SSRN Id4451752

Why Do Powerful Firms Hold Excessive Cash?
Dan Su†
CKGSB
This Version: May 17, 2023
Abstract
This paper investigates the cross-sectional relationship between corporate mar-

ket power and cash holdings. To begin with, we empirically show that firms with
higher markup tend to have larger cash holdings on their balance sheets. Then we
rationalize this empirical finding in a model with risky markup and costly external fi-
nancing. Changes in consumers’ taste for quality and producers’ marginal production
cost increase the earnings-quality gradient sharply in the right tail, which generates a
“winners-take-most” phenomenon but also makes current winners inherently riskier.
In the presence of external financing costs, firms with risky market power are inclined
to rely more on internal financing. Finally, when taken to the data, our model is able
to quantitatively match both the upward trend and the cross-sectional variation in
corporate cash.
JEL codes: D42; G31; G32; L12; L20
Keywords: market power; corporate cash; external financing cost; superstar firms
* For helpful comments and suggestions, I thank Lauren Cohen, Alexandre Corhay, Winston Wei Dou,
Howard Kung, and Erica Xuenan Li. Of course, all errors are mine own.
† Affiliation: Department of Finance, Cheung Kong Graduate School of Business. Oriental Plaza, 1st East
Chang’An Street, Beijing, 100738, China. Email address: dansu@ckgsb.edu.cn
Electronic copy available at: https://ssrn.com/abstract=4451752

M ARKET P OWER AND C ORPORATE C ASH
1. Introduction
Over the past several decades, we have observed two important macro-finance trends
in the data. First, corporate market power, measured as markup, has increased steadily
over time (De Loecker et al., 2020). Second, companies rely more and more on internal
financing by holding excessive cash on their balance sheets (Bates et al., 2009). As shown
in Figure 1, compared to 1980, now we are more likely to observe firms with excessive
market power and cash holdings. In this paper, we argue that these two phenomena are
deeply connected in both the time series and cross section, and our goal is to provide a
theoretical and quantitative framework to explain them jointly.
Figure 1: The Distributional Changes in Cash Holdings and Markups

10.0
1980 1980
2015 1.5 2015
7.5
1.0
Density
Density
5.0
0.5
2.5
0.0 0.0
0.00 0.25 0.50 0.75 1.00 0 1 2 3 4
Cash/Asset Markup
Notes: This figure presents the distributional changes in corporate cash holdings and markup between 1980
and 2015. The firm-level cash-to-asset ratio is simply calculated as the ratio of cash and cash equivalents
(Compustat series CHE) to total assets (Compustat series AT). Meanwhile, markup is estimated via the pro-
duction function approach proposed by De Loecker and Warzynski (2012). Data is obtained from Compustat.
To begin with, we empirically show that firms with higher markup are likely to hold
more cash on their balance sheets. Using a dataset containing U.S. public firms, our
main findings are twofold. First, although firms with different markup levels have all ob-
served a substantial increase in cash holdings, the magnitudes are quite different. More
specifically, the group of firms with the highest markup levels has experienced the most
rapid increase (or decrease) in cash holding (or net debt) since the 1970s. Second, cross-
sectionally, we find that powerful firms, i.e., the firms with higher markup levels, are

likely to hold more cash and less net debt on their balance sheets. There exists a statisti-
cally and economically significant relationship between firm-level cash-to-asset ratio (or
net-debt-to-asset ratio) and markup. However, this relationship is not always monotone
cross-sectionally. In some model specifications, we do find some degree of nonlinearity
and a U-shaped cash-markup relation: both left-tail and right-tail companies hold ex-
cessive cash, compared to the “representative” firms with a median markup level. Our
empirical findings suggest there could be two different opposing forces determining the
optimal cash-to-asset ratios.
In order to rationalize why powerful firms hold excessive cash, we provide a model
with risky market power, defaultable debt, and costly external financing. We argue that
the primitive drivers behind these macro-finance trends are some economic fundamen-
tal changes from both the demand and supply sides. The intuition is that nowadays, on
the demand side, consumers care more about product quality than quantity. In other
words, they are willing to pay more for high-quality products. On the supply side, some
new technology (e.g., digitalization) makes firms less costly to serve all the potential cus-
tomers, which leads to an increased returns-to-scale. These two primitive shocks can
directly impact the level and volatility of corporate earnings. As shown in Figure 2, with
these two fundamental changes, firms with the best product quality can obtain enormous
earnings and dominate the activities in which they engage. This static income level redis-
tribution effect has been widely recognized in the existing literature (e.g., Sattinger, 1993;
Tervio, 2008; Scheuer and Werning, 2017). However, when firms face uncertainty in their
product quality, from a dynamic perspective, superstars are inherently riskier. Compared
to the low-concentration traditional economy, in a superstar economy, a small variation
in product quality can translate into considerable earnings fluctuations, especially on the
right-tail side. In other words, these economic fundamental changes are also redistribu-
tive in risk. As shown in Section 3, this feature shows up in the model because corporate
earnings become a convex function of the underlying quality in the superstar economy.
With convexity, both income level and risk are redistributed towards right-tail firms.
In this way, these economic fundamental changes can substantially and heteroge-

Figure 2: Risky Superstar Economy
Traditional Economy
Superstar Economy
Earnings
Quality
neously affect corporate risk management policy. With external financing costs and pre-
cautionary saving incentives, an increased future earnings uncertainty makes firms op-
timally choose to accumulate more cash and rely more on internal financing. Cross-
sectionally, such an incentive is stronger for right-tail firms with more volatile earnings
processes. Our risky market power channel is different from the conventional default
risk channel, where the defaultable bond price is increasing in market power. To provide
additional support for our key model mechanism, we also implement the reduced-form
empirical investigation and show superstar firms are indeed riskier as they face more
volatile fluctuations in earnings.
Finally, we investigate the model’s quantitative implications. We estimate the struc-
tural parameters of the model through the simulated method of moments (SMM) ap-
proach. When taken to data, our model is able to quantitatively match both the aggregate
trend and cross-sectional variation in the data. In addition, we implement several coun-
terfactual exercises to evaluate the relative importance of supply versus demand origin,
as well as that of the precautionary saving versus bond price channel. Our conclusions
from these exercises are twofold. First, the most important driver behind these trends in
corporate cash hoarding behaviors is the change in production-technology-related supply

factors, although demand factors also play a non-negligible role. Second, the precaution-
ary saving channel is more important in explaining the behaviors of right-tail firms while
the bond price is more important in explaining the left-tail. In other words, there indeed
exist two different mechanisms for why companies hold cash, and the relative importance
of each channel is also different for firms with different markup levels. More importantly,
the risky market power story in our paper is important for our understanding of why
powerful firms hold excessive cash in the data.
Literature Review Our study is related to three different branches of research. First,
our paper speaks to the corporate cash literature. Bates et al. (2009) empirically show
that U.S. companies have changed their risk management policies substantially over the
past several decades and accumulated a massive amount of cash. This secular increase
in corporate cash holdings is quite puzzling, and a large literature in corporate finance
has attempted to answer this question. For instance, Nikolov and Whited (2014) point out
that the manager-shareholder conflicts are the main reason behind. They use a dynamic
structural model to explore the links between corporate cash holdings and three different
agency issues. In contrast, Azar et al. (2016) argue that the reduction in interest rate is the
main driver behind because interest rates are positively associated with the cost of holding
cash. Other possible explanations include riskier cash flows (e.g., Opler et al., 1999; Bates
et al., 2009), firm-level uncertainty (Im et al., 2017), foreign tax differentials (Gu, 2017),
changing industry composition (Begenau and Palazzo, Forthcoming) and so on. Although
our story on the origins of increasing corporate cash holdings is closely related to risky
technology or uncertainty stories, our key contribution here is to investigate the cross-
sectional link between market power and corporate cash.
Second, our research is closely related to the growing literature on superstar firms.
Rosen (1981) is the first to bring our attention to the economics of superstars. He points out
that technical change would allow the most talented entrepreneur to serve a larger group
of people and dominate this economy. Many following works attempt to use this idea
to explain why earnings distribution is more right-skewed compared to the underlying

talent distribution (e.g., Gabaix and Landier, 2008; Tervio, 2008). Recent studies can be
classified into three categories. In the first category, people are interested in investigating
the origins of superstars. Possible explanations include asset prices (Gomez, 2019), low
risk-free rate (Liu et al., 2019; Kroen et al., 2021), random growth (Luttmer, 2012), and so
on. The second group of works focuses on the macro-finance implications of superstar
firms. For instance, Autor et al. (2020) and Kehrig and Vincent (2020) discuss how these
superstar firms contribute to the decline of labor share; De Ridder (2019) shows that the
increasing market power of large firms discourages innovation and leads to the decline
in business dynamism. The third category focuses on how different are today’s firms
compared to their counterparts several decades ago. For example, Hoberg and Phillips
(2021) find that U.S. firms have expanded their scope and scale of operations in the past
several decades, and this scope expansion significantly increases corporate valuation.
Third, our paper also relates to the capital structure and product market competition
literature. The theoretical works on this topic can date back to Brander and Lewis (1986)
and Maksimovic (1988), which point out the role of capital structure in committing to cer-
tain product market strategies. In terms of empirical studies, MacKay and Phillips (2005)
show that leverage is higher for industries with higher degrees of concentration. Gao
(2021) finds that the input-output production network also affects a firm’s optimal choice
of internal financing. Recently, Jung et al. (2021) provide both a theory and some empirical
evidence to show that different types of competition can have contrasting implications on
optimal leverage. In addition, Dou and Ji (Forthcoming) use a monopolistic competition
framework with customer capital to provide an interesting interpretation of the financial
origin of markup.
Layout The rest of the paper is organized as follows. Section 2 documents the empirical
link between corporate market power and cash holdings. Section 3 investigates the origins
of risky market power and its implications on corporate cash-holding behaviors. Section
4 takes the model to the data and studies our model’s quantitative performance. Finally,
Section 5 concludes.

2. Empirics
This section provides several stylized facts on the cross-sectional relationship between
corporate market power and cash holdings. As mentioned before, we focus on the distri-
butional changes and cross-sectional link as the existing literature à la Bates et al. (2009)
and De Loecker et al. (2020) has separately studied the secular time-series trends on these
two topics.
2.1 Data and Variable
Firm-level balance sheet information is obtained from the Compustat North America Fun-
damentals datafile with a sample period from 1970 to 2015. Following the standard prac-
tice, we exclude financial firms (SIC codes 6000-6999) and regulated utilities (SIC codes
4900-4999). In addition, we keep all the entries with a foreign incorporation code of
“USA” and those traded on the three major exchanges: NYSE, AMEX, and NASDAQ.
For non-ratio indicators, we use the gross output price indices to deflate firms’ output-
related variables (e.g., sales) and adopt the industry-level capital price indices to deflate
firms’ investment-related variables. The historical data on these price indices are obtained
from Integrated Industry-Level Production Account (KLEMS).
Our two key variables – cash holdings and markup – are measured as follows. Fol-
lowing the existing literature, the firm-level cash-to-asset ratio is simply calculated as the
ratio of cash and cash equivalents (Compustat series CHE) to total assets (Compustat se-
ries AT). To obtain an estimate on the firm-level markup, we adopt the recent production
approach proposed by De Loecker et al. (2020) and De Loecker and Warzynski (2012).
The idea is mainly based on cost minimization decisions (e.g., Hall, 1988), and corporate
markup is measured as the product of output elasticity and revenue share of any certain
variable input. The production approach is advantageous to the previous accounting and
demand approach as it does not need to impose any strong assumptions on cost structures
or demand systems.
For robustness checks, we use Tobin’s Q or Peters and Taylor (2017)’s total Q as the

alternative proxies for corporate market power, although these two are less suitable mea-
sures. Following Kaplan and Zingales (1997) and references thereafter, we measure To-
bin’s Q as asset market value divided by its book value. The market value of assets is com-
puted as the book value of assets (Compustat series AT) plus the market value of common
stocks (Compustat series PRCC_C times Compustat series CSHO) minus the book value
of equity, where the book value of equity is estimated as the sum of shareholder equity
(Compustat series SEQ), deferred taxes (Compustat series TXDB), and investment tax
credit (Compustat series ITCB), minus the value of preferred stocks (coalesce outcomes
of Compustat series PSTKRV, PSTKL, and PSTK). In contrast, total Q is measured as the
ratio between a firm’s asset market value over the replacement cost of its intangible cap-
ital. Following Peters and Taylor (2017), we estimate the intangible capital replacement
cost by accumulating past investments in Research and Development (Compustat series
XRD) and Selling, General, and Administrative Expenses (Compustat series XSGA).
In addition, in order to alleviate the concern that some firms might borrow to accumu-
late cash due to the tax advantages of debt, we also use net debt as an alternative measure
for corporate internal financing behaviors. Net debt is simply calculated as the difference
between the debt-to-asset ratio and cash-to-asset ratio, where debt is the sum of long-term
debt (Compustat series DLTT) and current liabilities (Compustat series DLC).
Other variables are constructed by directly following the standard empirical corporate
finance literature. Size is the natural logarithm of total assets (Compustat series AT); re-
turn of the asset is calculated as the income before extraordinary items (Compustat series
IB) scaled by total assets; asset tangibility is measured as the ratio of physical assets (Com-
pustat series PPENT) to total assets; book leverage is computed as the ratio of total debts
DLTT+DLC
to the sum of total debts and common equity, i.e., book leverage = DLTT+DLC+CEQ , where
DLTT, DLC, and CEQ represent the Compustat data series on long-term debt, current li-
abilities, and common equity, respectively; dividend payout is measured as dividends
(Compustat series DVC) scaled by total assets; physical capital investment rate is com-
puted as the share of capital expenditures (Compustat series CAPX) in total assets; and
finally, customer capital investment rate is calculated as the relative ratio of the sum of

selling, general and administrative expenditures (Compustat series XSGA) to total assets.
2.2 Distributional Changes in the Time-Series
We start by showing that market power is an important reason behind the secular rise
of corporate cash holdings. In Graph (a) of Figure 3, we plot the time series of cash-to-
asset ratios for firms with different levels of markup. The figure is constructed with the
following two steps. To begin with, in each year, we classify all the firms with available
markup measures into five different quantiles according to their markup level. After that,
we calculate the annual median cash-to-asset ratio for each quantile. As shown in this
figure, all five groups have observed a substantial increase in cash holdings (at least until
2010), which indicates that this secular rise documented in Bates et al. (2009) is not a
unique phenomenon to any specific group of firms. However, the magnitudes are quite
different among firms with different markup levels. According to Graph (a), right-tail
firms, i.e., the group of firms with the highest markup levels, have experienced the most
rapid increase in cash holdings since the 1970s. The median cash-to-asset ratio for right-
tail firms was only 0.054 in 1970, but it increased to 0.217 in 2015. The peak happens in
2005 with a value of 0.242. In contrast, the long-term trend is modest for left-tail firms,
i.e., the group of companies with the lowest markups. The median cash-to-asset ratio for
left-tail firms has changed from 0.064 in 1970 to 0.069 in 2015 with a roof value of 0.096
in 2010. Therefore, our exercise here indicates the long-term increase in corporate cash is
mainly driven by powerful firms.
Besides, our conclusion remains valid even if we use the net debt measure. As ex-
plained in the previous section, the net-debt-to-asset ratio is measured as the difference
between the debt-to-asset ratio and the cash-to-asset ratio. The corresponding time-series
plot for companies with different levels of markup is presented in Graph (b) of Figure 3.
The figure shows that the decline in net debt is the sharpest for firms with the highest
markup levels. The median net-to-debt ratio for right-tail firms was 0.190 in 1970, but
now this number has become -0.047 in 2015 with a bottom value of -0.130 in 2011. In con-
trast, the net debt for the left-tail firms has changed slightly from 0.223 to 0.219 during

this sample period.

Generally speaking, these two graphs in Figure 3 show that a non-negligible fraction of
increasing corporate cash hoardings or declining net debt comes from powerful firms. In
Figures A1 and A2 in the appendix, we redo our exercise here but use Tobin’s Q or total
Q as our alternative measures for market power. According to these figures, our main
conclusion still holds for these robustness checks. Therefore, in order to fully understand
why companies hold excessive cash, we also need to explain why powerful firms rely
more on internal financing.
Figure 3: Changes in Median Corporate Cash/Net Debt at Different Markup Quintiles
(a) Cash (b) Net Debt
Notes: This figure presents the time-series plot of corporate cash and net debt for firms with different levels
of markup. The firm-level cash-to-asset ratio is simply calculated as the ratio of cash and cash equivalents
(Compustat series CHE) to total assets (Compustat series AT). Meanwhile, net debt is calculated as the dif-
ference between the debt-to-asset ratio and the cash-to-asset ratio, where debt is the sum of long-term debt
(Compustat series DLTT) and current liabilities (Compustat series DLC). Markup is estimated via the produc-
tion function approach proposed by De Loecker and Warzynski (2012). Data is obtained from Compustat.
2.3 Cross-Sectional Relationship
One possible drawback of our previous time-series analysis is that the group composition
has changed substantially over the past decades. Therefore, in this section, we turn to
investigate the cross-sectional relationship between firm-level markup and cash holdings.

