.0

528

0.0

173

0.1

345

0.0

965

0.0

380

0.0

207

9222

0.0

636

0.0

408

0.0

228

0.1

175

0.0

713

0.0

462

0.0

234

More

than

80%

from

equilib

rium

26919

0.0

536

0.0

959

-0.0

423

0.0

341

0.0

242

0.0

099

0.0

521

21634

0.0

575

0.0

904

-0.0

329

0.0

394

0.0

222

0.0

172

0.0

501

Over

lever

aged

Obse

rvati

ons

NG

BD

oC

Do

NB

Do

GB

De

CD

eN

BD

eD

Wo

NG

BD

oC

Do

NB

Do

GB

De

CD

eN

BD

eD

Wo

Wit

hin

5%

ofeq

uilib

rium

1234

0.1

455

0.1

080

0.0

376

0.1

431

0.1

045

0.0

386

0.0

011

1550

0.1

351

0.0

867

0.0

484

0.1

320

0.0

835

0.0

485

0.0

001

Bet

wee

n5%

-10%

from

equilib

rium

1320

0.1

460

0.1

095

0.0

366

0.1

384

0.0

996

0.0

388

0.0

022

1488

0.1

395

0.0

914

0.0

481

0.1

305

0.0

820

0.0

485

0.0

004

Bet

wee

n10%

-20%

from

equilib

rium

2458

0.1

567

0.1

199

0.0

368

0.1

415

0.0

999

0.0

415

0.0

047

2805

0.1

481

0.1

003

0.0

478

0.1

306

0.0

809

0.0

497

0.0

018

Bet

wee

n20%

-40%

from

equilib

rium

3476

0.1

715

0.1

398

0.0

317

0.1

379

0.0

998

0.0

381

0.0

064

4088

0.1

543

0.1

130

0.0

413

0.1

228

0.0

762

0.0

466

0.0

053

Bet

wee

n40%

-60%

from

equilib

rium

2666

0.1

763

0.1

551

0.0

212

0.1

265

0.0

909

0.0

356

0.0

144

3097

0.1

570

0.1

270

0.0

299

0.1

118

0.0

678

0.0

440

0.0

141

Bet

wee

n60%

-80%

from

equilib

rium

1597

0.1

834

0.1

705

0.0

129

0.1

175

0.0

846

0.0

329

0.0

200

2035

0.1

540

0.1

360

0.0

180

0.0

977

0.0

592

0.0

386

0.0

205

More

than

80%

from

equilib

rium

22770

0.0

605

0.1

118

-0.0

513

0.0

222

0.0

158

0.0

065

0.0

578

17493

0.0

679

0.1

112

-0.0

433

0.0

275

0.0

165

0.0

111

0.0

543

Under

lever

aged

Obse

rvati

ons

NG

BD

oC

Do

NB

Do

GB

De

CD

eN

BD

eD

Wu

NG

BD

oC

Do

NB

Do

GB

De

CD

eN

BD

eD

Wu

Wit

hin

5%

ofeq

uilib

rium

1663

0.1

451

0.1

054

0.0

397

0.1

500

0.1

089

0.0

411

0.0

014

1731

0.1

357

0.0

847

0.0

510

0.1

391

0.0

880

0.0

511

0.0

001

Bet

wee

n5%

-10%

from

equilib

rium

1715

0.1

347

0.0

944

0.0

402

0.1

463

0.1

044

0.0

418

0.0

016

1858

0.1

265

0.0

761

0.0

504

0.1

364

0.0

855

0.0

509

0.0

005

Bet

wee

n10%

-20%

from

equilib

rium

3594

0.1

265

0.0

879

0.0

385

0.1

501

0.1

092

0.0

410

0.0

024

3939

0.1

197

0.0

705

0.0

492

0.1

405

0.0

899

0.0

506

0.0

014

Bet

wee

n20%

-40%

from

equilib

rium

8075

0.1

056

0.0

694

0.0

362

0.1

520

0.1

096

0.0

423

0.0

062

8889

0.1

008

0.0

544

0.0

464

0.1

434

0.0

915

0.0

519

0.0

055

Bet

wee

n40%

-60%

from

equilib

rium

8218

0.0

754

0.0

461

0.0

293

0.1

503

0.1

079

0.0

424

0.0

131

8870

0.0

707

0.0

330

0.0

377

0.1

395

0.0

880

0.0

515

0.0

138

Bet

wee

n60%

-80%

from

equilib

rium

6634

0.0

428

0.0

244

0.0

184

0.1

386

0.0

994

0.0

392

0.0

209

7187

0.0

380

0.0

138

0.0

241

0.1

231

0.0

747

0.0

484

0.0

242

More

than

80%

from

equilib

rium

4149

0.0

159

0.0

087

0.0

073

0.0

993

0.0

708

0.0

285

0.0

213

4141

0.0

135

0.0

027

0.0

108

0.0

896

0.0

464

0.0

432

0.0

324

46

Panel

B:G

roupin

gby

per

centa

ge

ofeq

uilib

rium

IOB

from

obse

rved

IOB

Analy

sis

Base

don

(9)

