INDUSTRY INFLUENCES ON CORPORATE FINANCIAL POLICIES

by

Jun Zhou

A thesis submitted in conformity with the requirements

for the degree of Doctor of Philosophy

Graduate Department of Joseph L. Rotman School of Managemen

University of Toronto

© Copyright by Jun Zhou (2010)

ii

Industry Influences on Corporate Financial Policies

Jun Zhou

Doctor of Philosophy

Joseph L. Rotman School of Management

University of Toronto

2010

Abstract

This thesis examines how industry differences affect both corporate financial policies and

valuation.

Chapter 1 studies the impact of a firm‟s product market power, through the channel of business

risk, on its dividend policy. Using three measures of market power – the Herfindahl-Hirschman

index, the degree of import competition and the Lerner Index, I find that market power

positively affects a firm‟s dividend decision, both in terms of the probability of paying a

dividend and the amount of the dividend. I also provide evidence that the route through which

market power affects the dividend decision is business risk: a firm with greater market power is

less risky and hence more likely to pay dividends and pay more dividends.

Chapter 2 examines industry differences on the level of corporate cash holdings since the 1970s

with a focus on high-tech versus non-high-tech firms. In contrast to the average cash-to-assets

ratio of non-high-tech firms, which remained stable at a level close to that of the 1970s, the

iii

average cash ratio of high-tech firms more than tripled from 1980 to 2007. I find that this

difference can be explained by changing firm characteristics across these two industrial sectors.

This is due to high-tech new listings, whose changing characteristics and increasing proportion

have caused the population characteristics of the high-tech sector to tilt toward those typical of

firms that hold more cash.

Chapter 3 investigates the industry impact on the marginal value of corporate cash holdings and

how it has evolved over time. I find that on average the difference in the marginal value of cash

between high-tech and non-high-tech firms has become larger during the sub-period which

covers the 1990s and 2000s, as compared to earlier time periods. Furthermore, I show that this

increase can be explained by changing firm characteristics related to the precautionary demand

for holding cash.

Overall, this thesis shows that industry differences, represented by varying degrees of market

power and changing firm characteristics, have significantly affected corporate financial policies,

both in terms of dividend policy and optimal cash holdings.

iv

Acknowledgments

I would like to express my sincerest thanks to all, who have helped, encouraged, and advised me

during my journey in putting together this thesis.

In particular, my first word of thanks goes to my supervisor, Laurence Booth. Working under

his supervision has been an excellent learning experience for me. I feel grateful to him for his

close guidance, thoughtful advice, and constant encouragement. I am deeply indebted to my

committee members, Craig Doidge and David Goldreich, for their constant support and valuable

suggestions. I also thank Tom McCurdy, Jan Mahrt-Smith, Sergei Davydenko, Esther Eiling,

Ling Cen, for inspiration and support. Finally, I would like to thank my colleagues in the PhD

program for all their help and support throughout these years.

I dedicate this thesis to my parents and Christos, for their love and support!

v

Table of Contents

_

List of Tables .................................................................................................................................. vi

List of Figures ............................................................................................................................... vii

List of Appendices ....................................................................................................................... viii

Chapter 1 Market Power and Dividend Policy: a Risk-Based Perspective ..................................... 1

1.1 Literature Review ......................................................................................................................... 5

1.2 Sample Selection and Variable Definitions.................................................................................. 8

1.3 Empirical Results ....................................................................................................................... 14

1.4 Conclusion.................................................................................................................................. 20

Chapter 2 Increase in Cash Holdings: Pervasive or Sector-Specific ............................................. 35

2.1 Literature Review ....................................................................................................................... 38

2.2 Time Trends in Corporate Cash Holdings .................................................................................. 41

2.2.1 Sample ................................................................................................................................. 41

2.2.2 Cash Trends ......................................................................................................................... 42

2.2.3 Robustness Tests ................................................................................................................. 43

2.3 Explaining the Different Trends in Corporate Cash Holdings ................................................... 46

2.3.1 The Determinants of Cash Holdings ................................................................................... 47

2.3.2 Difference in Changing Firm Characteristics across Two Sectors ...................................... 52

2.3.3 Can Difference in Changing Characteristics Explain Different Cash Trends? ................... 56

2.4 Conclusion.................................................................................................................................. 58

Chapter 3 The Value of Cash: Industry and Temporal Effect ....................................................... 85

3.1 Related Literature ....................................................................................................................... 87

3.2 Empirical Methodology.............................................................................................................. 89

3.3 Sample and Summary Statistics ................................................................................................. 91

3.4 Empirical Results ....................................................................................................................... 92