2.3.1 Initial evidence with Binscatter plots
Before conducting the formal regression analysis, we provide some initial evidence by
exploiting the advantage of Binscatter plots to help clarify the possibly nonlinear rela-
tionship between firm-level cash holdings and markup. The blue dashed line in Figure 4
presents the estimated relation between markup and cash-to-asset ratio without adding
any other control variables. According to this figure, on average, we find a U-shaped
relationship between these two variables. A firm’s cash-to-asset ratio is increasing in its
markup, provided that its markup is not too low. In other words, powerful firms do
have stronger incentives to hold more cash on their balance sheets, compared to the me-
dian “representative” ones. At the same time, this relationship is not always monotone.
Among the firms with the lowest markup, their cash holdings are decreasing in market
power. Figure 4 suggests there could be two different opposing forces determining the
equilibrium corporate cash-to-asset ratios. This pattern for the left-tail firms is consistent
with the financial frictions story, which argues that firms accumulate cash because they
have a higher probability of default or are more likely to face market frictions such as bor-
rowing constraints. As a result, there exists a negative relationship between a firm’s cash
holding and its financial wealth. However, this force seems less convincing when applied
to firms in the right-tail distribution.
At the same time, the orange dashed line in Figure 4 presents the estimated relation-
ship between markup and net debt. It shows that powerful firms hold substantially less
net debt. In other words, the right-tail firms do not simply borrow to accumulate cash
because debt has tax advantages. In addition, different from the U-shaped relationship
between cash and markup, the estimated link between net debt and markup is monoton-
ically decreasing for most cases.
Finally, we also use alternative markup measures such as Tobin’s Q or Peters and
Taylor (2017)’s total Q, and the corresponding results are reported in Graph (a) and (b) in
Figure A3, respectively. The key message from these exercises is that our main findings
in Figure 4 are robust to the alternative choices of measuring corporate market power.
The only difference is that with these two measures, there exists an inverted-U-shaped
10

Figure 4: Binscatter Plot for Markup, Cash, and Net Debt
Notes: This figure presents the Binscatter plot for corporate markup and cash-to-asset ratio or net-debt-to-
asset ratio. The firm-level cash-to-asset ratio is simply calculated as the ratio of cash and cash equivalents
(Compustat series DLTT) and current liabilities (Compustat series DLC). Markup is estimated via the produc-
tion function approach proposed by De Loecker et al. (2020). Data is obtained from Compustat.
relationship between market power and net debt, which indicates that both the left-tail
and right-tail companies hold substantially less net debt.
2.3.2 Regression analysis
Now we turn to regression analysis to investigate whether the previous conclusion still
holds after controlling for other possible determinants of corporate cash. Table 1 shows
the regression results for investigating the relationship between markup and cash hold-
ings. We adopt two-way fixed-effects regressions to estimate the impacts of markup on
corporate cash-holding behaviors. The general model specification in Table 1 is shown as
follows:
cashi,t = α + β × markupi,t + γ × markup2i,t + ΓXi,t + δi + ηt + ε it
11

Throughout this section, i and t refer to firm and year, respectively. cash is the firm’s
cash-to-asset ratio, while markup represents our empirical measure for firm-level markup.
We are primarily interested in the sign and statistical significance of the estimated coeffi-
cient β. However, as observed in the previous Binscatter plots, there might be some non-
linearity in the relationship between markup and cash holdings. Hence, in some model
specifications, we also add the square markup term markup2 as an additional independent
variable. X represents a group of firm-level control variables that could affect corporate
cash holdings. Following the standard practice in the existing literature, we include some
other firm-level indicators such as the return of assets, tangibility, investment, size, prof-
itability, R&D, book leverage, and dividend payout. In addition, we introduce both firm
and year-fixed effects to account for the unobserved firm and year characteristics. All
standard errors are clustered at the firm level.
Columns (1) - (5) in Table 1 present our baseline results on the relationship between
corporate cash and markup using various fixed-effect regression models with differences
in the use of control variables or fixed effects in each column. We treat the R&D variable
differently as it has a relatively smaller data availability in the Compustat dataset. For
the first three columns, we control for the firm- and year-fixed effects. In Column (4), we
include the 3-digit-NAICS-industry- and year-fixed effects; Meanwhile, in Column (5),
we introduce the industry, year, and industry-year fixed effects.
Based on the results shown in the first five columns in Table 1, we find that in all model
specifications, the estimated coefficients of markup enter with a positive sign at the 1% sig-
nificance level. This result suggests that firms with higher markup are indeed associated
with higher cash-to-asset ratios. The economic significance is also quite considerable. Our
estimated coefficient varies from 0.0402 to 0.0753, which indicates that a one-standard-
deviation difference in market power is associated with a 0.37–0.69 standard-deviation
difference in corporate cash-to-asset ratios. In addition, similar to what we have observed
in the Binscatter plot, with some model specifications, we are able to find some degree of
nonlinearity in the cash-markup relationship. Column (5) shows that the estimated coef-
ficient of markup2 term is also positive at the 10% significance level, even after controlling
12

for a set of other variables.

The findings are similar if we turn to investigate the relation between corporate market
power and net debt. Columns (6) - (10) in Table 1 present similar fixed-effect regression
results but this time we use net debt as our dependent variable. Similarly, we find that
across various model specifications, the estimated coefficients of markup enter with a neg-
ative sign at the 1% significance level, which indicates that firms with higher markup tend
to have lower net-debt-to-asset ratios. As for the economic significance, a one-standard-
deviation difference in market power is associated with a 0.29–1.15 standard-deviation
difference in corporate net-debt-to-asset ratios.
In Tables A1 and A2 in the appendix, we provide the corresponding regression results
for using Tobin Q or total Q as alternative proxies for market power. Similarly, these
two tables show that across different model specifications, we can always find a positive
(negative) and significant association between a firm’s cash-to-asset ratio (net-debt-to-
asset ratio) and its Q value. This conclusion does not depend on which Q-measure we
use.
2.4 Alternative Stories
Multinational Corporations Many studies have shown that multinational corporations

in the U.S. hold significantly more cash than domestic companies. In order to alleviate
the concern that our previously documented pattern is mainly driven by multinationality
instead of markup, we implement the same exercises for both multinational and domestic
firms. Following the existing literature (e.g., Foley et al., 2007; Pinkowitz et al., 2014; Gu,
2017), we define a firm as a multinational one if its reported value for foreign income and
foreign sales are positive and non-missing. Otherwise, we treat the firm as a domestic
one.
The similar Binscatter plots for the multinational-firms subsample and domestic-firms
subsample are presented in Graph (a) and (b) of Figure 5, respectively. As we can see from
these two pictures, our previous conclusions on the cash-markup and net debt-markup
relationships remain unchanged in both subsamples. As a matter of fact, the estimated
13

Table 1: Reduced-form Evidence for Markup, Cash, and Net Debt
Cash/Asset Net Debt/Asset
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Markup 0.0487*** 0.0700*** 0.0753*** 0.0402*** 0.0588*** -0.1045*** -0.1612*** -0.4191*** -0.2927*** -0.3255***
(28.703) (29.157) (5.449) (3.140) (4.624) (-16.818) (-17.604) (-7.959) (-6.750) (-7.435)
Markup2 -0.0015 0.0132*** 0.0074* 0.0755*** 0.0441*** 0.0520***
(-0.386) (3.476) (1.946) (4.973) (3.418) (3.997)
Return of Asset -0.0018*** 0.0005* 0.0005* 0.0004 0.0005 0.0145*** 0.0186*** 0.0186*** -0.0006 -0.0009
(-7.493) (1.874) (1.877) (0.927) (1.384) (16.457) (17.394) (17.365) (-0.414) (-0.660)
Profitability -0.0030*** -0.0028*** -0.0028*** -0.0006 0.0000 -0.0284*** -0.0129*** -0.0128*** -0.0033** -0.0079***
(-8.406) (-6.365) (-6.368) (-1.466) (0.088) (-21.848) (-7.654) (-7.613) (-2.277) (-4.683)
Investment 0.0769*** 0.0679*** 0.0678*** 0.1744*** 0.1799*** -0.1724*** -0.3229*** -0.3219*** -0.4959*** -0.5113***
(15.530) (7.993) (7.991) (16.902) (17.652) (-9.497) (-9.975) (-9.947) (-14.174) (-14.579)
Tangibility -0.3693*** -0.5211*** -0.5213*** -0.4783*** -0.4723*** 0.6619*** 0.8680*** 0.8754*** 0.9034*** 0.9193***
(-106.426) (-95.906) (-95.685) (-98.634) (-96.918) (52.058) (41.895) (42.151) (54.957) (54.814)
Book Leverage -0.0020*** -0.0004*** -0.0004*** -0.0019*** -0.0016*** 0.0679*** 0.0525*** 0.0525*** 0.0617*** 0.0606***
(-15.136) (-3.394) (-3.385) (-12.151) (-10.209) (140.589) (104.158) (104.053) (114.782) (111.110)
Payout 0.0078*** 0.0068** 0.0068** 0.0104*** 0.0095** -0.0332*** -0.0881*** -0.0882*** 0.1481*** 0.1499***
(3.188) (2.223) (2.224) (2.686) (2.492) (-3.712) (-7.504) (-7.516) (11.281) (11.472)
Size -0.0099*** -0.0025*** -0.0025*** -0.0050*** -0.0044*** -0.0634*** -0.0842*** -0.0841*** -0.0279*** -0.0288***
(-20.545) (-3.791) (-3.794) (-16.445) (-14.407) (-35.827) (-32.908) (-32.880) (-26.767) (-27.315)
R&D -0.0240*** -0.0240*** 0.0495*** 0.0449*** 0.2075*** 0.2071*** 0.1083*** 0.1032***
(-9.627) (-9.623) (21.605) (18.914) (21.826) (21.788) (13.948) (12.627)
Fixed effects
Year Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Firm Yes Yes Yes Yes Yes Yes
SIC3 Yes Yes Yes Yes
SIC3 × Year Yes Yes
N 154,330 79,225 79,225 80,923 80,923 154,330 79,225 79,225 80,923 80,923
Adjusted R2 0.641 0.727 0.727 0.358 0.385 0.573 0.596 0.596 0.274 0.282
Notes: This table presents the association between markup and corporate cash or net debt with different fixed-effect model
specifications. The dependent variables are corporate markup, and we measure it by following De Loecker et al. (2020)’s
method. The firm-level cash-to-asset ratio is simply calculated as the ratio of cash and cash equivalents (Compustat series
CHE) to total assets (Compustat series AT). Meanwhile, net debt is calculated as the difference between the debt-to-asset
ratio and the cash-to-asset ratio, where debt is the sum of long-term debt (Compustat series DLTT) and current liabilities
(Compustat series DLC). Size is the natural logarithm of total assets (Compustat series AT); return of the asset is calculated
as the income before extraordinary items (Compustat series IB) scaled by total assets; asset tangibility is measured as the
ratio of physical assets (Compustat series PPENT) to total assets; book leverage is computed as the ratio of total debts to
DLTT+DLC
the sum of total debts and common equity, i.e., book leverage = DLTT +DLC+CEQ , where DLTT, DLC, and CEQ represent
the Compustat data series on long-term debt, current liabilities, and common equity, respectively; dividend payout is mea-
sured as dividends (Compustat series DVC) scaled by total assets; and finally, the corporate investment rate is computed
as the share of capital expenditures (Compustat series CAPX) in total assets. Data used in this table is at the firm-year
frequency and obtained from Compustat. T-statistics are in parentheses. *, **, and *** represent results significant at the
10%, 5%, and 1% levels, respectively. Standard errors are clustered at the firm level.
14

correlations are stronger in the domestic firms subsample. In addition, the formal two-
way fixed-effect regression results are shown in Table A3 and A4 in the appendix. These
two tables show that our previously documented strong empirical link between markup
and cash/net debt is not unique to multinational corporations.
R&D-intensive Firms Begenau and Palazzo (Forthcoming) empirically show that only
R&D-intensive firms have increased their cash holdings over the past several decades.
In order to argue that our previous findings are not entirely driven by R&D expenses,
we conduct some similar exercises for both R&D-intensive and R&D-non-intensive firms.
We follow the work of Begenau and Palazzo (Forthcoming) and define an R&D-intensive
firm as one belonging to an industry whose average R&D investment-to-asset ratio is
equal to or above 2%. We classify all the industries at 3-digit SIC level. The corresponding
Binscatter plots for the R&D-intensive and R&D-non-intensive subsamples are provided
in Graph (c) and (d) of Figure 5, respectively. As we can see from these two pictures,
our previous conclusions on the cash-markup and net debt-markup relationships are still
there in both subsamples. However, it is true that the estimated correlations are stronger
in the R&D-intensive subsample, which is consistent with Begenau and Palazzo (Forth-
coming)’s main findings and also the fact that innovation is likely to be the source of
market power. Similarly, we prepare the formal regression results in Table A9 and A10
in the appendix. These two tables show that our previously documented significant asso-
ciation between markup and cash/net debt exists within both R&D-intensive and R&D-
non-intensive firms.
Besides, for these two alternative stories, we have also adopted Tobin’s Q or Peters
and Taylor (2017)’s total Q as alternative measures for market power. The corresponding
Binscatter plots are provided in Figures A4 and A5 in the appendix. Meanwhile, the
regression results are shown in Tables A3–A14. Again, these results show that our main
findings do not rely on any specific choices of market power measures.
15

Figure 5: Binscatter Plot for Alternative Stories
(a) Multinational (b) Domestic
(c) R&D intensive (d) R&D non-intensive
Notes: This figure presents the Binscatter plot for corporate markup and cash-to-asset ratio or net-debt-to-
(Compustat series DLTT) and current liabilities (Compustat series DLC). Markup is estimated via the pro-
duction function approach proposed by De Loecker et al. (2020). We define a firm as a multinational one if
its reported value for foreign income and foreign sales are positive and non-missing. Otherwise, we treat the
firm as a domestic one. We define an R&D-intensive firm as one belonging to an industry whose average
R&D investment-to-asset ratio is equal to or above 2%. Otherwise, we label the firm as R&D-non-intensive.
All the industries are classified at the 3-digit SIC level. Data is obtained from Compustat.
3. Model
16

3.1 Setup
Now we rationalize our previous empirical findings by using a partial-equilibrium, dy-

namic model. Generally speaking, this infinite-horizon discrete-time economy consists
of a unit-measure continuum of companies. During each period, the risk-neutral man-
agers of these companies aim to maximize the expected value of all future dividends to
shareholders. They achieve this by selecting the optimal quantities of sales, investment in
customer capital, cash, and debt, subject to some constraints. In this way, we can explore
the interconnected patterns of corporate investment, cash accumulation, and external fi-
nancing.
Corporate earnings We start by describing corporate earning profiles. A firm with an

idiosyncratic product quality z wants to sell its product. We assume that idiosyncratic
quality is i.i.d. across firms and follows any ergodic process. In this section, we don’t
impose any functional forms as we focus on exploring the theoretical implications. We
discuss the quantitative performance of our model in Section 4.
The price function p (z, y) is the unit price charged by a seller of quality z selling y
units. Competition in the market for services implies that the pricing function is taken as
given by any individual company. We impose two assumptions on this exogenous price
∂p ∂p
function: ∂z > 0 and ∂y < 0. These two conditions are quite intuitive and consistent
with reality: the equilibrium price p is increasing in product quality z while decreasing in
product quantity y.
In addition to the market size and product quality effects from the demand side, the
other important element affecting corporate decisions is the direct cost of production.
We assume that C (y) denotes the out-of-pocket costs of producing y units, which have a
fixed cost component and a variable operating cost component. Generally, we assume that
C ′ ≥ 0 and C ′′ ≥ 0, which indicates that there are non-decreasing costs of production, i.e.,
internal diseconomies. As we will see later, the shape of price function p (z, y) and that
of cost function C (y) is essential to corporate risk profiles and hence the corresponding
optimal internal financing reliance.
17

At the same time, following the existing literature on customer capital and the impor-
tance of customer base (e.g., Dou and Ji, Forthcoming; Dou et al., Forthcoming; Morlacco
and Zeke, 2021), we assume that the final sales are limited by the company’s existing cus-
tomer base ω. Therefore, the equilibrium quantity of sales y∗ should be the minimum
of its existing customer base level ω and the optimal level ỹ, i.e., y∗ = min{ỹ, ω }. The
optimal level of output is chosen to maximize net revenue, i.e.,
π (z) ≡ max p (z, y) y − C (y) (1)

y≤ω
Therefore, ỹ (z) is chosen to satisfy:
∂p
ỹ |y=ỹ + p (z, ỹ) − C ′ (ỹ) = 0 (2)
∂y
∂2 p
as long as 2 ∂y |y=ỹ + y ∂y2 |y=ỹ − C ′′ (ỹ) < 0.
∂p
Customer capital investment The firm augments its customer base by investment ι,
which is given by
ι = ω ′ − (1 − δ ) ω (3)
where ω ′ represents the next period’s customer base level and δ denotes the depreciation
rate of the existing customer base. As mentioned before, the final sale should be less
or equal to the firm’s existing customer capital base, i.e., y∗ ≤ ω. Following Hayashi
(1982) and references thereafter, we assume that when converting final goods into new
customer capital, firms need to pay additional adjustment costs. These additional costs
can be specified by Ψ ω, ωι , which is a function of both the existing customer base level

ι
ω and the investment rate ω.
External financing At each period, any certain company has access to four distinct fund-
ing sources to support its expenses: current-period profits, internal cash reserves, issuance
of risky debt, and equity issuance with additional financing costs. For simplicity, we as-
18

sume that the cash balance, represented as c, does not generate any interest. Therefore,
the cost of holding cash is equivalent to the exogenously determined risk-free interest rate
denoted as r.
Furthermore, the company can utilize credit markets to obtain financing, whereby it
can raise funds through one-period risky debt denoted as b. This debt must be repaid or
renewed at the end of each period as it reaches maturity. The price of debt, represented
as q(z, ω ′ , c′ , b′ ), is determined endogenously and depends on the exogenous shock z, as
well as the firm’s endogenous choice variables for the next period, namely, customer base
ω ′ , cash c′ and debt b′ .
At the start of each period, following the observation of the product quality shock
∆z, the manager faces a decision regarding whether to default on the firm’s outstanding
debt obligations denoted as b. In the event of default, the firm’s assets are liquidated,
and the internal resources are allocated to creditors, while any remaining unpaid debt is
discharged. On the other hand, if the manager chooses not to default, the firm proceeds
with its production activities. Upon receiving revenue for the current period, the firm
determines the optimal levels of debt and cash, repays its existing debt, covers fixed op-
erating costs, and potentially issues new debt. Consequently, the firm’s cash flow during
this period can be expressed as follows:
ι
d = π (z, y∗ ) − ι − Ψ , ω + c − c′ + q z, ω ′ , c′ , b′ b′ − b

(4)
ω
Following Hennessy and Whited (2005, 2007), we denote d ≤ 0 as costly equity is-
suance and d > 0 as shareholder distributions. In addition, we assume that issuing equity
incurs both a fixed cost component λ0 > 0 and a linear cost component λ1 > 0.
Risky debt pricing We assume that the firm borrows from a competitive, risk-neutral
lender. The lender offers a state-contingent contract that is priced to compensate the
lender for the potential loss incurred in the event of the firm’s default. More specifi-
cally, if the firm continues its operations in the subsequent period, it fully repays the debt
amount of b′ . However, in the case of default, the firm pays the lender c′ + χπ (z′ , (y∗ )′ ),
19

where χ ∈ (0, 1) represents the recovery rate and governs the bankruptcy costs. Under
these conditions, the price of debt can be derived directly from the lender’s zero-profit
condition:

1
′ ′ ′ ′
E 1V ′ ≥0 b′ + 1V ′ <0 c′ + χ max{π z′ , (y∗ )′ , 0}

q z, ω , c , b b =
1 + r (1 − τ )
(5)
where the term 1V ′ ≥0 represents an indicator function that takes a value of one when the
equity value is nonnegative and zero otherwise. The equation above simply shows that
the market value of debt should be equal to the present value of the default probability-
weighted average of the principal repayment and the recovery in the event of default. The
term 1 − τ introduces a tax advantage for debt.
Summary of firm’s optimization problem To describe the firm’s optimization problem,

it is important to note that the firm defaults when the equity value V (z, ω, c, b) falls below
zero. Therefore, we can express the condition for default as follows:
V (z, ω, c, b) = max {0, V c (z, ω, c, b)} (6)
where V c (z, ω, c, b) denotes the continuation value, and it can further be expressed as

1
V c (z, ω, c, b) = max d + 1d≤0 (−λ0 + λ1 d) + E V z′ , ω ′ , c′ , b′

(7)
ι,b′ ,c′ ,y∗ 1+r
20

subject to the following five different constraints:
ι
d = π (z, y∗ ) − ι − Ψ , ω + c − c′ + q z, ω ′ , c′ , b′ b′ − b