MC

i,t

=a

+bx

i,t+P c∈C

θcc i

,tx

i,t+

ξi,

tA

naly

sis

Base

don

(5)

MC

i,t

=a

+bx

i,t+P c∈C

δcc i

,t+

ξi,

t

Over

lever

aged

Obse

rvati

ons

NG

BD

oC

Do

NB

Do

GB

De

CD

eN

BD

eD

Wo

NG

BD

oC

Do

NB

Do

GB

De

CD

eN

BD

eD

Wo

Wit

hin

5%

ofobse

rved

1308

0.1

454

0.1

078

0.0

376

0.1

433

0.1

041

0.0

392

0.0

015

1619

0.1

351

0.0

867

0.0

484

0.1

319

0.0

833

0.0

485

0.0

001

Bet

wee

n80%

-95%

ofobse

rved

4807

0.1

574

0.1

215

0.0

359

0.1

415

0.1

012

0.0

403

0.0

043

5461

0.1

483

0.1

014

0.0

469

0.1

310

0.0

821

0.0

489

0.0

020

Bet

wee

n60%

-80%

ofobse

rved

5688

0.1

754

0.1

512

0.0

242

0.1

294

0.0

934

0.0

360

0.0

119

6695

0.1

556

0.1

226

0.0

330

0.1

141

0.0

698

0.0

443

0.0

113

Bet

wee

n40%

-60%

ofobse

rved

4194

0.1

738

0.1

772

-0.0

034

0.1

000

0.0

713

0.0

287

0.0

321

5419

0.1

385

0.1

385

-0.0

001

0.0

788

0.0

468

0.0

319

0.0

320

Bet

wee

n20%

-40%

ofobse

rved

2884

0.1

308

0.1

696

-0.0

388

0.0

568

0.0

390

0.0

178

0.0

566

4103

0.0

916

0.1

349

-0.0

433

0.0

365

0.0

214

0.0

150

0.0

584

Les

sth

an

20%

ofobse

rved

16640

0.0

267

0.0

884

-0.0

618

0.0

018

0.0

014

0.0

004

0.0

622

9259

0.0

274

0.0

874

-0.0

600

0.0

027

0.0

018

0.0

009

0.0

609

Under

lever

aged

Obse

rvati

ons

NG

BD

oC

Do

NB

Do

GB

De

CD

eN

BD

eD

Wu

NG

BD

oC

Do

NB

Do

GB

De

CD

eN

BD

eD

Wu

Wit

hin

5%

ofobse

rved

1589

0.1

448

0.1

053

0.0

395

0.1

495

0.1

086

0.0

409

0.0

014

1621

0.1

361

0.0

849

0.0

511

0.1

393

0.0

881

0.0

512

0.0

001

Bet

wee

n105%

-120%

ofobse

rved

3992

0.1

302

0.0

911

0.0

390

0.1

466

0.1

055

0.0

411

0.0

021

4448

0.1

232

0.0

737

0.0

496

0.1

375

0.0

871

0.0

504

0.0

008

Bet

wee

n120%

-160%

ofobse

rved

8279

0.1

102

0.0

734

0.0

368

0.1

512

0.1

092

0.0

420

0.0

052

9078

0.1

049

0.0

578

0.0

471

0.1

428

0.0

913

0.0

515

0.0

044

Bet

wee

n160%

-200%

ofobse

rved

5232

0.0

860

0.0

539

0.0

321

0.1

520

0.1

098

0.0

422

0.0

101

5678

0.0

811

0.0

404

0.0

408

0.1

424

0.0

911

0.0

512

0.0

105

Bet

wee

n200%

-300%

ofobse

rved

6329

0.0

635

0.0

377

0.0

258

0.1

490

0.1

067

0.0

423

0.0

165

6852

0.0

579

0.0

242

0.0

337

0.1

351

0.0

833

0.0

518

0.0

181

Bet

wee

n300%

-400%

ofobse

rved

2762

0.0

437

0.0

250

0.0

187

0.1

455

0.1

045

0.0

409

0.0

222

3061

0.0

374

0.0

135

0.0

239

0.1

257

0.0

767

0.0

490

0.0

251

Bet

wee

n400%

-500%

ofobse

rved

1716

0.0

283

0.0

155

0.0

128

0.1

207

0.0

859

0.0

347

0.0

219

1726

0.0

254