3.5 Explanations ............................................................................................................................... 96

3.5.1 Fundamental Explanation .................................................................................................... 96

3.5.2 Potential Impact of Mispricing ............................................................................................ 99

3.6 Conclusion................................................................................................................................ 101

References.................................................................................................................................... 124

vi

List of Tables

Table 1.1: Market Power Measures........................................................................................................ 21

Table 1.2: Summary Statistics ................................................................................................................ 22

Table 1.3: The Decision to Pay Dividends as a Function of Market Power .......................................... 23

Table 1.4: Level of Dividend Payment as a Function of Market Power ................................................ 25

Table 1.5: Testing the Risk-Based Explanation ..................................................................................... 26

Table 1.6: Subsample Analysis .............................................................................................................. 29

Table 2.1: The Distribution of Firms ..................................................................................................... 60

Table 2.2: Trends in Cash Holdings: Fama-French Industries and GICS Economic Sectors ................ 61

Table 2.3: Descriptive Statistics ............................................................................................................. 62

Table 2.4: Correlation Matrix................................................................................................................. 64

Table 2.5: Determinants of Corporate Cash Holdings: 1974-2007 ........................................................ 68

Table 2.6: Changes in Firm Characteristics ........................................................................................... 71

Table 2.7: Determinants of Corporate Cash Holdings: Estimation vs. Forecast Periods ....................... 74

Table 2.8: Actual and Predicted Cash Holdings: Forecast Period ......................................................... 76

Table 3.1: Sample Distribution and Descriptive Statistics ................................................................... 102

Table 3.2: Value of Cash ...................................................................................................................... 106

Table 3.3: Robustness Check – Alternative Industry Classifications and Sub-periods ....................... 108

Table 3.4: Robustness Check--- Faulkender and Wang (2006) Model ................................................ 110

Table 3.5: Robustness Check – Benchmark Returns based on the DGTW portfolios ......................... 112

Table 3.6: Fundamental Explanation ................................................................................................... 114

Table 3.7: Mispricing Explanation – Impact of Changes in Sentiment ............................................... 115

Table 3.8: Mispricing Explanation – Periods of Large and Small Changes in Sentiment ................... 117

Table 3.A.1: Replication of Faulkender and Wang (2006) – Summary Statistics ............................... 119

Table 3.A.2: Replication of Faulkender and Wang (2006) – Regression Results ................................ 120

vii

List of Figures

Figure 2.1: Trends in Cash Holdings and Net Leverage: 1974-2007 ..................................................... 79

Figure 2.2: Trends in Cash Holdings – R&D Adjusted Assets as the Denominator .............................. 81

Figure 2.3: Annual Number of IPOs ...................................................................................................... 82

Figure 2.4: Trends in Cash Holdings of Newly IPO Firms and Seasoned Firms ................................... 83

Figure 3.1: Rolling Windows ............................................................................................................... 121

viii

List of Appendices

Appendix 1.A: Sample Selection ........................................................................................................... 31

Appendix 1.B: Import Penetration in Manufacturing Industries ............................................................ 32

Appendix 1.C: Variable Definitions....................................................................................................... 34

Appendix 2.A: Variable Definitions ...................................................................................................... 84

Appendix 3.A: Variable Definitions .................................................................................................... 122

Appendix 3.B: Replicating Faulkender and Wang (2006) ................................................................... 123

1

Chapter 1 Market Power and Dividend Policy: a Risk-Based Perspective

1

The strength of Intel's competitive position combined with our solid financials allows us to

again reward Intel shareholders with an increase in the quarterly dividend.

Craig Barrett, Intel's chairman, Wall Street Journal, March 21, 20082

The dividend policy of Corporate America has gone through significant changes over the past

two decades. Fama and French (2001) document that the proportion of listed firms paying cash

dividends fell from 66.5% in 1978 to 20.8% in 1999. Even after controlling for firm

characteristics such as size, growth opportunities, and profitability, the propensity to pay a

dividend still declined. DeAngelo et al. (2004) report that dividends have become highly

concentrated among a small group of firms that experienced substantial earnings during this

period. Further, Skinner (2008) argues that dividends are gradually being replaced by stock

repurchases while even the big dividend payers have become increasingly conservative in their

dividend payments.

Over the same period, the business environment faced by American firms has changed

remarkably. Alan Greenspan (2002), at the time the Governor of the Federal Reserve Board,

pointed out that

“a wave of innovation across a broad range of technologies, combined with considerable

deregulation and a further lowering of barriers to trade, fostered a pronounced expansion of

competition and creative destruction.”