ω
1
q z, ω ′ , c′ , b′ b′ = E 1V ′ ≥0 b′ + 1V ′ <0 c′ + χ max{π z′ , (y∗ )′ , 0}

1 + r (1 − τ )
′
ω = ι + (1 − δ ) ω
y∗ ≤ ω
ω ′ , b′ , c′ ≥ 0
3.2 Demand and Supply Origins of The Risky Superstar Economy
After outlining the model setup, now we turn to discuss two key model implications. The
first key implication is that shifts in the demand curve p (z, y) and supply curve C (y) are
redistributive in not only the level of corporate earnings or markup but also the risk.
We begin with discussing how a firm’s markup µ and earnings π are linked to its
product quality z. In this paper, the definition of a firm’s markup µ is the ratio of profits
py
to total costs, i.e., µ = C (y)
. Meanwhile, the definition of a firm’s earnings π is the same
as before, i.e., the difference between profits and total costs, i.e., π = py − C (y). With the
optimization problem shown in (7), the results on how markup µ and earnings depend
on product quality z are summarized in Lemma 1.
∂p
Lemma 1. If the demand curve slope is relatively steep, i.e., ∂z is large enough, and/or the cost
function is relatively flat, i.e., C ′′ is small enough, then markup µ and earnings π become a convex
∂µ ∂π ∂2 µ ∂2 π
function of product quality z, i.e., ∂z > 0, ∂z > 0, ∂z2
> 0, and ∂z2
> 0.
All proofs are provided in Appendix A. Lemma 1 summarizes our key results related
to the origins of markup. Three important conclusions can be drawn from this lemma.
First, both markup and earnings are increasing in capital quality z. This outcome is quite
intuitive given the model setup used in this paper. To begin with, we assume that con-
∂p
sumers in this economy have a taste for quality, i.e., ∂z > 0. As a result, entrepreneurs
with better product quality face a higher demand, which makes them able to charge
higher prices and sell more products. At the same time, if the total production cost does
21

not increase rapidly when the number of products sold increases, i.e., C ′′ being small,
then firms with better product quality can suffer less from increasing operating scale,
which also contributes to their increased profits. Hence, with these model setups, it is
not surprising that markup is increasing in product quality. As markup is positively and
closely related to earnings, we can obtain a similar conclusion for the earnings-quality
relationship.
Second, the emergence of superstar firms is attributed to shifts in both demand and
supply curves. Lemma 1 demonstrates that the relationship between earnings and markup
is ultimately determined by consumer preferences and production technology parame-
ters. Changes in the supply and demand curves have substantial effects on the distribu-
tion of markup. For instance, in an economy where there is a low preference for product
quality and significant supply costs, we would expect the markup and earnings to ex-
hibit a concave or linear relationship with product quality. In such an economy, although
markup and earnings still increase with product quality, no single entrepreneur dom-
inates the market. However, if there are permanent shifts in demand and supply, the
markup and earnings become a convex function of product quality. In this new economic
environment, consumers display a strong preference for high-quality products, and it is
relatively inexpensive for entrepreneurs to serve all potential customers. In this scenario,
superstars have the ability to leverage their market power and establish dominance across
the entire industry.
Third, our story is consistent with the evidence from the new intangible economy.
Studies such as Hsieh and Rossi-Hansberg (2019) argue that in the context of intangibles
like software, deployment across different markets is possible only after incurring fixed
costs such as R&D. Additionally, De Ridder (2019) directly incorporates intangibles as
inputs that shift the cost structure from marginal costs to fixed costs. This interpretation of
the intangible economy helps explain the slowdown in productivity growth, the decline
in business dynamism, and the rise of market power. Importantly, our model setup is
consistent with De Ridder (2019)’s empirical findings, which indicate a significant increase
in fixed costs for U.S. firms. Specifically, among publicly traded U.S. firms, the proportion
22

of fixed costs in total costs has risen from 13.9% in 1980 to 24.5% in 2015. However, it is
worth noting that our story is not limited to intangible capital alone, as a similar pattern
emerges for manufacturing firms as well. For instance, the adoption of robotic automation
in recent years has led to operational efficiencies but necessitates increased investment in
fixed production costs.
Then we investigate how this convexity affects both expected earnings level and risk.
The result is summarized in the following lemma.
Lemma 2. (Risky Superstars) If markup µ and earnings π are a convex function of product quality
z, then the expected level and volatility of earnings are both increasing in quality z.
The proof is also provided in the appendix. The economics behind Lemma 2 is quite
intuitive. The corporate earnings risk means how much do earnings vary with some ex-
ogenous changes in product quality. Once corporate earnings π (z) become a convex
function of z, then it means the slope π ′ (z) also becomes steepest in the right tail. There-
fore, for right-tail firms, corporate earnings become very sensitive to even small changes
in product quality.
The most important implication from Lemma 2 is that when firms face uncertainty
in their product quality, from a dynamic perspective, superstars are inherently riskier.
Compared to the low-concentration traditional economy, in a superstar economy, a small
variation in product quality can translate into considerable earnings fluctuations, espe-
cially on the right-tail side. In other words, these economic fundamental changes are also
redistributive in risk. The risky superstar economy essentially comes from the fact that
earnings become a convex function of the underlying quality. With convexity, both in-
come level and risk are redistributed towards right-tail firms. Therefore, if there were
some permanent changes in the shape of preference and supply functions, the expected
future earnings and expected volatility of future earnings vary across firms with different
capital qualities. As we will see in Section 3.3, the changed risk profile also explains why
powerful firms hold excessive cash in the new economy.
23

Continuous-time version Our conclusion here does not depend on the specific choice
of the underlying capital quality process. The generalization can be best illustrated with
Ito’s lemma in a continuous-time setup. Assume that each entrepreneur’s ability z follows
any stochastic differential equation dz = f (z) dt + g (z) dW , where f and g can be any
functions of z. In addition, we assume that corporate earnings π are a function of z, i.e.,
π = π (z). By Ito’s Lemma, we can derive the firm’s earnings through the following
stochastic process:

1 ′′ 2 1 ′′ 2
dπ = π (z) dz + π (z) g (z) dt = π (z) f (z) + π (z) g (z) dt + π ′ (z) g (z) dW
′ ′
2 2
(8)
Based on the equation above, as long as π is not a linear function of z, i.e., π ′ is not
independent of z, regardless of the underlying capital quality process, the actual earnings
process that eventually affects entrepreneur’s investment and cash holding decisions is a
quality-based and non-homogeneous one.
Again, mathematically, the rise of these risky superstars comes from the fact that shifts
in demand and supply curves make earnings become a convex function of capital quality.
More specifically, due to the convexity of π with respect to z, two implications arise di-
rectly from the earnings process (8). On the one hand, π ′ (z) on the drift coefficient shows
that star firms can become superstars, which is also the common focus of superstar effects
in the existing literature. On the other hand, π ′ (z) on the diffusion coefficient means that
superstar firms are inherently riskier. Therefore, changes in economic fundamentals af-
fect both the drift and diffusion parts simultaneously, which means that superstar firms
can take the most but at the cost of bearing more earnings uncertainty in the future. In
this way, these fundamental changes are redistributive in both income and risks. In other
words, superstar firms are not merely large versions of small firms as they face the highest
expected future earnings but also the most volatile income fluctuations. This unique char-
acteristic of a superstar economy substantially changes the entrepreneur’s cash holdings
and investment decisions.
24

3.3 Risky Market Power and Cash Holdings
The second key implication from our model is that the changed risk profile is also going to
affect the firms’ incentives for cash holdings, especially for the right-tail powerful firms.
Lemma 3. The condition for optimal cash holdings can be shown as follows:
  
 bond price channel 
1   z}|{
′

′ 
1 + 1d <0 λ 1 =  E 1V ′ ≥ 0 ( 1 + 1 d ′ < 0 λ 1 ) + ( 1 + 1 d ′ < 0 λ 1 ) q c′ b +µ  (9)
1+r 
| {z } 

precautionary saving channel
where qc′ is the first derivative of the price of debt with respect to cash, and µ is the Lagrange
multiplier on the cash nonnegativity constraint.
Equation (9) can be directly obtained if we calculate the derivative of Equation (7) with
respect to cash c. The left-hand side of the equation represents the marginal cost associated
with holding an additional dollar of cash. When shareholder distributions are positive,
this cost corresponds to the marginal value of forgone distributions, which is equal to
one. However, in cases where distributions are negative (i.e., the firm issues equity), the
marginal cost is given by one plus the linear cost of equity issuance. Although the fixed
cost of equity issuance is not directly included in the equation, it does impact the firm’s
optimal cash policy by influencing the extent to which equity issuance is pursued.
The right-hand side of the equation represents the discounted value of the firm’s
marginal benefit derived from holding cash, comprising several components. The most
important implication here is that there are two opposing forces at play in determining the
optimal cash holdings. The first force is the precautionary motive, where the firm holds
cash to mitigate the need for external equity financing. By holding cash, the firm reduces
the probability of entering states that require costly external financing. This precautionary
motive for cash accumulation is a standard feature in dynamic models of this nature (e.g.,
Riddick and Whited, 2009; Begenau and Palazzo, Forthcoming). The second force comes
from the impact of increased internal liquidity on the price of debt. The price of debt is
positively related to cash holdings, indicating that firms with greater internal liquidity
25

face lower borrowing costs. This pattern is consistent with the financial frictions story
and some recent studies (e.g., Gao et al., Forthcoming), where firms accumulate cash due
to a higher probability of default or the presence of market frictions such as borrowing
constraints. As a result, a negative relationship exists between a firm’s cash holdings and
its financial vulnerability for firms at the lower end of the distribution. We label the first
channel as the precautionary saving channel and the second as the bond price channel. As we
can see from Equation (9), these two channels also interact with each other.
In this paper, our focus lies on cross-sectional differences. With these two different
channels, we can discuss how market power affects equilibrium corporate cash holdings.
In a superstar economy, the precautionary saving channel is increasing in markup, mean-
while, the default risk channel is decreasing in markup, given the same leverage level.
If the markup increases, default risk decreases, so the risky bond price is higher. In this
way, the second bond price channel indicates that holding cash becomes less valuable to
the firm with a higher markup. However, on the other hand, as the markup increases in
the risky superstar economy, the earnings volatility increases. Therefore, the precaution-
ary motive for holding cash increases, as it allows the firm to avoid high external equity
issuance costs. Eventually, the shape of the cash-markup relationship depends on the pa-
rameters and the relative importance of each channel, which becomes our main focus in
the following quantitative exercise.
3.4 Reduced-form Evidence for the Risky Superstar Economy
One testable implication from our model is that superstar firms are riskier than other
firms. Before our quantitative exercise, here we provide some empirical evidence to sup-
port this claim. In Figure 6, we plot the time series of markup volatility for five groups
of firms with different levels of markup. Consistent with the model, we define super-
star firms to be those with the highest markup. We obtain the results in Figure 6 using
the following steps. First, for each individual company in each period, we compute its
markup volatility by using a five-year rolling window. Second, in each year, we classify
all the firms with available markup measures into five different quantiles according to
26

their markup level. Finally, we calculate the annual average markup volatility for each
quantile of firms. As shown in Figure 6, on average, the group of firms with the highest
markup level also has the largest degree of markup volatility. Over the sample period,
the average markup volatility for the lowest quantile group is 0.129. In contrast, super-
star firms have average markup volatility as high as 1.097, which is a substantially higher
number. In addition, there is a clear upward trend until 2004 in the measured degree of
riskiness for the superstar firms. It indicates that compared to the rest of the firms, su-
perstars are over time becoming much riskier, which is also consistent with the model
implications.
Figure 6: Risky Superstars
Notes: This figure presents the time-series plots of the markup volatility for five groups of firms
with different levels of markups. The main data source for this figure is the Compustat North
American Annual data file. Firm-level markup is estimated with De Loecker et al. (2020)’s pro-
duction cost function approach, and markup volatility is estimated as a five-year rolling window.
In addition, our empirical finding here on the positive association between market
power and cash flows riskiness is consistent with some recent asset pricing studies (e.g.,
Corhay et al., 2020; Dou et al., 2021, Forthcoming). These papers provide the same sup-
porting evidence – powerful firms are riskier – but from a stock return perspective.
27

Two caveats are worth noting for interpreting the results here. First, our finding seems
to contradict Herskovic et al. (2016), which found that large firms have small total return
or sale growth volatility. This difference comes from the fact that we have different defini-
tions of superstar firms and riskiness. In Herskovic et al. (2016), large firms are defined as
firms with more total assets. In contrast, we define superstar firms as those with a higher
markup. Moreover, Herskovic et al. (2016) measures the riskiness of firms by using the
total return volatility or sales growth volatility. In contrast, we adopt markup volatility as
the measure of riskiness.
Second, there is a crucial difference between what the model actually implies and what
we are able to observe from the data. In the model, we argue that superstar firms should
carry out more risk management because they are inherently riskier ex-ante. However,
the model does not imply that after adopting these risk management policies, superstar
firms are still riskier ex-post. In addition, in practice firms could develop different abilities
to hedge their idiosyncratic risks with various financial instruments, which could lead
to diametrically opposite conclusions if we focus on different aspects of an equilibrium
outcome. For example, the low turnover rate of dominant firms could come from their
successful risk management instead of the fact that they are not risky.
4. Quantitative Analysis
In this section, we investigate to what extent our previous model is able to quantitatively
explain the link between market power and cash holdings. In addition, we also study
the relative importance of the supply versus demand origin of corporate market power,
as well as the relative importance of the precautionary saving channel and bond price
channel shown in our previous model section.
4.1 Parameterization
Functional forms We start by explaining the functional forms used for our following
quantitative exercises. To begin with, we assume that idiosyncratic product quality fol-
28

lows a standard AR-1 autoregressive process in logs:
ln z′ = ρz ln z + ε′z , ε z ∼ N 0, σz2

(10)
where a prime denotes a variable in the next period. The parameter ρz governs the per-
sistence of z, and the innovation to z, i.e., ε z , is normally distributed with variance σz2 .
Then we assume that the price function and the production cost function adopt the
following functional forms:
p (z, y) = zα y−γ (11)

1
C ( y ) = c0 + c1 y β (12)
where α and γ measure how sensitive is price to variations in quality and quantity, respec-
tively. More specifically, α reflects customers’ taste for products with different qualities.
Two conclusions are worth noting here. First, price is increasing in product quality and
decreases in quantity. Second, changes in customer’s taste α can directly affect the (rela-
tive) prices of products with different qualities. Following the existing literature (e.g., De
Ridder, 2019), we assume that the cost function contains a fixed cost of c0 and a variable
1
cost component with marginal cost c1 > 0 and a scale parameter 0 < β ≤ 1. Here β
denotes the curvature of the supply curve, i.e., how costly it is for firms to expand their
operating scale. In addition, it could also represent how much the current technology
admits the joint consumption or the degree of degradation of services as entrepreneurs
1
increase their scale. One typical example of changes in β is digitization: the availabil-
ity of the internet largely reduces the costs for companies to serve a massive number of
potential buyers.
Furthermore, we assume that the customer capital adjustment cost function uses a
functional form shown as follows:
ι ψ ι 2
Ψ ω, = ψ0 ω1ι̸=0 + 1 ω (13)
ω 2 ω
29

where both ψ0 and ψ1 measure the degree of investment inflexibility and ψ0 , ψ1 > 0.
External calibration The model is calibrated at an annual frequency. To reduce the com-
putational burden, we externally calibrate a subset of parameters, then consider the rest
estimated within the model. For estimating those structural parameters, we adopt the
SMM approach (e.g., McFadden, 1989; Nikolov and Whited, 2014), as there are no closed-
form solutions. In addition, in our main cross-sectional analysis, following some recent
studies (e.g., Farhi and Gourio, 2018), we choose 2000 as the midpoint and split the histor-
ical dataset into two different subsamples: the 1980-1999 subsample is interpreted as the
traditional economy, and the 2000-2015 subsample is labeled as the superstar economy. As
for the time-series analysis in Section 4.2.2, we do this exercise for each 5-year subsample
and compare the model-implied average cash-to-asset ratio and the corresponding one in
the data.
Those externally calibrated parameters are shown in Panel (A) of Table 2. Following
Riddick and Whited (2009), we set the risk-free interest rate r to be 4%. The fixed external
equity financing cost λ0 and the proportional cost λ1 are set to be 0.007 and 0.054, respec-
tively. This choice is based on Gao et al. (Forthcoming), which computes these parameters
by regressing the firm-level issuance-fees-to-asset ratio on proceeds-to-asset ratio with the
Securities Data Corporation database. We also choose the deadweight cost to be 0.665. As
our focus is on the right-tail firms’ behaviors, our main conclusions are not sensitive to
the specific choices of this parameter’s value. Finally, following the standard literature
(e.g., Nikolov and Whited, 2014), we set the tax rate τ to be 0.2.
Internal estimation The rest of the parameters are jointly calibrated with the SMM ap-
proach by targeting some moments in the data. More specifically, eleven parameters are
calibrated through this approach: product quality persistence ρz , product quality shock
variance σz , price-quality sensitivity α, price-quantity sensitivity γ, supply curvature β,
fixed production cost c0 , variable production cost c1 , fixed adjustment cost ψ0 , variable
adjustment cost ψ1 , fixed production cost c0 , variable production cost c1 , fixed adjustment
cost ψ0 , variable adjustment cost ψ1 , and customer capital depreciation rate δ. The data
30

Table 2: Parameterization
(A) External Calibration
Parameter Description Value Source/Reference

r risk-free interest rate 0.04 Riddick and Whited (2009)
λ0 fixed external financing cost 0.007
λ1 proportional equity financing cost 0.054 Gao et al. (Forthcoming)
χ deadweight default costs 0.665
τ tax advantage 0.2 Nikolov and Whited (2014)
(B) Internal Estimation
Parameter Description Value