0.0

082

0.0

173

0.1

110

0.0

661

0.0

448

0.0

276

Bet

wee

n500%

-600%

ofobse

rved

1503

0.0

167

0.0

091

0.0

076

0.0

863

0.0

611

0.0

252

0.0

176

1249

0.0

186

0.0

046

0.0

141

0.1

012

0.0

551

0.0

461

0.0

321

Bet

wee

n600%

-700%

ofobse

rved

921

0.0

179

0.0

098

0.0

082

0.1

078

0.0

754

0.0

324

0.0

242

915

0.0

133

0.0

019

0.0

114

0.0

856

0.0

406

0.0

450

0.0

336

Bet

wee

n700%

-800%

ofobse

rved

691

0.0

165

0.0

091

0.0

075

0.1

098

0.0

783

0.0

315

0.0

240

760

0.0

112

0.0

021

0.0

091

0.0

825

0.0

418

0.0

407

0.0

316

Bet

wee

n800%

-900%

ofobse

rved

599

0.0

127

0.0

067

0.0

060

0.1

047

0.0

739

0.0

309

0.0

249

659

0.0

103

0.0

005

0.0

098

0.0

867

0.0

387

0.0

480

0.0

382

More

than

900%

ofobse

rved

435

0.0

125

0.0

068

0.0

056

0.1

024

0.0

785

0.0

238

0.0

182

568

0.0

088

0.0

029

0.0

059

0.0

821

0.0

509

0.0

312

0.0

253

47

Tab

le11

:A

naly

sis

onqu

alifi

ed50

perc

ent

offin

anci

ally

unco

nstr

aine

dfir

ms

usin

gfo

urfin

anci

alco

nstr

aint

mea

sure

s:(i

)lo

ngte

rmde

btor

equi

tyis

suan

ceor

repu

rcha

se,(

ii)K

Zin

dex,

(iii)

CL

inde

x,(i

v)W

Win

dex.

GM

Mes

tim

atio

nof

the

coeffi

cien

tsin

equa

tion

s5

and

9fo

ral

lind

ustr

ies

exce

ptut

iliti

es,fi

nanc

e,an

dpu

blic

adm

inst

rati

onfo

rsa

mpl

esD

thro

ugh

G.T

heer

ror

func

tion

sar

ede

fined

acco

rdin

gto

equa

tion

s6

and

10w

here

y∗ i,

tis

the

‘equ

ilibr

ium

’mar

gina

lben

efit/

cost

leve

l,x∗ i,

tis

the

obse

rved

or‘e

quili

briu

m’i

nter

est

expe

nses

over

book

valu

e(I

OB

)an

dC

isth

ese

tof

cost

cont

rol

vari

able

s.G

MM

mom

ents

are

obta

ined

byin

tera

ctin

gth

eer

rorfu

ncti

onw

ith

the

follo

win

gin

stru

men

ts:

the

cons

tant

term

,the

vari

atio

nof

the

mar

gina

lbe

nefit

curv

eA

i,t,a

ndea

chof

the

cont

rolv

aria

bles

.T

hese

tof

cont

rolv

aria

bles

,C,i

nclu

des{P

PE

,LS

AL

ES,B

TM

,DD

IV,C

F,C

AS

H,P

TP}w

here

PP

Eis

plan

t,pr

oper

ty,a

ndeq

uipm

ent

over

tota

lboo

kva

lues

,LSA

LE

Sis

log

ofne

tsa

les,

BT

Mis

book

equi

tyto

mar

ket

equi

ty,D

DIV

isan

indi

cato

rfo

rdi

vide

ndpa

ying

firm

,CF

isne

tca

shflo

wov

erto

talb

ook

valu

es,C

ASH

isca

shho

ldin

gsov

erto

talb

ook

valu

es,a

ndP

TP

isth

epe

rson

alta

xpe

nalty

asm

easu

red

inG

raha

m(1

999)

.T

heco

ntro

lva

riab

les

are

stan

dard

ized

toha

vem

ean

zero

and

stan

dard

devi

atio

non

eba

sed

onSa

mpl

eA

(and

are

not

re-s

tand

ardi

zed

acro

sssa

mpl

es).

Rob

ust

GM

Mst

anda

rder

rors

are

repo

rted

inth

epa

rent

hese

s.Si

gnifi

canc

eat

the

10%

leve

lis

indi

cate

dby

*,5%

leve

lby

**,an

d1%

leve

lby

***.

(9)

MC

i,t

=a

+bx

i,t+P c∈C

θcc i

,tx

i,t+

ξi,

t(5

)M

Ci,

t=

a+

bxi,

t+P c∈C

δcc i

,t+

ξi,

t

Sam

ple

DSam

ple

ESam

ple

FSam

ple

GSam

ple

DSam

ple

ESam

ple

FSam

ple

G

Const

ant

0.1

90

***

0.2

73

***

0.2

35

***

0.2

61

***

0.1

20

***

0.2

37

***

0.2

02

***

0.1

93

***

(0.0

04)

(0.0

02)

(0.0

03)

(0.0

03)

(0.0

05)

(0.0

03)

(0.0

04)

(0.0

04)

IOB

4.1

92

***

2.6

15

***

3.7

68

***

2.2

63

***

5.0

86

***

3.3

01

***

3.9

21

***

3.8

23

***

(0.1

57)

(0.1

13)

(0.1

38)

(0.1

24)

(0.1

28)

(0.0

90)

(0.1

11)

(0.1

10)

PP

E*IO

B-0

.470

***

0.3

79

***

-0.1

84

***

-0.2

07

***

(0.0

46)

(0.0

59)

(0.0

52)

(0.0

40)

LSA

LE

S*IO

B0.4

05

***

0.1

36

***

0.1

53

***

0.1

58

***

(0.0

45)

(0.0

44)

(0.0

55)

(0.0

46)

BT

M*IO

B-0

.518

***

-0.2

43

***

-0.5

36

***

-0.2

55

***

(0.0

39)

(0.0

45)

(0.0

52)

(0.0

48)

DD

IV*IO

B1.0

66

***

0.8

85

***

1.2

67

***

0.9

52

***

(0.0

34)

(0.0

39)

(0.0

43)

(0.0

30)

CF*IO

B1.8

37

***

1.2

36

***

0.9

24

***

2.0

43

***

(0.2

06)

(0.1

46)

(0.1

53)

(0.2

15)

CA

SH

*IO

B1.5

04

***

0.7

48

***

0.9

25

***

0.6

61

***

(0.0

89)

(0.0

54)

(0.0

76)

(0.0

71)

PT

P*IO

B0.9

15

***

1.3

76

***

1.4

34

***

1.0

86

***

(0.0

31)

(0.0

37)

(0.0

48)

(0.0

32)

PP

E-0

.019

***

0.0

12

***

-0.0

06

***

-0.0

08

***

(0.0

02)

(0.0

01)

(0.0

01)

(0.0

01)

LSA

LE

S0.0

15

***

0.0

07

***

0.0

08

***

0.0

00

(0.0

01)

(0.0

01)

(0.0

01)

(0.0

01)

BT

M-0

.029

***

-0.0

15

***

-0.0

23

***

-0.0

20

***

(0.0

02)

(0.0

02)

(0.0

02)

(0.0

02)

DD

IV0.0

47

***

0.0

33

***

0.0

43

***

0.0

37

***

(0.0

01)

(0.0

01)

(0.0

01)

(0.0

01)

CF

0.1

12

***

0.0

51

***

0.0

63

***

0.0

92

***

(0.0

05)

(0.0

04)

(0.0

04)

(0.0

05)

CA

SH

0.0

44

***

0.0

22

***

0.0

24

***

0.0

23

***

(0.0

02)

(0.0

01)

(0.0

01)

(0.0

01)

PT

P0.0

35

***

0.0

43

***

0.0

43

***

0.0

36

***

(0.0

01)

(0.0

01)

(0.0

01)

(0.0

01)

No.

Obs.

25977

22387

19942

20176

25977

22387

19942

20176

48

Tab

le12

:A

naly

sis

onqu

alifi

ed25

perc

ent

offin

anci

ally

unco

nstr

aine

dfir

ms

usin

gfo

urfin

anci

alco

nstr

aint

mea

sure

s:(i

)lo

ngte

rmde

btor

equi

tyis

suan

ceor

repu

rcha

se,(

ii)K

Zin

dex,

(iii)

CL

inde

x,(i

v)W

Win

dex.