1980-1999 subsample 2000-2015 subsample
product quality process
ρz product quality persistence 0.64 0.55
σz product quality shock variance 0.54 0.60
demand factors
α price-quality sensitivity 0.50 0.66
γ price-quantity sensitivity 0.81 0.69
supply factors
1
β supply curvature 1.88 1.45
c0 fixed production cost 0.71 0.92
c1 variable production cost 0.58 0.32
others
ψ0 fixed adjustment cost 0.15 0.50
ψ1 variable adjustment cost 0.97 1.22
δ customer capital depreciation rate 0.11 0.25
Notes: This table presents the values of parameters used in our model. The sample is constructed from
Compustat annual database and covers from 1980 to 2015.
31

moments we choose are listed in Table 3, which includes mean and dispersion of markup,
investment-to-output ratio, cash-to-output ratio, and debt-to-output ratio. As our story
is much closer to the intangible capital story instead to a physical capital story, therefore
we choose the sum of selling, general, and administrative expenditures (Compustat data
item XSGA) as a proxy for investment. In addition, following the previous studies such
as Han and Phillips (2010) and Bazdresch et al. (2018), we include the serial correlation of
operating income, the regression coefficient of cash on debt, and the regression coefficient
of cash on markup. All of these moments are informative for our identification.
Table 3: Target Moments
Moments Model Counterpart 1980-1999 subsample 2000-2015 subsample

data model data model
mean cash-to-output c
0.32 0.38 0.89 0.72
c+ω
standard deviation cash-to-output 1.31 1.60 3.42 2.39
mean debt-to-output b
0.38 0.41 0.57 0.49
c+ω
standard deviation debt-to-output 0.73 0.80 1.53 1.00
mean markup py 1.46 1.50 2.05 1.88
c0 + c1 y
standard deviation markup 0.37 0.42 0.50 0.46
mean XGSA-to-output ω
0.31 0.33 0.47 0.39
c+ω
standard deviation XGSA-to-output 0.35 0.41 1.02 0.53
mean production-cost-to-output c0 + c1 y 0.82 0.85 1.14 1.09
c+ω
standard deviation production-cost-to-output 1.11 1.32 2.58 2.17
serial correlation of operating income corr (πt , πt−1 ) 0.36 0.36 0.17 0.16
c b−c
coefficient of cash on debt reg coeff of c+ω on c+ω 0.38 0.43 0.49 0.46
c py
coefficient of cash on markup reg coeff of c+ω on c0 + c1 y 0.27 0.30 0.15 0.20
Notes: This table presents the 13 moments from the estimation of our model parameters. The sample is
constructed from Compustat annual database and covers from 1980 to 2015.
According to Table 3, our model fits the data moments reasonably well. Panel (B) in
Table 2 presents the point estimates of these structural parameters. The long-term changes
in these parameter values are consistent with our intuition. For instance, in the traditional
economy, the estimated value of α is 0.50. In contrast, with the subsample dataset on the
superstar economy, α is estimated to have a value of 0.66. Therefore, when taking the
previous model to the data, it shows that people indeed have stronger preferences for
32

product quality in today’s economy. At the same time, the estimated marginal product
cost has declined from 0.58 to 0.32. One possible reason behind this is certain scale-related
technical changes such as digitization, which allows individual firms to easily serve a
large group of buyers with nearly zero marginal cost. In addition, the estimated fixed cost
has increased from 0.71 to 0.92, which is consistent with what has been documented in De
Ridder (2019).
4.2 Quantitative Performance
4.2.1 Cross-sectional relationship
In this section, we provide some cross-sectional evidence to validate the model and the
choice of parameters. It is worth noting that when parameterizing the model, we do not
directly target these moments. Therefore, the goodness of fit in these moments can be
informative for evaluating the model’s validity.
We focus on how the model fits the cross-sectional cash-to-output ratios. The results
are presented in Figure 7, where we include both the data and model for better com-
parison. The data part is created as follows. To begin with, for each year between 1980
and 1999, we split all the firms into five different groups according to their markup level.
Throughout this section, Q1 represents firms with the smallest markup, and Q5 means the
group of firms with the highest values of markup. After that, we compute the firm-level
cash-to-output ratios, then take averages for each group in each year. Finally, we compute
the subsample average for each group throughout the subsample period. At the same
time, the model part is generated by simulating a panel of 2,000 firms. The firms behave
optimally according to the first-order conditions derived in our model section. Then we
classify all the firms into five different groups according to their markup level and com-
pute the model-implied average cash-to-output ratios for each group. Finally, the bottom
picture in Figure 7 is obtained by computing the difference in these numbers between the
traditional and the new economy.
Figure 7 shows that our model is able to quantitatively match the cross-sectional pat-
33

Figure 7: Cross-Sectional Validation: Data v.s. Model
Notes: This table presents the average cash-to-output ratios for firms with different levels of
markup. Q1 represents firms with the smallest markup, and Q5 means the group of firms with
the highest values of markup. We classify the whole sample into two subsamples: the traditional
economy sample (1980-1999) and the superstar economy subsample (2000-2015). For different sub-
samples, the model solution is computed with parameter values listed in Table 2. The data sample
is constructed from Compustat and covers from 1980 to 2015.
terns of cash-to-output ratios in both subsamples. Our model implies that all the firms
in different groups tend to save more in the superstar economy, especially for the firms
with the highest markups. This cross-sectional pattern is consistent with what we find
in the data. As previously explained, the underlying mechanism for this outcome is that
the shifts in demand and supply curves raise both the expected profits and uncertainty in
the future; therefore, this risk redistribution channel makes all firms save more internally,
especially the superstar firms. In terms of magnitudes, what is less satisfying here is that
the model tends to overestimate the cash-to-output ratio for the firms with the smallest
markup, and underestimate it for the firms with the highest markup. One possible expla-
nation is that we assume all firms face the same investment opportunity and tax rate. In
reality, although, these assumptions may not hold. As mentioned in Gu (2017) and Bege-
34

nau and Palazzo (Forthcoming), firms may have other incentives to accumulate cash such
as multinational income sources and R&D investment needs. These incentives might be
stronger for firms with higher markup levels.
4.2.2 Time-series relationship
Now we turn to investigate the time-series performance of our model. As mentioned

before, here we redo our exercise for each of the 5-year-window subsamples between
1980 and 2015. Then we compare the trends in the average cash-to-output ratio both in
the data and in the model. Our main results with both time series are presented in Figure
8.
Figure 8: Time-Series Validation: Data v.s. Model
Notes: This table presents the time-series median cash-to-output ratio both in the model and in the
data. For each of the 5-year-window subsamples between 1980 and 2015, we implement an SMM
estimation to obtain the parameter values and then use these estimated parameters to calculate
the median cash-to-output ratio in the model. The data sample is constructed from Compustat and
covers from the period from 1980 to 2015.
According to Figure 8, we can draw two conclusions. First, our model is able to repro-
duce the increasing trend of the average cash-to-output ratio in the data. It means that the
35

increasing riskiness of market power, arising from shifts in demand and supply curves,
is likely to be the reason why (powerful) firms hold excessive cash. Second, compared to
the data counterpart, our model tends to overestimate the incentives of cash hoarding, es-
pecially after 2000. One possible explanation is that in our quantitative exercise, in order
to reduce the computational burden in the SMM estimation, we assume that the external
financing cost is the same throughout our sample. However, in reality, due to the devel-
opment of new financial technologies and reduced regulation requirement, these external
financing costs should be significantly smaller today. Under this situation, firms will have
fewer incentives to accumulate cash.
4.3 Counterfactual Exercises
4.3.1 Supply origin versus demand origin
As mentioned in Section 3, there are two origins of this risky superstar economy, and
also two channels through which market power affects optimal cash holdings. In order
to show the relative contributions of each origin and each channel, here we conduct some
counterfactual exercises. Our first exercise is to explore the relative importance of the sup-
ply versus demand origin in explaining the joint rise in corporate market power and cash
holdings. Our main results on both time-series and cross-sectional analyses are provided
in Figure 9. The way we implement time-series and cross-sectional analyses is similar to
what we have done before. The only difference here is that in order to evaluate the rela-
tive importance of each story, we fix all related parameters and see how much our model
is able to match the patterns in the data. For instance, when investigating the importance
of demand factors, we fix the parameter values of price-quality sensitivity α and price-
quantity sensitivity γ as they were in the 1980 (for the time-series analysis) or 1980-1999
subsample (for the cross-sectional analysis), then we compute the model equilibrium. The
process is similar for evaluating the importance of supply factors.
The main findings from Figure 9 are twofold. First, for both the time-series and cross-
sectional analyses, the most important driver behind these trends in corporate cash hoard-
36

Figure 9: Counterfactual Exercise: Demand v.s. Supply
(a) time-series analysis (b) cross-sectional analysis
Notes: This table presents the counterfactual exercises where either the demand factor or supply
factor is fixed. Supply-related parameters are supply curvature β1 , fixed production cost c0 , and
variable production cost c1 . Demand-related parameters are price-quality sensitivity α and price-
quantity sensitivity γ. The sample is constructed from Compustat annual database and covers from
1980 to 2015.
ing behaviors is the change in production-technology-related supply factors. As shown

in our counterfactual exercise in Figure 9, if the production technology were unchanged
across the past several decades, the changes in the demand side can only account for a
relatively small portion of these secular changes in corporate cash holdings. Therefore,
our quantitative exercise here suggests that the emergence of digitization and other tech-
nical changes bring us a society of low marginal cost, which fundamentally changes the
relationship between earnings and product quality. Consequently, it also affects the long-
run trends in some important corporate behaviors such as corporate reliance on internal
financing.
Second, the change in the demand side is also relatively important, and it has become
increasingly important since 2000. Our exercise here implies that the increasing prefer-
ence of consumers over high-quality products also contributes significantly to the rise of
superstar firms and their cash-holding behaviors. Therefore, technology is not the only
important reason behind the risky superstar economy, the demand side is also quantita-
37

tively important. One possible explanation for the increasing preference for high-quality
products is the secular increase in inequality. As rich people have a higher taste for quality,
the increased wealth inequality boosts the aggregate price-quality sensitivity α. However,
exploring the precise reason behind this secular change in the demand side is beyond the
scope of this paper, and we leave it for further research.
4.3.2 Precautionary saving channel versus bond price channel
Our second exercise is to explore the relative importance of the precautionary saving
channel and bond price channel through which market power affects equilibrium cash
levels. Our main results on both time-series and cross-sectional analyses are provided in
Figure 10. The way we conduct the counterfactual exercise here is similar to what we have
done in Figure 9.
Figure 10: Counterfactual Exercise: Precautionary Saving v.s. Bond Price
(a) time-series analysis (b) cross-sectional analysis
Notes: This table presents the counterfactual exercises where either corporate earnings volatility or
bond prices are fixed. The sample is constructed from Compustat annual database and covers from
1980 to 2015.
Based on the results in Figure 9, two conclusions are worth noting. First, for explaining
the time-series upward trend, both channels play a non-negligible role, despite that the
precautionary saving channel is relatively more important. This result is not surprising
38

as our key mechanism is that shifts in demand and supply curves increase the riskiness
of corporate earnings. With increased risk, both channels will make firms hold more cash
and rely more on internal financing. As we can see from Graph (a), in the time-series
analysis, both channels can explain a certain fraction of the upward trend in the equilib-
rium cash-to-output ratio. In addition, the bond price channel has become increasingly
important since 2000.
Second, in our cross-sectional analysis shown in Graph (b), these two channels play
different roles. More specifically, the precautionary saving channel is more important in
explaining the right tail while the bond price is more important in explaining the left tail.
Therefore, there indeed exist two different mechanisms for why companies hold cash, and
the relative importance of each channel is also different for firms with different markup
levels. More importantly, the risky market power story in our paper is important for our
understanding of why powerful firms hold excessive cash in the data.
5. Conclusion
Over the past several decades, we have observed both the increasing corporate market
power and increasing corporate internal financing in the data. In this paper, we argue
that these two phenomena are deeply connected, and we provide a theoretical and quan-
titative framework to explain them jointly. The underlying mechanism comes from that
two economic fundamental changes from both the demand and supply sides can directly
impact the level and volatility of corporate earnings. More specifically, the changes in
consumers’ taste for quality and producers’ marginal supply cost increase the earnings-
quality gradient sharply in the right tail, which generates a “winners-take-most” phe-
nomenon and makes current winners inherently riskier. This income and risk redistribu-
tion generates a positive correlation between markup and cash value, prompting super-
star firms to rely more on internal financing. Finally, when taken to data, our model is
able to quantitatively match the corresponding empirical patterns.
For simplicity, our paper here uses a partial equilibrium framework to focus on how
39

changing economic fundamentals affect market power and cash holdings jointly. How-
ever, it is also interesting to investigate whether those changes in corporate risk manage-
ment policy could also lead to some aggregate impacts on the whole macroeconomy. We
leave these studies for future research.
References
Autor, David, David Dorn, Lawrence F. Katz, Christina Patterson, and John Van Reenen, “The Fall
of the Labor Share and the Rise of Superstar Firms,” Quarterly Journal of Economics, 2020, 135 (2),
645–709.
Azar, Jose A., Jean-François Kagy, and Martin C. Schmalz, “Can Changes in the Cost of Carry
Explain the Dynamics of Corporate “Cash” Holdings?,” Review of Financial Studies, 2016, 29 (8),
2194–2240.
Bates, Thomas W., Kathleen M. Kahle, and Rene M. Stulz, “Why Do U.S. Firms Hold So Much
More Cash than They Used To?,” Journal of Finance, 2009, 64 (5), 1985–2021.
Bazdresch, Santiago, R. Jay Kahn, and Toni M. Whited, “Estimating and Testing Dynamic Corpo-
rate Finance Models,” Review of Financial Studies, 2018, 31 (1), 322–361.
Begenau, Juliane and Berardino Palazzo, “Firm Selection and Corporate Cash Holdings,” Forth-
coming.
Brander, James A. and Tracy R. Lewis, “Oligopoly and Financial Structure: The Limited Liability
Effect,” American Economic Review, 1986, 76 (5), 956–970.
Corhay, Alexandre, Howard Kung, and Lukas Schmid, “Competition, Markups, and Predictable
Returns,” Review of Financial Studies, 2020, 33 (12), 5906–5939.
De Loecker, Jan and Frederic Warzynski, “Markups and Firm-Level Export Status,” American Eco-
nomic Review, 2012, 102 (6), 2437–2471.
, Jan Eeckhout, and Gabriel Unger, “The Rise of Market Power and the Macroeconomic Impli-
cations,” Quarterly Journal of Economics, 2020, 135, 561–644.
De Ridder, Maarten, “Market Power and Innovation in the Intangible Economy,” 2019.
Dou, Winston Wei and Yan Ji, “External Financing and Customer Capital: A Financial Theory of
Markups,” Management Science, Forthcoming.
, , and Wei Wu, “Competition, Profitability, and Discount Rates,” Journal of Financial Eco-
nomics, 2021, 140, 582–620.
, , and , “The Oligopoly Lucas Tree,” Review of Financial Studies, Forthcoming.
Farhi, Emmanuel and Francois Gourio, “Accounting for Macro-Finance Trends: Market Power,
Intangibles, and Risk Premia,” Brookings Papers on Economic Activity, 2018, Fall.
40

Foley, C. Fritz, Jay C. Hartzell, Sheridan Titman, and Garry Twite, “Why do firms hold so much
cash? A tax-based explanation,” Journal of Financial Economics, 2007, 86 (3), 579–607.
Gabaix, Xavier and Augustin Landier, “Why has CEO Pay Increased So Much?,” Quarterly Journal
of Economics, 2008, 123 (1), 49–100.
Gao, Janet, “Managing Liquidity in Production Networks: The Role of Central Firms,” Review of
Finance, 2021, 25 (3), 819–861.
Gao, Xiaodan, Toni M. Whited, and Na Zhang, “Corporate Money Demand,” Review of Financial
Studies, Forthcoming.
Gomez, Matthieu, “Asset Prices and Wealth Inequality,” 2019.
Gu, Tiantian, “U.S. multinationals and cash holdings,” Journal of Financial Economics, 2017, 125,
344–368.
Hall, Robert E., “The Relation between Price and Marginal Cost in U.S. Industry,” Journal of Political
Economy, 1988, 96 (5), 921–947.
Han, Chirok and Peter C. B. Phillips, “Estimating and Testing Dynamic Corporate Finance Mod-
els,” Econometric Theory, 2010, 26 (1), 119–151.
Hayashi, Fumio, “Tobin’s Marginal q and Average q: A Neoclassical Interpretation,” Econometrica,

1982, 50 (1), 213–224.
Hennessy, Christopher A. and Toni M. Whited, “Debt Dynamics,” Journal of Finance, 2005, 60 (3),
1129–1165.
and , “How Costly Is External Financing? Evidence from a Structural Estimation,” Journal of
Finance, 2007, 62 (4), 1705–1745.
Herskovic, Bernard, Bryan Kelly, Hanno Lustig, and Stijn Van Nieuwerburgh, “The common fac-
tor in idiosyncratic volatility: Quantitative asset pricing implications,” Journal of Financial Eco-
nomics, 2016, 119 (2), 249–283.
Hoberg, Gerard and Gordon M. Phillips, “Scope, Scale and Competition: The 21st Century Firm,”
2021.
Hsieh, Chang-Tai and Esteban Rossi-Hansberg, “The Industrial Revolution in Services,” 2019.
NBER Working Paper w25968.
Im, Hyun Joong, Heungju Park, and Gege Zhao, “Uncertainty and the value of cash holdings,”
Economics Letters, 2017, 155, 43–48.
Jung, Hae Won (Henny), Dalida Kadyrzhanova, and Ajay Subramanian, “Capital Structure under
Imperfect Product Market Competition: Theory and Evidence,” 2021.
Kaplan, Steven N. and Luigi Zingales, “Do Investment-Cash Flow Sensitivities Provide Useful
Measures of Financing Constraints?,” Quarterly Journal of Economics, 1997, 112 (1), 169–215.
41

Kehrig, Matthias and Nicolas Vincent, “The Micro-Level Anatomy of the Labor Share Decline,”
2020. NBER Working Paper No. 25275.
Kroen, Thomas, Ernest Liu, Atif R. Mian, and Amir Sufi, “Falling Rates and Rising Superstars,”
2021.
Liu, Ernest, Atif Mian, and Amir Sufi, “Low Interest Rates, Market Power, and Productivity
Growth,” 2019.
Luttmer, Erzo G.J., “Technology diffusion and growth,” Journal of Economic Theory, 2012, 147 (2),
602–622.
MacKay, Peter and Gordon Phillips, “How Does Industry Affect Firm Financial Structure?,” Review
of Financial Studies, 2005, 18 (4), 1433–1466.
Maksimovic, Vojislav, “Capital Structure in Repeated Oligopolies,” RAND Journal of Economics,

1988, 19 (3), 389–407.
McFadden, Daniel, “A Method of Simulated Moments for Estimation of Discrete Response Models
Without Numerical Integration,” Econometrica, 1989, 57 (5), 995–1026.
Morlacco, Monica and David Zeke, “Monetary policy, customer capital, and market power,” Jour-
nal of Monetary Economics, 2021, 121, 116–134.
Nikolov, Boris and Toni M. Whited, “Agency Conflicts and Cash: Estimates from a Dynamic
Model,” Journal of Finance, 2014, 69 (5), 1883–1921.
Opler, Tim, Lee Pinkowitz, Rene Stulz, and Rohan Williamson, “The determinants and implica-
tions of corporate cash holdings,” Journal of Financial Economics, 1999, 52 (1), 3–46.
Peters, Ryan and Lucian Taylor, “Intangible capital and the investment-q relation,” Journal of Fi-
nancial Economics, 2017, 123 (2), 251–272.
Pinkowitz, Lee, René M. Stulz, and Rohan Williamson, “Is there a U.S. High Cash Holdings Puzzle
after the Financial Crisis?,” 2014.
Riddick, Leigh A. and Toni Whited, “The Corporate Propensity to Save,” Journal of Finance, 2009,
64 (4), 1729–1766.
Rosen, Sherwin, “The Economics of Superstars,” American Economic Review, 1981, 71 (5), 845–858.
Sattinger, Michael, “Assignment Models of the Distribution of Earnings,” Journal of Economic Lit-
erature, 1993, 31 (2), 831–880.
Scheuer, Florian and Iván Werning, “The Taxation of Superstars,” Quarterly Journal of Economics,
2017, 132 (1), 211–270.
Tervio, Marko, “The Difference That CEOs Make: An Assignment Model Approach,” American
Economic Review, 2008, 98 (3), 642–668.
42

ONLINE APPENDIX
A. Proof
Proof of Lemma 1
Proof. We first investigate the convexity conditions for corporate earnings π. By using the
envelope theorem, we know that
∂p ∗
π ′ (z) = y >0 (A1)
∂z
Further differentiation gives the following equation:
∂2 p ∗ ∂p ∂y∗
π ′′ (z) = y + (A2)
∂z2 ∂z ∂z
Differentiating Equation (2) to z leads to the equation below:
∂y∗ ∂p ∂2 p ∗ ∂p ∂p ∂y∗ 2
∗ ∂ p ∂y
∗
′′ ∗ ∂y
∗
+ y + + + y − C ( y ) =0 (A3)
∂z ∂y∗ ∂y∗ ∂z ∂z ∂y∗ ∂z ∂y∗2 ∂z ∂z
Then we know that
∂2 p∗ ∂p
∂y∗ ∂y∗ ∂z y + ∂z
= ∂2 p ∂p
(A4)
∂z C ′′ (y∗ ) − y∗ ∂y∗2 − 2 ∂y∗
∂2 p
The optimization of y∗ requires that C ′′ (y∗ ) − y∗ ∂y∗2 − 2 ∂y∗ > 0.
∂p
∂y∗
is sufficiently large and/or C ′′ (y∗ ) is sufficiently low, we have
∂p
Therefore, if ∂z ∂z >0
and π ′′ (z) > 0. In this way, corporate earnings π become a convex function of z.
We can obtain similar results for markup π. By using the envelope theorem, we know
that
A1

∂p y∗
µ′ (z) = >0 (A5)
∂z C (y∗ )
Further differentiation gives the following equation:
∂2 p ∗ ∂y∗
h i
∂z2
y + ∂p
∂z − C ′ (y∗ ) ∂z
µ′′ (z) = (A6)
C ( y ∗ )2
∂y∗
is sufficiently large and/or C ′′ (y∗ ) is sufficiently low, we have
∂p
Similarly, if ∂z ∂z >0
and µ′′ (z) > 0. In this way, corporate earnings µ also become a convex function of z.
Proof of Lemma 2
Proof. It is easy to show E (π (z) |z) is increasing in z as π (z) is a convex function of
z, so any linear combination of a convex function is still a convex function. Therefore,
E (π (z) |z) is also a convex function of z, which means that E (π (z) |z) is increasing in z.
As for the volatility of earnings, we have Var (π (z) |z) = E [π (z) |z]2 − [ E (π (z) |z)]2 .
Therefore, we have
∂Var (π (z) |z)

= 2E π (z) π ′ (z) |z − 2E [π (z) |z] E π ′ (z) |z

∂z ˆ
= 2 π (z) π ′ (z) − E π ′ (z) f (z) dz

(A7)
´
As π (z) is a convex transformation of z, therefore we have π (z) (π ′ (z) − E (π ′ (z))) f (z) dz >
∂Var(π (z)|z)
0, which gives ∂z > 0. As a result, Var (π (z) |z) is also increasing in z.
A2

B. Additional Figures and Tables
Figure A1: Changes in Median Corporate Cash/Net Debt at Different Tobin’s Q Quintiles
of Tobin’s Q. The firm-level cash-to-asset ratio is simply calculated as the ratio of cash and cash equivalents
(Compustat series DLTT) and current liabilities (Compustat series DLC). We measure Tobin’s Q as asset mar-
ket value divided by its book value. In terms of the market value of assets, we compute it as the book value
of assets (Compustat series AT) plus the market value of common stocks (Compustat series PRCC_C times
Compustat series CSHO) minus the book value of equity, where the book value of equity is estimated as the
sum of shareholder equity (Compustat series SEQ), deferred taxes (Compustat series TXDB), and investment
tax credit (Compustat series ITCB), minus the value of preferred stocks (coalesce outcomes of Compustat
series PSTKRV, PSTKL, and PSTK). Data is obtained from Compustat.
A3

Figure A2: Changes in Median Corporate Cash or Net Debt at Different Total Q Quintiles
of total Q. The firm-level cash-to-asset ratio is simply calculated as the ratio of cash and cash equivalents
(Compustat series DLTT) and current liabilities (Compustat series DLC). Peters and Taylor (2017)’s total Q is
measured as the ratio between a firm’s market value over the replacement cost of its intangible capital. The
market value of assets is computed as the book value of assets (Compustat series AT) plus the market value
of common stocks (Compustat series PRCC_C times Compustat series CSHO) minus the book value of equity,
where the book value of equity is estimated as the sum of shareholder equity (Compustat series SEQ), de-
ferred taxes (Compustat series TXDB), and investment tax credit (Compustat series ITCB), minus the value of
preferred stocks (coalesce outcomes of Compustat series PSTKRV, PSTKL, and PSTK). Meanwhile, the intan-
gible capital replacement cost is estimated by accumulating past investments in Research and Development
(Compustat series XRD) and Selling, General, and Administrative Expenses (Compustat series XSGA). Data
is obtained from Compustat.
A4

Figure A3: Binscatter Plot for Q, Cash, and Net Debt
(a) Tobin’s Q (b) Peters and Taylor (2017)’s Total Q
Notes: This figure presents the Binscatter plot for corporate Q or total Q and cash-to-asset ratio or net-debt-
to-asset ratio. The firm-level cash-to-asset ratio is simply calculated as the ratio of cash and cash equivalents
series PSTKRV, PSTKL, and PSTK). Meanwhile Peters and Taylor (2017)’s total Q is measured as the ratio
between a firm’s market value over the replacement cost of its intangible capital, and the intangible capital
replacement cost is estimated by accumulating past investments in Research and Development (Compustat
series XRD) and Selling, General, and Administrative Expenses (Compustat series XSGA). Data is obtained
from Compustat.
A5

Figure A4: Binscatter Plot for Alternative Stories: Tobin’s Q
(c) Multinational (d) Domestic
Notes: This figure presents the Binscatter plot for corporate Tobin’s Q and cash-to-asset ratio or net-debt-to-
series PSTKRV, PSTKL, and PSTK). We define a firm as a multinational one if its reported value for foreign
income and foreign sales are positive and non-missing. Otherwise, we treat the firm as a domestic one. We
define an R&D-intensive firm is the one belonging to an industry whose average R&D investment-to-asset
ratio is equal to or above 2%. Otherwise, we label the firm as R&D-non-intensive. All the industries are
classified at 3-digit SIC level. Data is obtained from Compustat.
A6

Figure A5: Binscatter Plot for Total Q, Cash, and Net Debt
(c) Multinational (d) Domestic
Notes: This figure presents the Binscatter plot for corporate total Q and cash-to-asset ratio or net-debt-to-
(Compustat series DLTT) and current liabilities (Compustat series DLC). Peters and Taylor (2017)’s total Q is
measured as the ratio between a firm’s market value over the replacement cost of its intangible capital. The
market value of assets is computed as the book value of assets (Compustat series AT) plus the market value
of common stocks (Compustat series PRCC_C times Compustat series CSHO) minus the book value of equity,
where the book value of equity is estimated as the sum of shareholder equity (Compustat series SEQ), de-
ferred taxes (Compustat series TXDB), and investment tax credit (Compustat series ITCB), minus the value of
preferred stocks (coalesce outcomes of Compustat series PSTKRV, PSTKL, and PSTK). Meanwhile, the intan-
gible capital replacement cost is estimated by accumulating past investments in Research and Development
(Compustat series XRD) and Selling, General, and Administrative Expenses (Compustat series XSGA). We
define a firm as a multinational one if its reported value for foreign income and foreign sales are positive and
non-missing. Otherwise, we treat the firm as a domestic one. We define an R&D-intensive firm is the one
belonging to an industry whose average R&D investment-to-asset ratio is equal to or above 2%. Otherwise,
we label the firm as R&D-non-intensive. All the industries are classified at 3-digit SIC level. Data is obtained
from Compustat.
A7

Table A1: Reduced-form Evidence for Tobin’s Q, Cash, and Net Debt
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
log Q 0.0552*** 0.0585*** 0.0556*** 0.0476*** 0.0561*** -0.0937*** -0.0897*** -0.0997*** -0.1131*** -0.1203***
(67.990) (47.989) (21.842) (16.485) (19.465) (-82.023) (-55.489) (-29.528) (-30.305) (-32.243)
(log Q)2 0.0023 0.0303*** 0.0258*** 0.0080*** -0.0145*** -0.0107***
(1.295) (14.190) (12.134) (3.368) (-5.247) (-3.895)
Return of Asset 0.0073*** 0.0106*** 0.0106*** 0.0467*** 0.0468*** -0.0103*** -0.0130*** -0.0131*** -0.0454*** -0.0452***
(5.278) (5.170) (5.154) (17.948) (18.177) (-5.320) (-4.756) (-4.797) (-13.493) (-13.569)
Profitability -0.0254*** -0.0531*** -0.0530*** -0.1265*** -0.1255*** 0.0570*** 0.0554*** 0.0560*** 0.1439*** 0.1424***
(-12.525) (-15.891) (-15.824) (-35.137) (-35.233) (20.007) (12.471) (12.601) (30.931) (30.875)
Investment -0.0406*** -0.0356*** -0.0351*** 0.0029 0.0049 0.0497*** 0.0188* 0.0207* 0.0028 -0.0061
(-9.924) (-4.467) (-4.392) (0.313) (0.523) (8.634) (1.778) (1.956) (0.235) (-0.501)
Tangibility -0.3001*** -0.3973*** -0.3977*** -0.3091*** -0.3048*** 0.3969*** 0.4833*** 0.4818*** 0.3886*** 0.3882***
(-112.316) (-79.950) (-79.863) (-74.422) (-72.879) (105.627) (73.273) (72.900) (72.406) (71.694)
Book Leverage -0.1098*** -0.1239*** -0.1239*** -0.1836*** -0.1810*** 0.7189*** 0.7228*** 0.7229*** 0.7810*** 0.7770***
(-96.806) (-72.200) (-72.168) (-109.874) (-108.774) (450.581) (317.286) (317.319) (361.617) (360.684)
Payout 0.0375*** 0.0477*** 0.0477*** 0.0276*** 0.0228*** -0.0834*** -0.1088*** -0.1090*** -0.1626*** -0.1543***
(9.968) (8.773) (8.761) (3.980) (3.324) (-15.749) (-15.072) (-15.103) (-18.156) (-17.364)
Size -0.0127*** -0.0113*** -0.0113*** -0.0046*** -0.0046*** 0.0422*** 0.0426*** 0.0426*** 0.0041*** 0.0042***
(-31.087) (-16.861) (-16.862) (-16.472) (-16.465) (73.692) (47.676) (47.677) (11.386) (11.710)
R&D -0.0850*** -0.0850*** 0.2252*** 0.2056*** -0.0497*** -0.0498*** -0.3839*** -0.3662***
(-14.356) (-14.357) (40.041) (36.773) (-6.328) (-6.331) (-52.813) (-50.608)
Fixed effects
N 165,284 84,077 84,077 85,804 85,804 165,284 84,077 84,077 85,804 85,804
Adjusted R2 0.725 0.752 0.752 0.469 0.483 0.865 0.866 0.866 0.726 0.732
Notes: This table presents the association between Tobin’s Q and corporate cash or net debt with different fixed-effect model
specifications. The dependent variables are corporate Tobin’s Q. We measure Tobin’s Q as asset market value divided by
its book value. In terms of the market value of assets, we compute it as the book value of assets (Compustat series AT)
plus the market value of common stocks (Compustat series PRCC_C times Compustat series CSHO) minus the book value
of equity, where the book value of equity is estimated as the sum of shareholder equity (Compustat series SEQ), deferred
taxes (Compustat series TXDB), and investment tax credit (Compustat series ITCB), minus the value of preferred stocks
(coalesce outcomes of Compustat series PSTKRV, PSTKL, and PSTK). The firm-level cash-to-asset ratio is simply calculated
as the ratio of cash and cash equivalents (Compustat series CHE) to total assets (Compustat series AT). Meanwhile, net
debt is calculated as the difference between the debt-to-asset ratio and the cash-to-asset ratio, where debt is the sum of
long-term debt (Compustat series DLTT) and current liabilities (Compustat series DLC). Size is the natural logarithm of
total assets (Compustat series AT); return of the asset is calculated as the income before extraordinary items (Compustat
series IB) scaled by total assets; asset tangibility is measured as the ratio of physical assets (Compustat series PPENT)
to total assets; book leverage is computed as the ratio of total debts to the sum of total debts and common equity, i.e.,
DLTT+DLC
book leverage = DLTT +DLC+CEQ , where DLTT, DLC, and CEQ represent the Compustat data series on long-term debt,
current liabilities, and common equity, respectively; dividend payout is measured as dividends (Compustat series DVC)
scaled by total assets; and finally, corporate investment A8rate is computed as the share of capital expenditures (Compustat
series CAPX) in total assets. Data used in this table is at firm-year frequency, and obtained from Compustat. T-statistics are
in parentheses. *, **, and *** represent results significant at the 10%, 5%, and 1% levels, respectively. Standard errors are
clustered at the firm level.

Table A2: Reduced-form Evidence for Total Q, Cash, and Net Debt
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
log Total Q 0.0148*** 0.0161*** 0.0212*** 0.0297*** 0.0313*** -0.0031** -0.0040* -0.0140*** -0.0163*** -0.0172***
(52.785) (39.272) (45.786) (57.533) (61.025) (-2.051) (-1.941) (-5.956) (-7.292) (-7.619)
(log Total Q)2 0.0018*** 0.0028*** 0.0028*** -0.0034*** -0.0043*** -0.0042***
(23.389) (27.171) (27.576) (-8.941) (-9.722) (-9.651)
Return of Asset -0.0003*** 0.0010*** 0.0010*** 0.0033*** 0.0033*** 0.0453*** -0.0060*** -0.0061*** -0.0278*** -0.0282***
(-4.525) (2.851) (2.938) (7.215) (7.357) (119.346) (-3.520) (-3.551) (-14.102) (-14.338)
Profitability 0.0008*** -0.0011** -0.0011** 0.0013** 0.0012** -0.1094*** -0.0579*** -0.0578*** -0.0617*** -0.0612***
(4.930) (-2.328) (-2.431) (2.125) (2.011) (-126.493) (-24.689) (-24.664) (-24.046) (-23.883)
Investment -0.0015 0.0318*** 0.0224*** 0.0751*** 0.0793*** -0.1102*** -0.4934*** -0.4753*** -0.6733*** -0.6885***
(-0.499) (4.091) (2.896) (7.994) (8.511) (-6.962) (-12.556) (-12.086) (-16.528) (-16.843)
Tangibility -0.3403*** -0.4573*** -0.4479*** -0.3993*** -0.3950*** 0.6296*** 0.8473*** 0.8290*** 0.8596*** 0.8794***
(-128.184) (-92.396) (-90.496) (-92.321) (-90.984) (44.269) (33.811) (32.988) (45.848) (46.153)
Book Leverage -0.0003*** -0.0002 -0.0001 -0.0016*** -0.0015*** 0.0461*** 0.0520*** 0.0520*** 0.0613*** 0.0609***
(-4.604) (-1.416) (-1.352) (-11.550) (-10.923) (122.165) (92.991) (93.010) (99.599) (99.054)
Payout 0.0070*** 0.0116*** 0.0112*** 0.0048* 0.0038 -0.2246*** -0.3681*** -0.3674*** -0.4320*** -0.4282***
(4.500) (5.132) (4.984) (1.664) (1.338) (-26.872) (-32.168) (-32.120) (-34.688) (-34.443)
Size -0.0102*** -0.0050*** -0.0050*** -0.0089*** -0.0088*** -0.1184*** -0.1489*** -0.1490*** -0.0399*** -0.0398***
(-26.157) (-8.076) (-8.015) (-33.384) (-33.144) (-56.713) (-47.171) (-47.227) (-34.622) (-34.177)
R&D -0.0092*** -0.0088*** 0.0405*** 0.0377*** -0.0820*** -0.0828*** -0.1053*** -0.1015***
(-6.249) (-5.982) (22.928) (21.583) (-10.943) (-11.057) (-13.758) (-13.236)
Fixed effects
N 172,733 90,797 90,797 92,471 92,471 172,733 90,797 90,797 92,471 92,471
Adjusted R2 0.701 0.722 0.724 0.374 0.393 0.525 0.560 0.561 0.279 0.283
Notes: This table presents the association between total Q and corporate cash or net debt with different fixed-effect model
specifications. The dependent variables are corporate total Q. Peters and Taylor (2017)’s total Q is measured as the ratio
between a firm’s market value over the replacement cost of its intangible capital. The market value of assets is computed
as the book value of assets (Compustat series AT) plus the market value of common stocks (Compustat series PRCC_C
times Compustat series CSHO) minus the book value of equity, where the book value of equity is estimated as the sum
of shareholder equity (Compustat series SEQ), deferred taxes (Compustat series TXDB), and investment tax credit (Com-
pustat series ITCB), minus the value of preferred stocks (coalesce outcomes of Compustat series PSTKRV, PSTKL, and
PSTK). Meanwhile, the intangible capital replacement cost is estimated by accumulating past investments in Research and
Development (Compustat series XRD) and Selling, General, and Administrative Expenses (Compustat series XSGA). The
firm-level cash-to-asset ratio is simply calculated as the ratio of cash and cash equivalents (Compustat series CHE) to total
assets (Compustat series AT). Meanwhile, net debt is calculated as the difference between the debt-to-asset ratio and the
cash-to-asset ratio, where debt is the sum of long-term debt (Compustat series DLTT) and current liabilities (Compustat se-
ries DLC). Size is the natural logarithm of total assets (Compustat series AT); return of the asset is calculated as the income
before extraordinary items (Compustat series IB) scaled by total assets; asset tangibility is measured as the ratio of physical
assets (Compustat series PPENT) to total assets; book leverage is computed as the ratio of total debts to the sum of total
DLTT+A9
DLC
debts and common equity, i.e., book leverage = DLTT +DLC+CEQ , where DLTT, DLC, and CEQ represent the Compustat data
series on long-term debt, current liabilities, and common equity, respectively; dividend payout is measured as dividends
(Compustat series DVC) scaled by total assets; and finally, corporate investment rate is computed as the share of capital
expenditures (Compustat series CAPX) in total assets. Data used in this table is at firm-year frequency, and obtained from
Compustat. T-statistics are in parentheses. *, **, and *** represent results significant at the 10%, 5%, and 1% levels, respec-
tively. Standard errors are clustered at the firm level.