GM

Mes

tim

atio

nof

the

coeffi

cien

tsin

equa

tion

s5

and

9fo

ral

lind

ustr

ies

exce

ptut

iliti

es,fi

nanc

e,an

dpu

blic

adm

inis

trat

ion

for

sam

ples

Hth

roug

hK

.The

erro

rfu

ncti

ons

are

defin

edac

cord

ing

toeq

uati

ons

6an

d10

whe

rey∗ i,

tis

the

‘equ

ilibr

ium

’mar

gina

lben

efit/

cost

leve

l,x∗ i,

tis

the

obse

rved

or‘e

quili

briu

m’i

nter

est

expe

nses

over

book

valu

e(I

OB

)an

dC

isth

ese

tof

cost

cont

rol

vari

able

s.G

MM

mom

ents

are

obta

ined

byin

tera

ctin

gth

eer

rorfu

ncti

onw

ith

the

follo

win

gin

stru

men

ts:

the

cons

tant

term

,the

vari

atio

nof

the

mar

gina

lbe

nefit

curv

eA

i,t,a

ndea

chof

the

cont

rolv

aria

bles

.T

hese

tof

cont

rolv

aria

bles

,C,i

nclu

des{P

PE

,LS

AL

ES,B

TM

,DD

IV,C

F,C

AS

H,P

TP}w

here

PP

Eis

plan

t,pr

oper

ty,a

ndeq

uipm

ent

over

tota

lboo

kva

lues

,LSA

LE

Sis

log

ofne

tsa

les,

BT

Mis

book

equi

tyto

mar

ket

equi

ty,D

DIV

isan

indi

cato

rfo

rdi

vide

ndpa

ying

firm

,CF

isne

tca

shflo

wov

erto

talb

ook

valu

es,C

ASH

isca

shho

ldin

gsov

erto

talb

ook

valu

es,a

ndP

TP

isth

epe

rson

alta

xpe

nalty

asm

easu

red

inG

raha

m(1

999)

.T

heco

ntro

lva

riab

les

are

stan

dard

ized

toha

vem

ean

zero

and

stan

dard

devi

atio

non

eba

sed

onSa

mpl

eA

(and

are

not

re-s

tand

ardi

zed

acro

sssa

mpl

es).

Rob

ust

GM

Mst

anda

rder

rors

are

repo

rted

inth

epa

rent

hese

s.Si

gnifi

canc

eat

the

10%

leve

lis

indi

cate

dby

*,5%

leve

lby

**,an

d1%

leve

lby

***.

(9)

MC

i,t

=a

+bx

i,t+P c∈C

θcc i

,tx

i,t+

ξi,

t(5

)M

Ci,

t=

a+

bxi,

t+P c∈C

δcc i

,t+

ξi,

t

Sam

ple

HSam

ple

ISam

ple

JSam

ple

KSam

ple

HSam

ple

ISam

ple

JSam

ple

K

Const

ant

0.1

89

***

0.2

99

***

0.2

75

***

0.2

77

***

0.1

20

***

0.2

82

***

0.2

43

***

0.2

15

***

(0.0

05)

(0.0

02)

(0.0

03)

(0.0

04)

(0.0

07)

(0.0

04)

(0.0

04)

(0.0

07)

IOB

4.2

86

***

2.8

08

***

2.7

13

***

1.9

89

***

4.9

94

***

2.6

90

***

2.9

45

***

3.5

71

***

(0.1

96)

(0.1

70)

(0.1

65)

(0.1

85)

(0.1

78)

(0.1

11)

(0.1

20)

(0.1

55)

PP

E*IO

B-0

.465

***

1.0

57

***

-0.0

65

-0.0

84

(0.0

58)

(0.1

14)

(0.0

68)

(0.0

52)

LSA

LE

S*IO

B0.4

24

***

-0.0

53

0.3

85

***

0.2

20

***

(0.0

62)

(0.0

70)

(0.0

80)

(0.0

68)

BT

M*IO

B-0

.590

***

-0.1

50

*-0

.531

***

0.1

68

**

(0.0

51)

(0.0

83)

(0.0

77)

(0.0

74)

DD

IV*IO

B0.9

48

***

0.8

52

***

1.2

01

***

0.8

16

***

(0.0

45)

(0.0

59)

(0.0

61)

(0.0

55)

CF*IO

B1.2

38

***

0.7

88

***

0.4

66

***

2.0

73

***

(0.2

20)

(0.1

69)