Table A3: Reduced-form Evidence for Markup, Cash, and Net Debt: Multinational Cor-
porations
casha ssetr atio netdebta ssetr atio
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Markup 0.0766*** 0.0852*** 0.0366* 0.0629*** 0.0712*** -0.0481*** -0.0459*** 0.0766** 0.0474* 0.0542*
(26.373) (23.602) (1.718) (3.338) (3.748) (-8.163) (-8.027) (2.269) (1.669) (1.886)
Markup2 0.0140** 0.0135** 0.0113** -0.0354*** -0.0387*** -0.0407***
(2.311) (2.465) (2.047) (-3.681) (-4.674) (-4.869)
Return of Asset -0.0068*** 0.0159*** 0.0159*** 0.0384*** 0.0375*** 0.0535*** -0.0361*** -0.0360*** -0.0618*** -0.0609***
(-4.088) (5.737) (5.731) (11.207) (11.058) (15.921) (-8.234) (-8.226) (-11.930) (-11.870)
Profitability -0.0043** -0.0346*** -0.0346*** -0.0644*** -0.0634*** -0.0221*** 0.1186*** 0.1185*** 0.1515*** 0.1501***
(-1.981) (-10.594) (-10.584) (-15.824) (-15.781) (-5.009) (22.943) (22.929) (24.683) (24.683)
Investment 0.0483*** 0.0912*** 0.0914*** 0.1918*** 0.2172*** -0.1627*** -0.2737*** -0.2740*** -0.3730*** -0.4138***
(4.213) (5.907) (5.916) (10.535) (11.942) (-7.003) (-11.191) (-11.206) (-13.582) (-15.048)
Tangibility -0.4440*** -0.5556*** -0.5545*** -0.5396*** -0.5494*** 0.6206*** 0.7297*** 0.7267*** 0.6706*** 0.6868***
(-69.530) (-62.710) (-62.481) (-68.177) (-68.629) (47.968) (52.014) (51.730) (56.153) (56.732)
Book Leverage -0.0211*** -0.0275*** -0.0276*** -0.0569*** -0.0550*** 0.4423*** 0.5509*** 0.5511*** 0.5779*** 0.5757***
(-22.515) (-21.689) (-21.721) (-41.718) (-40.592) (232.936) (273.954) (274.029) (280.593) (280.986)
Payout 0.0568*** 0.0658*** 0.0658*** 0.1003*** 0.0888*** -0.1487*** -0.2160*** -0.2159*** -0.3315*** -0.3136***
(7.709) (7.732) (7.728) (9.468) (8.449) (-9.960) (-16.027) (-16.022) (-20.744) (-19.736)
Size -0.0016** -0.0012 -0.0012 0.0030*** 0.0028*** 0.0198*** 0.0247*** 0.0248*** -0.0052*** -0.0049***
(-2.170) (-1.212) (-1.271) (6.592) (6.196) (12.979) (16.105) (16.198) (-7.630) (-7.196)
R&D -0.0493*** -0.0497*** 0.1865*** 0.1680*** -0.0764*** -0.0752*** -0.3480*** -0.3334***
(-5.659) (-5.712) (22.467) (20.330) (-5.543) (-5.457) (-27.791) (-26.681)
Fixed effects
N 50,579 37,306 37,306 37,617 37,617 50,579 37,306 37,306 37,617 37,617
Adjusted R2 0.647 0.661 0.661 0.363 0.379 0.796 0.858 0.858 0.756 0.761
specifications. We define a firm as a multinational one if its reported value for foreign income and foreign sales are positive
and non-missing. Otherwise, we treat the firm as a domestic one. The dependent variables are corporate markup, and
we measure it by following De Loecker et al. (2020)’s method. The firm-level cash-to-asset ratio is simply calculated
DLTT+DLC
scaled by total assets; and finally, corporate investment rate is computed as the share of capital expenditures (Compustat
series CAPX) in total assets. Data used in this table is atA10
firm-year frequency, and obtained from Compustat. T-statistics are

Table A4: Reduced-form Evidence for Markup, Cash, and Net Debt: Domestic Corpora-
tions
casha ssetr atio netdebta ssetr atio
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Markup 0.0371*** 0.0549*** 0.1182*** 0.0470*** 0.0561*** -0.1068*** -0.2095*** -0.5819*** -0.3769*** -0.4078***
(17.840) (17.238) (6.599) (2.748) (3.301) (-13.098) (-14.190) (-7.004) (-5.587) (-5.965)
Markup2 -0.0188*** 0.0064 0.0041 0.1102*** 0.0760*** 0.0810***
(-3.593) (1.250) (0.794) (4.555) (3.742) (3.941)
Return of Asset -0.0017*** 0.0006** 0.0006** 0.0007* 0.0009** 0.0141*** 0.0159*** 0.0159*** -0.0039** -0.0040**
(-6.660) (2.122) (2.156) (1.811) (2.214) (14.331) (12.534) (12.495) (-2.435) (-2.499)
Profitability -0.0030*** -0.0022*** -0.0022*** -0.0009** -0.0002 -0.0154*** 0.0010 0.0010 0.0016 0.0001
(-8.092) (-5.017) (-5.051) (-2.097) (-0.379) (-10.470) (0.466) (0.509) (0.898) (0.054)
Investment 0.0807*** 0.0468*** 0.0465*** 0.1420*** 0.1388*** -0.1369*** -0.1767*** -0.1748*** -0.3921*** -0.4006***
(14.483) (4.692) (4.662) (11.354) (11.246) (-6.271) (-3.820) (-3.781) (-7.947) (-8.070)
Tangibility -0.3383*** -0.4852*** -0.4872*** -0.4356*** -0.4283*** 0.6156*** 0.7798*** 0.7917*** 0.8411*** 0.8713***
(-81.314) (-71.288) (-71.353) (-70.160) (-68.162) (37.792) (24.705) (25.003) (34.327) (34.473)
Book Leverage -0.0018*** -0.0002* -0.0002 -0.0013*** -0.0010*** 0.0633*** 0.0482*** 0.0481*** 0.0559*** 0.0552***
(-12.773) (-1.705) (-1.607) (-7.930) (-6.368) (117.613) (80.029) (79.898) (85.741) (83.488)
Payout 0.0021 -0.0012 -0.0012 0.0002 0.0004 -0.0352*** -0.1072*** -0.1072*** 0.1826*** 0.1825***
(0.794) (-0.385) (-0.382) (0.051) (0.102) (-3.398) (-7.185) (-7.190) (10.910) (10.948)
Size -0.0158*** -0.0090*** -0.0090*** -0.0100*** -0.0088*** -0.0760*** -0.1245*** -0.1240*** -0.0327*** -0.0345***
(-25.384) (-9.355) (-9.441) (-23.441) (-20.488) (-31.106) (-28.042) (-27.933) (-19.443) (-20.028)
R&D -0.0209*** -0.0208*** 0.0317*** 0.0297*** 0.1904*** 0.1903*** 0.1064*** 0.1071***
(-8.205) (-8.199) (12.815) (11.636) (16.149) (16.144) (10.916) (10.446)
Fixed effects
N 103,751 41,919 41,919 43,298 43,298 103,751 41,919 41,919 43,298 43,298
Adjusted R2 0.641 0.777 0.777 0.410 0.439 0.574 0.608 0.609 0.277 0.285
and non-missing. Otherwise, we treat the firm as a domestic one. The dependent variables are corporate markup, and
we measure it by following De Loecker et al. (2020)’s method. The firm-level cash-to-asset ratio is simply calculated
DLTT+DLC
series CAPX) in total assets. Data used in this table is atA11
firm-year frequency, and obtained from Compustat. T-statistics are

Table A5: Reduced-form Evidence for Tobin’s Q, Cash, and Net Debt: Multinational Cor-
porations
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
log Q 0.0569*** 0.0614*** 0.0551*** 0.0507*** 0.0597*** -0.0962*** -0.0910*** -0.1066*** -0.1161*** -0.1239***
(37.931) (34.088) (14.013) (11.377) (13.386) (-49.535) (-40.071) (-21.496) (-20.729) (-22.052)
(log Q)2 0.0050* 0.0257*** 0.0206*** 0.0124*** -0.0018 0.0024
(1.795) (7.907) (6.342) (3.538) (-0.434) (0.585)
Return of Asset 0.0204*** 0.0152*** 0.0150*** 0.0566*** 0.0556*** -0.0033 -0.0012 -0.0017 -0.0528*** -0.0508***
(5.977) (3.729) (3.684) (11.844) (11.750) (-0.748) (-0.239) (-0.327) (-8.794) (-8.534)
Profitability -0.0611*** -0.0952*** -0.0946*** -0.1624*** -0.1619*** 0.0828*** 0.0637*** 0.0651*** 0.1635*** 0.1613***
(-13.115) (-16.444) (-16.317) (-25.284) (-25.433) (13.725) (8.731) (8.918) (20.231) (20.136)
Investment -0.0561*** -0.0522*** -0.0511*** 0.0043 0.0292* 0.0553*** 0.0242 0.0270 0.0092 -0.0175
(-5.258) (-3.525) (-3.446) (0.248) (1.690) (4.002) (1.296) (1.444) (0.422) (-0.803)
Tangibility -0.3200*** -0.4114*** -0.4126*** -0.3490*** -0.3514*** 0.4083*** 0.5122*** 0.5092*** 0.4225*** 0.4279***
(-54.157) (-49.702) (-49.688) (-48.917) (-48.970) (53.336) (49.091) (48.660) (47.070) (47.377)
Book Leverage -0.1266*** -0.1342*** -0.1341*** -0.1952*** -0.1914*** 0.7532*** 0.7496*** 0.7499*** 0.7926*** 0.7882***
(-58.863) (-51.102) (-51.054) (-78.000) (-76.496) (270.245) (226.387) (226.447) (251.702) (250.269)
Payout 0.0676*** 0.0774*** 0.0772*** 0.0905*** 0.0758*** -0.1303*** -0.1478*** -0.1481*** -0.2039*** -0.1877***
(8.983) (8.698) (8.680) (8.393) (7.077) (-13.355) (-13.175) (-13.211) (-15.025) (-13.926)
Size -0.0079*** -0.0075*** -0.0076*** -0.0011*** -0.0014*** 0.0387*** 0.0375*** 0.0375*** -0.0013*** -0.0011**
(-10.671) (-7.983) (-7.996) (-2.751) (-3.554) (40.455) (31.530) (31.509) (-2.683) (-2.209)
R&D -0.1603*** -0.1603*** 0.1533*** 0.1298*** -0.0605*** -0.0606*** -0.4133*** -0.3908***
(-15.225) (-15.229) (15.823) (13.446) (-4.558) (-4.566) (-33.897) (-32.163)
Fixed effects
N 53,331 39,399 39,399 39,704 39,704 53,331 39,399 39,399 39,704 39,704
Adjusted R2 0.674 0.687 0.687 0.436 0.450 0.852 0.850 0.850 0.730 0.737
Notes: This table presents the association between Tobin’s Q and corporate cash or net debt with different fixed-effect
model specifications. We define a firm as a multinational one if its reported value for foreign income and foreign sales are
positive and non-missing. Otherwise, we treat the firm as a domestic one. The dependent variables are corporate Tobin’s
Q. We measure Tobin’s Q as asset market value divided by its book value. In terms of the market value of assets, we
compute it as the book value of assets (Compustat series AT) plus the market value of common stocks (Compustat series
PRCC_C times Compustat series CSHO) minus the book value of equity, where the book value of equity is estimated as
the sum of shareholder equity (Compustat series SEQ), deferred taxes (Compustat series TXDB), and investment tax credit
(Compustat series ITCB), minus the value of preferred stocks (coalesce outcomes of Compustat series PSTKRV, PSTKL,
and PSTK). The firm-level cash-to-asset ratio is simply calculated as the ratio of cash and cash equivalents (Compustat
series CHE) to total assets (Compustat series AT). Meanwhile, net debt is calculated as the difference between the debt-
to-asset ratio and the cash-to-asset ratio, where debt is the sum of long-term debt (Compustat series DLTT) and current
liabilities (Compustat series DLC). Size is the natural logarithm of total assets (Compustat series AT); return of the asset
is calculated as the income before extraordinary items (Compustat series IB) scaled by total assets; asset tangibility is
measured as the ratio of physical assets (Compustat series PPENT) to total assets; book leverage is computed as the ratio
A12
of total debts to the sum of total debts and common equity, i.e., book leverage = DLTT DLTT+DLC
+DLC+CEQ , where DLTT, DLC, and
CEQ represent the Compustat data series on long-term debt, current liabilities, and common equity, respectively; dividend
payout is measured as dividends (Compustat series DVC) scaled by total assets; and finally, corporate investment rate is
computed as the share of capital expenditures (Compustat series CAPX) in total assets. Data used in this table is at firm-
year frequency, and obtained from Compustat. T-statistics are in parentheses. *, **, and *** represent results significant at
the 10%, 5%, and 1% levels, respectively. Standard errors are clustered at the firm level.

Table A6: Reduced-form Evidence for Tobin’s Q, Cash, and Net Debt: Domestic Corpora-
tions
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
log Q 0.0546*** 0.0567*** 0.0570*** 0.0530*** 0.0606*** -0.0926*** -0.0882*** -0.0942*** -0.1165*** -0.1229***
(56.832) (34.274) (17.216) (13.981) (15.989) (-65.409) (-38.122) (-20.357) (-23.249) (-24.510)
(log Q)2 -0.0002 0.0282*** 0.0244*** 0.0049 -0.0191*** -0.0160***
(-0.095) (10.007) (8.700) (1.501) (-5.117) (-4.313)
Return of Asset 0.0052*** 0.0093*** 0.0093*** 0.0430*** 0.0427*** -0.0137*** -0.0193*** -0.0193*** -0.0430*** -0.0426***
(3.580) (3.985) (3.986) (13.748) (13.818) (-6.369) (-5.933) (-5.942) (-10.400) (-10.424)
Profitability -0.0143*** -0.0234*** -0.0235*** -0.0998*** -0.0989*** 0.0480*** 0.0450*** 0.0453*** 0.1213*** 0.1201***
(-6.481) (-5.703) (-5.703) (-22.644) (-22.667) (14.755) (7.830) (7.874) (20.840) (20.824)
Investment -0.0366*** -0.0275*** -0.0275*** 0.0064 0.0002 0.0474*** 0.0129 0.0141 -0.0001 -0.0027
(-8.500) (-2.961) (-2.961) (0.563) (0.016) (7.462) (0.994) (1.081) (-0.009) (-0.181)
Tangibility -0.2928*** -0.3846*** -0.3845*** -0.2988*** -0.2940*** 0.3932*** 0.4634*** 0.4627*** 0.3797*** 0.3788***
(-99.612) (-62.539) (-62.446) (-56.783) (-55.036) (90.778) (53.909) (53.756) (54.641) (53.614)
Book Leverage -0.1005*** -0.1137*** -0.1137*** -0.1761*** -0.1750*** 0.7019*** 0.6996*** 0.6997*** 0.7735*** 0.7704***
(-75.886) (-50.484) (-50.480) (-78.347) (-78.049) (359.786) (222.176) (222.179) (260.608) (259.772)
Payout 0.0251*** 0.0289*** 0.0289*** -0.0037 -0.0023 -0.0636*** -0.0820*** -0.0821*** -0.1413*** -0.1389***
(5.860) (4.275) (4.276) (-0.411) (-0.263) (-10.102) (-8.666) (-8.680) (-11.889) (-11.784)
Size -0.0156*** -0.0171*** -0.0171*** -0.0079*** -0.0076*** 0.0447*** 0.0494*** 0.0495*** 0.0097*** 0.0097***
(-32.106) (-17.746) (-17.746) (-19.755) (-18.857) (62.514) (36.749) (36.757) (18.294) (18.033)
R&D -0.0436*** -0.0436*** 0.2488*** 0.2313*** -0.0440*** -0.0440*** -0.3629*** -0.3459***
(-6.206) (-6.206) (35.501) (33.199) (-4.473) (-4.474) (-39.213) (-37.545)
Fixed effects
N 111,953 44,678 44,678 46,091 46,091 111,953 44,678 44,678 46,091 46,091
Adjusted R2 0.750 0.798 0.798 0.509 0.523 0.869 0.877 0.877 0.730 0.738
Notes: This table presents the association between Tobin’s Q and corporate cash or net debt with different fixed-effect
model specifications. We define a firm as a multinational one if its reported value for foreign income and foreign sales are
positive and non-missing. Otherwise, we treat the firm as a domestic one. The dependent variables are corporate Tobin’s
Q. We measure Tobin’s Q as asset market value divided by its book value. In terms of the market value of assets, we
compute it as the book value of assets (Compustat series AT) plus the market value of common stocks (Compustat series
(Compustat series ITCB), minus the value of preferred stocks (coalesce outcomes of Compustat series PSTKRV, PSTKL,
and PSTK). The firm-level cash-to-asset ratio is simply calculated as the ratio of cash and cash equivalents (Compustat
series CHE) to total assets (Compustat series AT). Meanwhile, net debt is calculated as the difference between the debt-
to-asset ratio and the cash-to-asset ratio, where debt is the sum of long-term debt (Compustat series DLTT) and current
liabilities (Compustat series DLC). Size is the natural logarithm of total assets (Compustat series AT); return of the asset
is calculated as the income before extraordinary items (Compustat series IB) scaled by total assets; asset tangibility is
measured as the ratio of physical assets (Compustat series PPENT) to total assets; book leverage is computed as the ratio
A13
of total debts to the sum of total debts and common equity, i.e., book leverage = DLTT DLTT+DLC
+DLC+CEQ , where DLTT, DLC, and
CEQ represent the Compustat data series on long-term debt, current liabilities, and common equity, respectively; dividend
payout is measured as dividends (Compustat series DVC) scaled by total assets; and finally, corporate investment rate is
computed as the share of capital expenditures (Compustat series CAPX) in total assets. Data used in this table is at firm-
year frequency, and obtained from Compustat. T-statistics are in parentheses. *, **, and *** represent results significant at
the 10%, 5%, and 1% levels, respectively. Standard errors are clustered at the firm level.