(0.1

68)

(0.2

04)

CA

SH

*IO

B1.6

54

***

0.4

83

***

0.5

16

***

0.4

54

***

(0.1

18)

(0.0

63)

(0.0

82)

(0.1

06)

PT

P*IO

B0.9

29

***

1.5

49

***

1.9

65

***

1.2

43

***

(0.0

45)

(0.0

64)

(0.0

99)

(0.0

45)

PP

E-0

.020

***

0.0

24

***

-0.0

02

-0.0

04

***

(0.0

02)

(0.0

02)

(0.0

02)

(0.0

01)

LSA

LE

S0.0

17

***

0.0

04

***

0.0

10

***

0.0

04

***

(0.0

02)

(0.0

01)

(0.0

01)

(0.0

02)

BT

M-0

.035

***

-0.0

13

***

-0.0

23

***

-0.0

06

**

(0.0

02)

(0.0

02)

(0.0

02)

(0.0

03)

DD

IV0.0

44

***

0.0

32

***

0.0

39

***

0.0

29

***

(0.0

01)

(0.0

01)

(0.0

01)

(0.0

02)

CF

0.0

92

***

0.0

29

***

0.0

39

***

0.0

80

***

(0.0

07)

(0.0

04)

(0.0

05)

(0.0

06)

CA

SH

0.0

46

***

0.0

14

***

0.0

15

***

0.0

16

***

(0.0

02)

(0.0

01)

(0.0

02)

(0.0

02)

PT

P0.0

38

***

0.0

44

***

0.0

48

***

0.0

38

***

(0.0

02)

(0.0

01)

(0.0

02)

(0.0

02)

No.

Obs.

13455

10610

10815

8985

13455

10610

10815

8985

49

Tab

le13

:T

ime

seri

esan

alys

isus

ing

tax

regi

me

shift

s.G

MM

esti

mat

ion

ofth

eco

effici

ents

ineq

uati

ons

5an

d9

for

all

indu

stri

esex

cept

utili

ties

,fin

ance

,an

dpu

blic

adm

inis

trat

ion

for

sam

ples

Ath

roug

hD

.T

heer

ror

func

tion

sar

ede

fined

acco

rdin

gto

equa

tion

s6

and

10w

here

y∗ i,

tis

the

‘equ

ilibr

ium

’m

argi

nal

bene

fit/c

ost

leve

l,x∗ i,

tis

the

obse

rved

or‘e

quili

briu

m’

inte

rest

expe

nses

over

book

valu

e(I

OB

)an

dC

isth

ese

tof

cost

cont

rol

vari

able

s.G

MM

mom

ents

are

obta

ined

byin

tera

ctin

gth

eer

ror

func

tion

wit

hth

efo

llow

ing

inst

rum

ents

:th

eco

nsta

ntte

rm,

the

vari

atio

nof

the

corp

orat

eta

xra

tes

from

elev

enta

xbr

acke

tsac

ross

tim

e,an

dea

chof

the

cont

rol

vari

able

s.T

hese

tof

cont

rol

vari

able

s,C

,in

clud

es{P

PE

,LS

AL

ES,B

TM

,DD

IV,C

F,C

AS

H,P

TP}

whe

reP

PE

ispl

ant,

prop

erty

,an

deq

uipm

ent

over

tota

lbo

okva

lues

,LSA

LE

Sis

log

ofne

tsa

les,

BT

Mis

book

equi

tyto

mar

ket

equi

ty,D

DIV

isan

indi

cato

rfo

rdi

vide

ndpa

ying

firm

,CF

isne

tca

shflo

wov

erto

talb

ook

valu

es,C

ASH

isca

shho

ldin

gsov

erto

talb

ook

valu

es,a

ndP

TP

isth

epe

rson

alta

xpe

nalty

asm

easu

red

inG

raha

m(1

999)

.T

heco

ntro

lvar

iabl

esar

est

anda

rdiz

edto

have

mea

nze

roan

dst

anda

rdde

viat

ion

one

base

don

Sam

ple

A(a

ndar

eno

tre

-sta

ndar

dize

dac

ross

sam

ples

).R

obus

tG

MM

stan

dard

erro

rsar

ere

port

edin

the

pare

nthe

ses.