Table A7: Reduced-form Evidence for Total Q, Cash, and Net Debt: Multinational Corpo-
rations
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
log Total Q 0.0147*** 0.0155*** 0.0197*** 0.0294*** 0.0308*** -0.0067*** -0.0062*** -0.0104*** -0.0172*** -0.0186***
(31.016) (27.437) (31.087) (43.180) (45.414) (-5.133) (-5.050) (-7.519) (-12.869) (-13.848)
(log Total Q)2 0.0012*** 0.0019*** 0.0019*** -0.0012*** -0.0019*** -0.0019***
(14.474) (18.600) (18.910) (-6.601) (-9.632) (-9.699)
Return of Asset 0.0019 0.0189*** 0.0190*** 0.0422*** 0.0411*** -0.0648*** -0.2246*** -0.2246*** -0.2563*** -0.2547***
(1.620) (9.201) (9.246) (16.573) (16.360) (-20.291) (-50.182) (-50.219) (-51.175) (-51.135)
Profitability -0.0095*** -0.0339*** -0.0340*** -0.0600*** -0.0589*** 0.0896*** 0.2892*** 0.2893*** 0.3276*** 0.3263***
(-5.652) (-12.976) (-13.059) (-18.606) (-18.531) (19.427) (50.876) (50.926) (51.731) (51.803)
Investment -0.0068 0.0073 -0.0044 0.0709*** 0.0964*** -0.1724*** -0.2985*** -0.2869*** -0.4330*** -0.4703***
(-0.631) (0.494) (-0.301) (4.041) (5.534) (-5.847) (-9.314) (-8.942) (-12.566) (-13.624)
Tangibility -0.3797*** -0.4751*** -0.4657*** -0.4260*** -0.4267*** 0.5585*** 0.7133*** 0.7040*** 0.6625*** 0.6687***
(-63.609) (-57.193) (-56.053) (-58.242) (-58.187) (34.046) (39.439) (38.829) (46.093) (46.020)
Book Leverage -0.0104*** -0.0176*** -0.0175*** -0.0351*** -0.0341*** 0.3472*** 0.4254*** 0.4252*** 0.4590*** 0.4576***
(-19.809) (-22.280) (-22.177) (-39.906) (-39.219) (240.327) (246.953) (247.002) (265.251) (265.460)
Payout 0.0622*** 0.0812*** 0.0809*** 0.1013*** 0.0843*** -0.3998*** -0.5854*** -0.5851*** -0.7157*** -0.6937***
(8.874) (9.882) (9.874) (10.013) (8.419) (-20.745) (-32.732) (-32.734) (-36.001) (-34.950)
Size -0.0070*** -0.0069*** -0.0069*** -0.0064*** -0.0068*** 0.0066*** 0.0296*** 0.0296*** 0.0057*** 0.0061***
(-9.668) (-7.589) (-7.590) (-15.977) (-17.065) (3.345) (14.880) (14.879) (7.273) (7.761)
R&D -0.0248*** -0.0238*** 0.1350*** 0.1235*** 0.1876*** 0.1866*** -0.0615*** -0.0468***
(-4.664) (-4.476) (23.549) (21.768) (16.173) (16.089) (-5.462) (-4.158)
Fixed effects
N 55,665 41,526 41,526 41,806 41,806 55,665 41,526 41,526 41,806 41,806
Adjusted R2 0.648 0.660 0.662 0.370 0.389 0.758 0.810 0.810 0.711 0.714
and non-missing. Otherwise, we treat the firm as a domestic one. The dependent variables are corporate total Q. Peters and
Taylor (2017)’s total Q is measured as the ratio between a firm’s market value over the replacement cost of its intangible
capital. The market value of assets is computed as the book value of assets (Compustat series AT) plus the market value
of common stocks (Compustat series PRCC_C times Compustat series CSHO) minus the book value of equity, where the
book value of equity is estimated as the sum of shareholder equity (Compustat series SEQ), deferred taxes (Compustat se-
ries TXDB), and investment tax credit (Compustat series ITCB), minus the value of preferred stocks (coalesce outcomes of
Compustat series PSTKRV, PSTKL, and PSTK). Meanwhile, the intangible capital replacement cost is estimated by accumu-
lating past investments in Research and Development (Compustat series XRD) and Selling, General, and Administrative
Expenses (Compustat series XSGA). The firm-level cash-to-asset ratio is simply calculated as the ratio of cash and cash
equivalents (Compustat series CHE) to total assets (Compustat series AT). Meanwhile, net debt is calculated as the differ-
ence between the debt-to-asset ratio and the cash-to-asset ratio, where debt is the sum of long-term debt (Compustat series
DLTT) and current liabilities (Compustat series DLC). Size is the natural logarithm of total assets (Compustat series AT);
A14
return of the asset is calculated as the income before extraordinary items (Compustat series IB) scaled by total assets; asset
tangibility is measured as the ratio of physical assets (Compustat series PPENT) to total assets; book leverage is computed
DLTT+DLC
as the ratio of total debts to the sum of total debts and common equity, i.e., book leverage = DLTT +DLC+CEQ , where DLTT,
DLC, and CEQ represent the Compustat data series on long-term debt, current liabilities, and common equity, respectively;
dividend payout is measured as dividends (Compustat series DVC) scaled by total assets; and finally, corporate investment
rate is computed as the share of capital expenditures (Compustat series CAPX) in total assets. Data used in this table is at
firm-year frequency, and obtained from Compustat. T-statistics are in parentheses. *, **, and *** represent results significant
at the 10%, 5%, and 1% levels, respectively. Standard errors are clustered at the firm level.

Table A8: Reduced-form Evidence for Total Q, Cash, and Net Debt: Domestic Corpora-
tions
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
log Total Q 0.0146*** 0.0165*** 0.0281*** 0.0350*** 0.0366*** 0.0011 0.0029 -0.0274*** -0.0235*** -0.0248***
(41.650) (27.800) (37.975) (42.667) (44.688) (0.520) (0.798) (-5.923) (-5.667) (-5.924)
(log Total Q)2 0.0052*** 0.0073*** 0.0072*** -0.0136*** -0.0140*** -0.0140***
(25.913) (27.403) (27.525) (-10.884) (-10.401) (-10.399)
Return of Asset -0.0001 0.0011*** 0.0011*** 0.0033*** 0.0033*** 0.0398*** -0.0096*** -0.0098*** -0.0328*** -0.0339***
(-1.393) (3.154) (3.373) (6.939) (7.017) (95.033) (-4.565) (-4.653) (-13.489) (-13.971)
Profitability 0.0008*** -0.0003 -0.0004 0.0019*** 0.0018*** -0.0976*** -0.0488*** -0.0486*** -0.0546*** -0.0537***
(4.773) (-0.674) (-0.851) (2.969) (2.874) (-101.080) (-16.699) (-16.649) (-17.168) (-16.942)
Investment -0.0012 0.0327*** 0.0167* 0.0638*** 0.0605*** -0.0961*** -0.4198*** -0.3778*** -0.6203*** -0.6247***
(-0.405) (3.604) (1.847) (5.671) (5.410) (-5.328) (-7.462) (-6.709) (-10.888) (-10.916)
Tangibility -0.3262*** -0.4382*** -0.4181*** -0.3747*** -0.3714*** 0.6251*** 0.7787*** 0.7261*** 0.8142*** 0.8523***
(-111.079) (-71.128) (-67.862) (-67.802) (-66.389) (35.127) (20.377) (18.877) (29.111) (29.769)
Book Leverage -0.0001 0.0002** 0.0002** -0.0008*** -0.0008*** 0.0407*** 0.0461*** 0.0461*** 0.0543*** 0.0536***
(-1.470) (1.978) (2.199) (-5.608) (-5.294) (97.546) (66.544) (66.547) (70.997) (70.174)
Payout 0.0024 0.0037 0.0029 -0.0037 -0.0037 -0.1828*** -0.3083*** -0.3062*** -0.3503*** -0.3422***
(1.565) (1.576) (1.235) (-1.208) (-1.201) (-19.518) (-21.319) (-21.198) (-22.304) (-21.844)
Size -0.0131*** -0.0073*** -0.0067*** -0.0096*** -0.0092*** -0.1374*** -0.2226*** -0.2242*** -0.0585*** -0.0587***
(-28.263) (-8.356) (-7.733) (-26.035) (-24.752) (-48.829) (-40.872) (-41.202) (-31.287) (-30.889)
R&D -0.0068*** -0.0062*** 0.0323*** 0.0304*** -0.1128*** -0.1144*** -0.1316*** -0.1314***
(-4.466) (-4.096) (16.790) (15.990) (-11.881) (-12.066) (-13.532) (-13.504)
Fixed effects
N 117,068 49,271 49,271 50,656 50,656 117,068 49,271 49,271 50,656 50,656
Adjusted R2 0.728 0.764 0.768 0.414 0.432 0.528 0.563 0.564 0.281 0.287
and non-missing. Otherwise, we treat the firm as a domestic one. The dependent variables are corporate total Q. Peters and
Taylor (2017)’s total Q is measured as the ratio between a firm’s market value over the replacement cost of its intangible
capital. The market value of assets is computed as the book value of assets (Compustat series AT) plus the market value
of common stocks (Compustat series PRCC_C times Compustat series CSHO) minus the book value of equity, where the
book value of equity is estimated as the sum of shareholder equity (Compustat series SEQ), deferred taxes (Compustat se-
ries TXDB), and investment tax credit (Compustat series ITCB), minus the value of preferred stocks (coalesce outcomes of
Compustat series PSTKRV, PSTKL, and PSTK). Meanwhile, the intangible capital replacement cost is estimated by accumu-
lating past investments in Research and Development (Compustat series XRD) and Selling, General, and Administrative
Expenses (Compustat series XSGA). The firm-level cash-to-asset ratio is simply calculated as the ratio of cash and cash
equivalents (Compustat series CHE) to total assets (Compustat series AT). Meanwhile, net debt is calculated as the differ-
ence between the debt-to-asset ratio and the cash-to-asset ratio, where debt is the sum of long-term debt (Compustat series
DLTT) and current liabilities (Compustat series DLC). Size is the natural logarithm of total assets (Compustat series AT);
A15
return of the asset is calculated as the income before extraordinary items (Compustat series IB) scaled by total assets; asset
tangibility is measured as the ratio of physical assets (Compustat series PPENT) to total assets; book leverage is computed
DLTT+DLC
as the ratio of total debts to the sum of total debts and common equity, i.e., book leverage = DLTT +DLC+CEQ , where DLTT,
DLC, and CEQ represent the Compustat data series on long-term debt, current liabilities, and common equity, respectively;
dividend payout is measured as dividends (Compustat series DVC) scaled by total assets; and finally, corporate investment
rate is computed as the share of capital expenditures (Compustat series CAPX) in total assets. Data used in this table is at
firm-year frequency, and obtained from Compustat. T-statistics are in parentheses. *, **, and *** represent results significant
at the 10%, 5%, and 1% levels, respectively. Standard errors are clustered at the firm level.

Table A9: Reduced-form Evidence for Markup, Cash, and Net Debt: R&D-intensive firms
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Markup 0.0651*** 0.0802*** 0.0724*** 0.0430*** 0.0577*** -0.1624*** -0.1931*** -0.4738*** -0.3071*** -0.3200***
(25.177) (27.411) (4.239) (2.722) (3.676) (-16.243) (-16.884) (-7.106) (-5.617) (-5.865)
Markup2 0.0023 0.0156*** 0.0115** 0.0810*** 0.0395** 0.0429***
(0.469) (3.371) (2.496) (4.273) (2.459) (2.680)
Return of Asset -0.0013*** 0.0006** 0.0006** 0.0006 0.0007* 0.0102*** 0.0170*** 0.0170*** -0.0020 -0.0025*
(-4.063) (1.981) (1.978) (1.508) (1.707) (8.341) (14.166) (14.139) (-1.329) (-1.717)
Profitability -0.0034*** -0.0030*** -0.0030*** 0.0014** 0.0009 -0.0264*** -0.0092*** -0.0091*** -0.0070*** -0.0057***
(-7.252) (-6.147) (-6.143) (2.208) (1.466) (-14.702) (-4.815) (-4.782) (-3.211) (-2.651)
Investment 0.1606*** 0.1367*** 0.1367*** 0.2952*** 0.3049*** -0.2345*** -0.4273*** -0.4260*** -0.6590*** -0.6695***
(18.353) (10.843) (10.846) (19.416) (20.248) (-6.932) (-8.676) (-8.652) (-12.524) (-12.784)
Tangibility -0.4859*** -0.6331*** -0.6329*** -0.6260*** -0.6190*** 0.7898*** 1.0274*** 1.0337*** 1.2159*** 1.2158***
(-78.152) (-80.478) (-80.364) (-88.241) (-87.609) (32.865) (33.427) (33.598) (49.527) (49.474)
Book Leverage -0.0014*** -0.0002* -0.0002* -0.0013*** -0.0011*** 0.0608*** 0.0497*** 0.0497*** 0.0572*** 0.0563***
(-8.660) (-1.673) (-1.683) (-6.966) (-6.324) (99.973) (87.943) (87.845) (91.425) (90.943)
Payout 0.0058* 0.0068* 0.0068* 0.0056 0.0056 -0.0610*** -0.1140*** -0.1141*** 0.1546*** 0.1552***
(1.884) (1.919) (1.918) (1.282) (1.287) (-5.158) (-8.270) (-8.285) (10.167) (10.319)
Size -0.0058*** -0.0016* -0.0016* -0.0034*** -0.0026*** -0.1065*** -0.1018*** -0.1018*** -0.0350*** -0.0359***
(-7.783) (-1.879) (-1.878) (-8.890) (-6.611) (-36.882) (-30.857) (-30.853) (-26.095) (-26.709)
R&D -0.0231*** -0.0231*** 0.0534*** 0.0474*** 0.2112*** 0.2108*** 0.0988*** 0.1103***
(-8.261) (-8.265) (18.628) (16.720) (19.320) (19.285) (9.954) (11.192)
Fixed effects
N 82,803 54,972 54,972 56,605 56,605 82,803 54,972 54,972 56,605 56,605
Adjusted R2 0.644 0.725 0.725 0.353 0.374 0.563 0.592 0.592 0.277 0.293
specifications. We define an R&D-intensive firm is the one belonging to an industry whose average R&D investment-to-
asset ratio is equal to or above 2%. Otherwise, we label the firm as R&D-non-intensive. All the industries are classified at
3-digit SIC level. The dependent variables are corporate markup, and we measure it by following De Loecker et al. (2020)’s
DLTT+DLC
the Compustat data series on long-term debt, current liabilities, and common equity, respectively; dividend payout is
measured as dividends (Compustat series DVC) scaled by total assets; and finally, corporate investment rate is computed as
the share of capital expenditures (Compustat series CAPX) in total assets. Data used in this table is at firm-year frequency,
and obtained from Compustat. T-statistics are in parentheses. *, **, and *** represent results significant at the 10%, 5%, and
1% levels, respectively. Standard errors are clustered at A16
the firm level.

Table A10: Reduced-form Evidence for Markup, Cash, and Net Debt: non-R&D-intensive
firms
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Markup 0.0342*** 0.0466*** 0.0848*** 0.0501** 0.0556*** -0.0588*** -0.0392*** 0.0439 0.2511*** 0.2310***
(17.435) (10.645) (3.472) (2.428) (2.640) (-10.082) (-4.573) (0.919) (5.990) (5.409)
Markup2 -0.0120 -0.0031 -0.0045 -0.0261* -0.0839*** -0.0785***
(-1.588) (-0.472) (-0.681) (-1.768) (-6.317) (-5.817)
Return of Assets -0.0079*** 0.0027 0.0025 -0.0114*** -0.0118*** 0.1674*** 0.0595*** 0.0589*** 0.1147*** 0.1210***
(-12.822) (0.765) (0.687) (-2.790) (-2.911) (91.259) (8.496) (8.400) (13.856) (14.729)
Profitability 0.0039*** -0.0060 -0.0059 0.0098** 0.0099** -0.0573*** 0.0183* 0.0185* -0.1033*** -0.1088***
(5.906) (-1.221) (-1.203) (2.430) (2.459) (-29.045) (1.908) (1.928) (-12.561) (-13.271)
Investment 0.0029 0.0164* 0.0162* 0.0734*** 0.0675*** -0.0874*** -0.0939*** -0.0944*** -0.1953*** -0.1903***
(0.601) (1.726) (1.706) (6.867) (6.343) (-6.020) (-5.068) (-5.090) (-9.000) (-8.812)
Tangibility -0.2842*** -0.3666*** -0.3678*** -0.2574*** -0.2612*** 0.5589*** 0.4758*** 0.4733*** 0.3520*** 0.3635***
(-85.472) (-56.832) (-56.657) (-49.675) (-49.490) (56.489) (37.700) (37.270) (33.468) (33.955)
Book Leverage -0.0075*** -0.0196*** -0.0196*** -0.0274*** -0.0261*** 0.2064*** 0.5074*** 0.5074*** 0.5127*** 0.5141***
(-15.054) (-12.645) (-12.658) (-21.065) (-19.982) (139.433) (167.096) (167.085) (194.069) (194.226)
Payout 0.0286*** 0.0256*** 0.0257*** 0.0953*** 0.0826*** 0.0697*** -0.0969*** -0.0965*** -0.2514*** -0.2315***
(6.034) (3.430) (3.451) (10.069) (8.770) (4.941) (-6.639) (-6.616) (-13.096) (-12.118)
Size -0.0163*** -0.0099*** -0.0099*** -0.0096*** -0.0096*** 0.0001 0.0368*** 0.0368*** 0.0122*** 0.0121***
(-30.094) (-9.204) (-9.200) (-22.278) (-22.094) (0.062) (17.420) (17.425) (14.031) (13.638)
R&D -0.0026 -0.0027 0.0507*** 0.0479*** 0.0622*** 0.0620*** 0.0038 -0.0046
(-0.300) (-0.311) (6.274) (5.991) (3.598) (3.586) (0.229) (-0.286)
Fixed effects
N 70,012 23,547 23,547 24,312 24,312 70,012 23,547 23,547 24,312 24,312
Adjusted R2 0.644 0.655 0.655 0.262 0.285 0.681 0.856 0.856 0.697 0.707
3-digit SIC level. The dependent variables are corporate markup, and we measure it by following De Loecker et al. (2020)’s
DLTT+DLC
the Compustat data series on long-term debt, current liabilities, and common equity, respectively; dividend payout is
measured as dividends (Compustat series DVC) scaled by total assets; and finally, corporate investment rate is computed as
the share of capital expenditures (Compustat series CAPX) in total assets. Data used in this table is at firm-year frequency,
A17 *, **, and *** represent results significant at the 10%, 5%, and
and obtained from Compustat. T-statistics are in parentheses.
1% levels, respectively. Standard errors are clustered at the firm level.