Sign

ifica

nce

atth

e10

%le

velis

indi

cate

dby

*,5%

leve

lby

**,an

d1%

leve

lby

***.

(9)

MC

i,t

=a

+bx

i,t+P c∈C

θcc i

,tx

i,t+

ξi,

t(5

)M

Ci,

t=

a+

bxi,

t+P c∈C

δcc i

,t+

ξi,

t

Sam

ple

ASam

ple

BSam

ple

CSam

ple

DSam

ple

ASam

ple

BSam

ple

CSam

ple

D

Const

ant

0.1

93

***

0.2

50

***

0.1

90

***

0.2

42

***

0.1

60

***

0.1

70

***

0.1

72

***

0.1

72

***

(0.0

04)

(0.0

05)

(0.0

04)

(0.0

05)

(0.0

04)

(0.0

06)

(0.0

04)

(0.0

07)

IOB

2.8

02

***

1.8

44

***

2.6

29

***

2.1

14

***

3.2

20

***

3.7

38

***

2.6

67

***

3.5

28

***

(0.1

43)

(0.1

82)

(0.1

46)

(0.1

85)

(0.1

27)

(0.1

91)

(0.1

29)

(0.2

03)

PP

E*IO

B-0

.516

***

-0.4

15

***

-0.5

38

***

-0.5

16

***

(0.0

21)

(0.0

41)

(0.0

22)

(0.0

45)

LSA

LE

S*IO

B0.9

85

***

0.4

97

***

1.0

93

***

0.5

84

***

(0.0

29)

(0.0

39)

(0.0

30)

(0.0

42)

BT

M*IO

B-0

.438

***

-0.5

13

***

-0.4

41

***

-0.5

43

***

(0.0

17)

(0.0

29)

(0.0

19)

(0.0

33)

DD

IV*IO

B1.6

87

***

1.3

85

***

1.4

33

***

1.1

79

***

(0.0

28)

(0.0

32)

(0.0

28)

(0.0

32)

CF*IO

B2.0

88

***

3.4

03

***

1.9

69

***

3.0

25

***

(0.0

58)

(0.1

81)

(0.0

59)

(0.1

90)

CA

SH

*IO

B0.9

82

***

1.7

16

***

0.7

98

***

1.8

19

***

(0.0

71)

(0.1

02)

(0.0

72)

(0.1

09)

PT

P*IO

B1.0

53

***

1.2

05

***

0.9

76

***

1.0

64

***

(0.0

23)

(0.0

31)

(0.0

25)

(0.0

34)

PP

E-0

.016

***

-0.0

11

***

-0.0

17

***

-0.0

14

***

(0.0

01)

(0.0

01)

(0.0

01)

(0.0

01)

LSA

LE

S0.0

27

***

0.0

12

***

0.0

32

***

0.0

16

***

(0.0

01)

(0.0

01)

(0.0

01)

(0.0

01)

BT

M-0

.013

***

-0.0

22

***

-0.0

14

***

-0.0

25

***

(0.0

01)

(0.0

01)

(0.0

01)

(0.0

01)

DD

IV0.0

59

***

0.0

47

***

0.0

52

***

0.0

42

***

(0.0

01)

(0.0

01)

(0.0

01)

(0.0

01)

CF

0.0

75

***

0.1

05

***

0.0

71

***

0.0

96

***

(0.0

02)

(0.0

05)

(0.0

02)

(0.0

05)

CA

SH

0.0

32

***

0.0

34

***

0.0

26

***

0.0

32

***

(0.0

01)

(0.0

02)

(0.0

01)

(0.0

02)

PT

P0.0

39

***

0.0

41

***

0.0

40

***

0.0

40

***

(0.0

01)

(0.0

01)

(0.0

01)

(0.0

01)

No.

Obs.

69569

37978

50345

25977

69569

37978

50345

25977

50

Tab

le14

:Si

ngle

vari

ate

anal

ysis

.G

MM

esti

mat

ion

ofth

eco

effici

ents

ineq

uati

ons

5an

d9

usin

gon

eco

ntro

lat

ati

me

for

alli

ndus

trie

sex

cept

utili

ties

,fin

ance

,an

dpu

blic

adm

inis

trat

ion

for

sam

ple

D.

The

erro

rfu

ncti

ons

are

defin

edac

cord

ing

toeq

uati

ons

6an

d10

,w

here

y∗ i,

tis

the

‘equ

ilibr

ium

’m

argi

nalbe

nefit

/cos

tle

vel,

x∗ i,

tis

the

obse

rved

or‘e

quili

briu

m’in

tere

stex

pens

esov

erbo

okva

lue

(IO

B)

and

Cis

the

set

ofco

stco

ntro

lva

riab

les.