Table A11: Reduced-form Evidence for Tobin’s Q, Cash, and Net Debt: R&D-intensive
firms
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
log Q 0.0580*** 0.0598*** 0.0598*** 0.0485*** 0.0570*** -0.0939*** -0.0889*** -0.1052*** -0.1142*** -0.1211***
(49.103) (39.216) (17.465) (12.175) (14.309) (-58.558) (-44.889) (-23.649) (-22.699) (-24.047)
(log Q)2 -0.0000 0.0327*** 0.0281*** 0.0124*** -0.0124*** -0.0089**
(-0.004) (11.546) (9.959) (4.086) (-3.471) (-2.505)
Return of Assets 0.0075*** 0.0103*** 0.0103*** 0.0470*** 0.0470*** -0.0129*** -0.0128*** -0.0130*** -0.0476*** -0.0473***
(4.123) (4.327) (4.327) (15.628) (15.748) (-5.220) (-4.139) (-4.199) (-12.545) (-12.535)
Profitability -0.0322*** -0.0636*** -0.0636*** -0.1296*** -0.1295*** 0.0664*** 0.0620*** 0.0629*** 0.1433*** 0.1427***
(-11.440) (-16.057) (-16.044) (-30.893) (-31.096) (17.418) (12.064) (12.220) (27.045) (27.097)
Investment -0.0217*** 0.0053 0.0052 0.0607*** 0.0741*** 0.0036 -0.0515*** -0.0488*** -0.0791*** -0.0988***
(-3.112) (0.440) (0.439) (4.384) (5.367) (0.378) (-3.319) (-3.140) (-4.521) (-5.663)
Tangibility -0.3810*** -0.4717*** -0.4717*** -0.3838*** -0.3816*** 0.4767*** 0.5733*** 0.5709*** 0.4827*** 0.4813***
(-82.867) (-66.835) (-66.696) (-64.527) (-64.097) (76.417) (62.545) (62.163) (64.262) (63.963)
Book Leverage -0.1172*** -0.1294*** -0.1294*** -0.1957*** -0.1934*** 0.7079*** 0.7175*** 0.7176*** 0.7741*** 0.7711***
(-73.294) (-61.963) (-61.960) (-96.073) (-95.097) (326.160) (264.461) (264.527) (301.009) (299.997)
Payout 0.0523*** 0.0628*** 0.0628*** 0.0296*** 0.0261*** -0.1118*** -0.1434*** -0.1437*** -0.1816*** -0.1771***
(9.692) (9.149) (9.148) (3.447) (3.065) (-15.256) (-16.089) (-16.125) (-16.748) (-16.425)
Size -0.0121*** -0.0138*** -0.0138*** -0.0037*** -0.0038*** 0.0434*** 0.0441*** 0.0440*** 0.0029*** 0.0032***
(-19.449) (-16.347) (-16.347) (-10.758) (-10.846) (51.500) (40.093) (40.064) (6.690) (7.178)
R&D -0.0953*** -0.0953*** 0.2186*** 0.2032*** -0.0373*** -0.0373*** -0.3791*** -0.3644***
(-13.985) (-13.984) (34.432) (32.112) (-4.221) (-4.219) (-47.297) (-45.563)
Fixed effects
N 89,580 59,191 59,191 60,828 60,828 89,580 59,191 59,191 60,828 60,828
Adjusted R2 0.730 0.749 0.749 0.464 0.474 0.859 0.858 0.858 0.713 0.718
3-digit SIC level. The dependent variables are corporate Tobin’s Q. We measure Tobin’s Q as asset market value divided
by its book value. In terms of the market value of assets, we compute it as the book value of assets (Compustat series AT)
to total assets; book leverage is computed as the ratioA18 of total debts to the sum of total debts and common equity, i.e.,
DLTT+DLC

Table A12: Reduced-form Evidence for Tobin’s Q, Cash, and Net Debt: non-R&D-
intensive firms
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
log Q 0.0486*** 0.0501*** 0.0488*** 0.0515*** 0.0525*** -0.0909*** -0.0908*** -0.0952*** -0.1210*** -0.1197***
(44.561) (26.172) (14.692) (15.041) (15.309) (-55.018) (-32.193) (-19.488) (-23.852) (-23.576)
(log Q)2 0.0013 0.0131*** 0.0126*** 0.0043 -0.0089** -0.0105**
(0.505) (4.518) (4.339) (1.112) (-2.079) (-2.437)
Return of Assets -0.0001 0.0107** 0.0108** 0.0347*** 0.0372*** 0.0028 -0.0070 -0.0068 -0.0164** -0.0175**
(-0.063) (2.355) (2.367) (6.382) (6.906) (0.789) (-1.043) (-1.016) (-2.036) (-2.197)
Profitability -0.0056* -0.0028 -0.0027 -0.0890*** -0.0897*** 0.0316*** 0.0392*** 0.0395*** 0.1300*** 0.1273***
(-1.797) (-0.412) (-0.398) (-12.512) (-12.678) (6.693) (3.882) (3.911) (12.328) (12.143)
Investment -0.0459*** -0.0573*** -0.0571*** -0.0222** -0.0255** 0.0804*** 0.0826*** 0.0832*** 0.0691*** 0.0689***
(-9.966) (-6.427) (-6.401) (-2.212) (-2.556) (11.514) (6.294) (6.335) (4.648) (4.662)
Tangibility -0.2374*** -0.2911*** -0.2913*** -0.1911*** -0.1922*** 0.3360*** 0.3459*** 0.3453*** 0.2426*** 0.2492***
(-79.465) (-48.275) (-48.232) (-40.996) (-40.624) (74.216) (38.967) (38.852) (35.095) (35.560)
Book Leverage -0.0911*** -0.0953*** -0.0952*** -0.1235*** -0.1207*** 0.7419*** 0.7456*** 0.7458*** 0.7931*** 0.7890***
(-55.900) (-31.533) (-31.469) (-45.619) (-44.458) (300.403) (167.559) (167.409) (197.687) (196.255)
Payout 0.0115** -0.0028 -0.0029 0.0295*** 0.0205** -0.0333*** -0.0086 -0.0090 -0.0988*** -0.0810***
(2.241) (-0.348) (-0.364) (2.874) (2.005) (-4.275) (-0.723) (-0.758) (-6.497) (-5.359)
Size -0.0151*** -0.0065*** -0.0065*** -0.0078*** -0.0080*** 0.0417*** 0.0408*** 0.0409*** 0.0073*** 0.0076***
(-28.702) (-6.188) (-6.172) (-18.737) (-19.053) (52.302) (26.231) (26.252) (11.968) (12.205)
R&D -0.0501** -0.0499** 0.1808*** 0.1667*** -0.1009*** -0.1004*** -0.3485*** -0.3439***
(-2.551) (-2.544) (10.212) (9.464) (-3.494) (-3.477) (-13.275) (-13.179)
Fixed effects
N 74,193 24,188 24,188 24,970 24,970 74,193 24,188 24,188 24,970 24,970
Adjusted R2 0.666 0.684 0.684 0.349 0.369 0.866 0.870 0.870 0.722 0.731
3-digit SIC level. The dependent variables are corporate Tobin’s Q. We measure Tobin’s Q as asset market value divided
by its book value. In terms of the market value of assets, we compute it as the book value of assets (Compustat series AT)
to total assets; book leverage is computed as the ratioA19 of total debts to the sum of total debts and common equity, i.e.,
DLTT+DLC

Table A13: Reduced-form Evidence for Total Q, Cash, and Net Debt: R&D-intensive firms
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
log Total Q 0.0172*** 0.0171*** 0.0295*** 0.0405*** 0.0419*** -0.0012 -0.0008 -0.0269*** -0.0311*** -0.0320***
(42.103) (33.474) (45.663) (56.839) (59.083) (-0.507) (-0.300) (-7.909) (-9.721) (-9.992)
(log Total Q)2 0.0056*** 0.0084*** 0.0083*** -0.0118*** -0.0141*** -0.0139***
(30.959) (35.928) (35.762) (-12.436) (-13.518) (-13.363)
Return of Assets -0.0004*** 0.0011*** 0.0011*** 0.0029*** 0.0029*** 0.0453*** -0.0093*** -0.0094*** -0.0316*** -0.0323***
(-4.294) (2.905) (3.075) (5.772) (5.962) (89.410) (-4.759) (-4.824) (-14.133) (-14.477)
Profitability 0.0006*** -0.0013** -0.0014*** 0.0013** 0.0012* -0.1153*** -0.0505*** -0.0503*** -0.0513*** -0.0506***
(3.028) (-2.506) (-2.694) (2.064) (1.910) (-98.240) (-18.920) (-18.878) (-17.556) (-17.371)
Investment 0.0023 0.0640*** 0.0410*** 0.1338*** 0.1449*** -0.2023*** -0.5828*** -0.5341*** -0.7744*** -0.7967***
(0.393) (5.765) (3.713) (10.064) (10.963) (-6.141) (-10.030) (-9.183) (-12.989) (-13.354)
Tangibility -0.4171*** -0.5320*** -0.5069*** -0.4829*** -0.4782*** 0.7233*** 0.9737*** 0.9206*** 1.0618*** 1.0704***
(-93.484) (-78.195) (-74.586) (-80.209) (-79.562) (28.212) (27.334) (25.692) (39.331) (39.446)
Book Leverage -0.0004*** 0.0000 0.0000 -0.0016*** -0.0015*** 0.0461*** 0.0484*** 0.0484*** 0.0567*** 0.0561***
(-4.341) (0.114) (0.185) (-9.994) (-9.434) (91.474) (75.050) (75.122) (80.516) (79.923)
Payout 0.0084*** 0.0118*** 0.0109*** -0.0011 -0.0018 -0.2646*** -0.3919*** -0.3900*** -0.4355*** -0.4291***
(4.494) (4.657) (4.350) (-0.335) (-0.573) (-24.543) (-29.511) (-29.410) (-30.501) (-30.141)
Size -0.0082*** -0.0063*** -0.0060*** -0.0074*** -0.0074*** -0.1721*** -0.1691*** -0.1698*** -0.0485*** -0.0481***
(-14.334) (-8.201) (-7.845) (-22.554) (-22.759) (-52.309) (-41.989) (-42.211) (-33.178) (-32.768)
R&D -0.0122*** -0.0110*** 0.0464*** 0.0443*** 0.0028 0.0003 -0.0247*** -0.0228**
(-6.548) (-5.956) (22.053) (21.191) (0.288) (0.030) (-2.617) (-2.423)
Fixed effects
N 96,782 65,232 65,232 66,817 66,817 96,782 65,232 65,232 66,817 66,817
Adjusted R2 0.701 0.715 0.720 0.366 0.379 0.525 0.562 0.563 0.285 0.290
asset ratio is equal to or above 2%. Otherwise, we label the firm as R&D-non-intensive. All the industries are classified
at 3-digit SIC level. The dependent variables are corporate total Q. Peters and Taylor (2017)’s total Q is measured as
the ratio between a firm’s market value over the replacement cost of its intangible capital. The market value of assets is
computed as the book value of assets (Compustat series AT) plus the market value of common stocks (Compustat series
(Compustat series ITCB), minus the value of preferred stocks (coalesce outcomes of Compustat series PSTKRV, PSTKL, and
cash-to-asset ratio, where debt is the sum of long-term debt (Compustat series DLTT) and current liabilities (Compustat
series DLC). Size is the natural logarithm of total assets (Compustat series AT); return of the asset is calculated as the income
before extraordinary items (Compustat series IB) scaledA20 by total assets; asset tangibility is measured as the ratio of physical
DLTT+DLC
expenditures (Compustat series CAPX) in total assets. Data used in this table is at firm-year frequency, and obtained
from Compustat. T-statistics are in parentheses. *, **, and *** represent results significant at the 10%, 5%, and 1% levels,
respectively. Standard errors are clustered at the firm level.

Table A14: Reduced-form Evidence for Total Q, Cash, and Net Debt: non-R&D-intensive
firms
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
log Total Q 0.0094*** 0.0104*** 0.0136*** 0.0180*** 0.0185*** -0.0037*** -0.0007 -0.0033* -0.0094*** -0.0099***
(25.693) (16.742) (19.224) (25.494) (26.073) (-2.888) (-0.438) (-1.899) (-5.976) (-6.235)
(log Total Q)2 0.0006*** 0.0008*** 0.0008*** -0.0005*** -0.0009*** -0.0009***
(9.388) (10.619) (10.730) (-3.122) (-5.210) (-5.229)
Return of Assets 0.0007*** 0.0101*** 0.0104*** 0.0091*** 0.0089*** 0.0431*** -0.0407*** -0.0409*** -0.1230*** -0.1234***
(4.203) (4.252) (4.370) (4.409) (4.345) (75.639) (-7.031) (-7.069) (-26.707) (-26.895)
Profitability -0.0000 -0.0209*** -0.0215*** -0.0172*** -0.0166*** -0.0085*** 0.0377*** 0.0382*** 0.1899*** 0.1877***
(-0.061) (-6.174) (-6.361) (-7.186) (-6.989) (-6.771) (4.576) (4.635) (35.388) (35.151)
Investment 0.0042 0.0039 -0.0020 0.0311*** 0.0274*** -0.0088 -0.1369*** -0.1321*** -0.1788*** -0.1761***
(1.452) (0.441) (-0.224) (3.106) (2.757) (-0.876) (-6.307) (-6.072) (-7.976) (-7.884)
Tangibility -0.2663*** -0.3202*** -0.3159*** -0.2265*** -0.2288*** 0.5262*** 0.4777*** 0.4742*** 0.3520*** 0.3665***
(-90.479) (-52.093) (-51.362) (-47.643) (-47.425) (50.773) (32.025) (31.713) (33.107) (33.829)
Book Leverage -0.0016*** -0.0163*** -0.0162*** -0.0215*** -0.0207*** 0.1323*** 0.4420*** 0.4419*** 0.4688*** 0.4654***
(-6.461) (-14.859) (-14.753) (-19.964) (-19.073) (154.091) (165.932) (165.904) (194.863) (191.051)
Dividend 0.0115** 0.0131* 0.0118 0.0580*** 0.0467*** 0.1043*** -0.0208 -0.0198 -0.1812*** -0.1621***
(2.489) (1.810) (1.632) (6.376) (5.174) (6.435) (-1.184) (-1.124) (-8.902) (-7.991)
Size -0.0155*** -0.0073*** -0.0072*** -0.0107*** -0.0109*** -0.0177*** 0.0248*** 0.0247*** 0.0087*** 0.0090***
(-30.169) (-7.021) (-6.907) (-27.373) (-27.381) (-9.799) (9.773) (9.732) (9.882) (10.045)
R&D -0.0078 -0.0082* 0.0074 0.0065 -0.4752*** -0.4749*** -0.3923*** -0.3944***
(-1.639) (-1.724) (1.555) (1.379) (-41.126) (-41.106) (-36.964) (-37.184)
Fixed effects
N 74,417 24,847 24,847 25,649 25,649 74,417 24,847 24,847 25,649 25,649
Adjusted R2 0.638 0.654 0.656 0.289 0.313 0.664 0.836 0.836 0.726 0.734
asset ratio is equal to or above 2%. Otherwise, we label the firm as R&D-non-intensive. All the industries are classified
at 3-digit SIC level. The dependent variables are corporate total Q. Peters and Taylor (2017)’s total Q is measured as
the ratio between a firm’s market value over the replacement cost of its intangible capital. The market value of assets is
computed as the book value of assets (Compustat series AT) plus the market value of common stocks (Compustat series
(Compustat series ITCB), minus the value of preferred stocks (coalesce outcomes of Compustat series PSTKRV, PSTKL, and
cash-to-asset ratio, where debt is the sum of long-term debt (Compustat series DLTT) and current liabilities (Compustat
A21
series DLC). Size is the natural logarithm of total assets (Compustat series AT); return of the asset is calculated as the income
before extraordinary items (Compustat series IB) scaled by total assets; asset tangibility is measured as the ratio of physical
DLTT+DLC
expenditures (Compustat series CAPX) in total assets. Data used in this table is at firm-year frequency, and obtained
from Compustat. T-statistics are in parentheses. *, **, and *** represent results significant at the 10%, 5%, and 1% levels,
respectively. Standard errors are clustered at the firm level.

SSRN Id4451752

Uploaded by

Copyright:

Available Formats

SSRN Id4451752

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

SSRN Id4451752

Uploaded by

Copyright:

Available Formats

Why Do Powerful Firms Hold Excessive Cash?

This Version: May 17, 2023

This paper investigates the cross-sectional relationship between corporate mar-

JEL codes: D42; G31; G32; L12; L20

Chang’An Street, Beijing, 100738, China. Email address: dansu@ckgsb.edu.cn

Electronic copy available at: https://ssrn.com/abstract=4451752

Figure 1: The Distributional Changes in Cash Holdings and Markups

Electronic copy available at: https://ssrn.com/abstract=4451752

Electronic copy available at: https://ssrn.com/abstract=4451752

Figure 2: Risky Superstar Economy

Electronic copy available at: https://ssrn.com/abstract=4451752

Electronic copy available at: https://ssrn.com/abstract=4451752

Electronic copy available at: https://ssrn.com/abstract=4451752

2.1 Data and Variable

Electronic copy available at: https://ssrn.com/abstract=4451752

Electronic copy available at: https://ssrn.com/abstract=4451752

2.2 Distributional Changes in the Time-Series

Electronic copy available at: https://ssrn.com/abstract=4451752

this sample period.

Figure 3: Changes in Median Corporate Cash/Net Debt at Different Markup Quintiles

(a) Cash (b) Net Debt

2.3 Cross-Sectional Relationship

Electronic copy available at: https://ssrn.com/abstract=4451752

2.3.1 Initial evidence with Binscatter plots

Electronic copy available at: https://ssrn.com/abstract=4451752

Figure 4: Binscatter Plot for Markup, Cash, and Net Debt

2.3.2 Regression analysis

cashi,t = α + β × markupi,t + γ × markup2i,t + ΓXi,t + δi + ηt + ε it

Electronic copy available at: https://ssrn.com/abstract=4451752

Electronic copy available at: https://ssrn.com/abstract=4451752

for a set of other variables.

2.4 Alternative Stories

Multinational Corporations Many studies have shown that multinational corporations

Electronic copy available at: https://ssrn.com/abstract=4451752

Table 1: Reduced-form Evidence for Markup, Cash, and Net Debt

Cash/Asset Net Debt/Asset

Electronic copy available at: https://ssrn.com/abstract=4451752

Electronic copy available at: https://ssrn.com/abstract=4451752

Figure 5: Binscatter Plot for Alternative Stories

(a) Multinational (b) Domestic

(c) R&D intensive (d) R&D non-intensive

Electronic copy available at: https://ssrn.com/abstract=4451752

Now we rationalize our previous empirical findings by using a partial-equilibrium, dy-

Corporate earnings We start by describing corporate earning profiles. A firm with an

Electronic copy available at: https://ssrn.com/abstract=4451752

π (z) ≡ max p (z, y) y − C (y) (1)

Therefore, ỹ (z) is chosen to satisfy:

Electronic copy available at: https://ssrn.com/abstract=4451752

Electronic copy available at: https://ssrn.com/abstract=4451752

Summary of firm’s optimization problem To describe the firm’s optimization problem,

V (z, ω, c, b) = max {0, V c (z, ω, c, b)} (6)

Electronic copy available at: https://ssrn.com/abstract=4451752

subject to the following five different constraints:

3.2 Demand and Supply Origins of The Risky Superstar Economy

Electronic copy available at: https://ssrn.com/abstract=4451752

Electronic copy available at: https://ssrn.com/abstract=4451752

Electronic copy available at: https://ssrn.com/abstract=4451752

Electronic copy available at: https://ssrn.com/abstract=4451752