GM

Mm

omen

tsar

eob

tain

edby

inte

ract

ing

the

erro

rfu

ncti

onw

ith

the

follo

win

gin

stru

men

ts:

the

cons

tant

term

,the

vari

atio

nof

the

mar

gina

lben

efit

curv

eA

i,t,a

ndea

chof

the

cont

rolv

aria

bles

.T

hese

tof

cont

rolv

aria

bles

,C,i

nclu

des{P

PE

,LS

AL

ES,B

TM

,DD

IV,C

F,C

AS

H,P

TP}w

here

PP

Eis

plan

t,pr

oper

ty,an

deq

uipm

ent

over

tota

lbo

okva

lues

,LSA

LE

Sis

log

ofne

tsa

les,

BT

Mis

book

equi

tyto

mar

ket

equi

ty,SA

LE

Sis

net

sale

sov

erto

talbo

okva

lues

,D

DIV

isan

indi

cato

rfo

rdi

vide

ndpa

ying

firm

,C

Fis

net

cash

flow

over

tota

lbo

okva

lues

,C

ASH

isca

shho

ldin

gsov

erto

talbo

okva

lues

,and

PT

Pis

the

pers

onal

tax

pena

lty

asm

easu

red

inG

raha

m(1

999)

.T

heco

ntro

lvar

iabl

esar

est

anda

rdiz

edto

have

mea

nze

roan

dst

anda

rdde

viat

ion

one

base

don

Sam

ple

A(a

ndar

eno

tre

-sta

ndar

dize

dac

ross

sam

ples

).R

obus

tG

MM

stan

dard

erro

rsar

ere

port

edin

the

pare

nthe

ses.

Sign

ifica

nce

atth

e10

%le

velis

indi

cate

dby

*,5%

leve

lby

**,an

d1%

leve

lby

***.

(9)

MC

i,t

=a

+bx

i,t+P c∈C

θcc i

,tx

i,t+

ξi,

t(5

)M

Ci,

t=

a+

bxi,

t+P c∈C

δcc i

,t+

ξi,

t

No

PP

ELSA

LE

SB

TM

DD

IVC

FC

ASH

PT

PP

PE

LSA

LE

SB

TM

DD

IVC

FC

ASH

PT

PC

ontr

ols

Const

ant

0.0

81

***

0.0

78

***

0.0

85

***

0.1

24

***

0.1

05

***

0.1

24

***

0.0

93

***

0.0

80

***

0.0

70

***

0.0

82

***

0.1

20

***

0.0

99

***

0.0

64

***

0.0

39

***

0.0

80

***

(0.0

06)

(0.0

06)

(0.0

06)

(0.0

05)

(0.0

05)

(0.0

06)

(0.0

05)

(0.0

06)

(0.0

06)

(0.0

06)

(0.0

05)

(0.0

05)

(0.0

05)

(0.0

07)

(0.0

06)

IOB

8.1

84

***

8.2

20

***

7.8

09

***

7.0

40

***

7.2

73

***

5.9

75

***

8.6

23

***

8.2

13

***

8.4

19

***

7.8

63

***

6.8

54

***

7.2

68

***

6.6

26

***

9.8

68

***

8.2

25

***

(0.1

80)

(0.1

82)

(0.1

70)

(0.1

43)

(0.1

51)

(0.2

53)

(0.1

88)

(0.1

81)

(0.1

91)

(0.1

71)

(0.1

39)

(0.1

50)

(0.1

33)

(0.2

37)

(0.1

82)

PP

E*IO

B-0

.373

***

(0.0

57)

LSA

LE

S*IO

B0.8

62

***

(0.0

50)

BT

M*IO

B-1

.158

***

(0.0

41)

DD

IV*IO

B1.2

48

***

(0.0

37)

CF*IO

B2.8

65

***

(0.3

41)

CA

SH

*IO

B2.7

38

***

(0.1

16)

PT

P*IO

B0.1

73

***

(0.0

48)

PP

E-0

.021

***

(0.0

02)

LSA

LE

S0.0

29

***

(0.0

02)

BT

M-0

.070

***

(0.0

02)

DD

IV0.0

49

***

(0.0

01)

CF

0.1

83

***

(0.0

07)

CA

SH

0.1

05

***

(0.0

03)

PT

P0.0

01

(0.0

02)

No.

Obs.

29674

29670

29670

28184

29671

29645

29671

27354

29670

29670

28184

29671

29645

29671

27354

51

The Cost of Debt - Wharton...

Documents

Transcript of The Cost of Debt - Wharton...