This transformation of the competitive environment has brought new opportunities as well as

challenges for both incumbents and new entrants alike. In all likelihood it has increased the

business risk faced by the typical American firm. These two trends elicit an interesting question:

1 This paper is based on joint work with my supervisor, Professor Laurence Booth

2 “Intel to Boost Its Dividend About 10%”, The Wall Street Journal, March 21, 2008.

2

is a firm‟s dividend policy influenced by its competitive environment, and in particular by its

relative position in the market?

In this paper we conduct a comprehensive empirical study to investigate this linkage between

market structure, risk and dividend policy. Since firms decide their financial policies to

accommodate their business risk, in the sense that financial risk is layered on top of business

risk, changes in a firm or industry‟s competitive status should trigger changes in its financial

policies. Microeconomic theory has long shown that a firm‟s market power affects its business

risk, that is, the risk associated with operating earnings. Further survey studies (Lintner 1956

and Brav et al. 2005) have demonstrated that firms view the stability of future earnings as one of

the most important determinants of their dividend policy. Consequently, a firm‟s market power

should influence its dividend policy through its impact on business risk. This risk-based

perspective can help us understand the potential link between the above two trends, both in

dividend policy and the degree of market competition.

At the same time, a number of regulatory reforms and market innovations have brought in new

entrants to the capital market. Reforms in the financial industry have made it easier for some

firms to finance through the public markets via an initial public offering, rather than relying on

the private venture capital market; consistent with the new-listing phenomenon documented by

Fama and French (2001, 2004). These newly-listed firms, with predominantly unstable

profitability but strong growth opportunities, are typically firms in the growth stage of their life-

cycle and lack market power. Consequently, they are not proper candidates to pay dividends

according to the life-cycle theory of dividends advanced by Fama and French (2001), Grullon et

al. (2002), DeAngelo et al. (2006) and others.

On the other hand, challenges from domestically based new entrants, combined with an influx of

foreign products due to increased international trade, impair the market power of established

firms, thereby increasing their business risk. Escalating business risk experienced by existing

firms would impede both their incentives to initiate dividends and raise them, regardless of their

stage in the life-cycle. This view is supported by Skinner (2008) finding of increasing

conservatism even by dividend payers and the findings of Aivazian et al. (2006) and others that

3

dividends are increasingly “sticky” in the sense that dividend payers are reluctant to increase

their dividend payments.

The impact of competition on business risk may vary across firms. Studies have shown that a

firm with relatively higher market power should be less affected by either its competitive

environment or exogenous shocks. Since the firm can pass through the shocks to its customers,

it can attenuate the instability of its cash flows (Booth, 1981; Gaspar and Massa, 2006; Irvine

and Pontiff, 2009). This result is also consistent with DeAngelo et al. (2004) finding of an

increasing concentration of dividends among a small group of well-established firms. Overall,

the impact of market power on a firm‟s business risk provides a framework that enables us to

understand these recent changes in dividend policy documented in the literature.

The industrial organization and international trade literatures recommend three market power

measures which allows a relatively complete description of market power. At the industry level,

the Herfindahl-Hirschman index (HHI) is used to capture domestic market power, by providing

a more sophisticated measure of the degree of concentration within an industry than simple

concentration ratios. In contrast the level of import penetration is used to capture the market

power of firms exposed to challenges from foreign rivals, a feature of competition not captured

by the HHI. The assumption is that the average firm in an industry with higher HHI or lower

import penetration has higher market power compared to firms in other industries. Finally, firm-

level market power is measured by the accounting-based approximation to the classic Lerner

index, which directly captures a firm‟s ability to charge a market price above its marginal cost

and thus the elasticity of the demand curve faced by the firm.

The empirical findings from a sample of manufacturing firms during the period 1972-2002

generally support the prediction that a firm with higher market power is more likely to pay a

dividend and when it does pay more. The explanatory power of both the degree of import

penetration and the Lerner index are statistically significant even when controlling for

traditional firm characteristics associated with dividend policy such as the firm‟s profitability,

size, growth opportunities, retained earnings, and current risk. This in itself is a strong result,

since these firm characteristics are themselves endogenous to the firm‟s market structure. On the

4

other hand, the explanatory power of the HHI is less stable; a result that could be attributed to

the fact that the HHI is only available every five years and only for the short sample period

1982-1996. It may also be that the HHI is subject to the contestable market argument, associated

with the fundamental changes occurring as a result of globalization and deregulation.

Investigation of the mediating channel for the impact of market power on dividend policy

supports the risk-based explanation. Firms with higher market power experience less business

risk in the future, which is reflected in both higher profitability and more stable operations.

Hence, the persistence and stability of earnings in the future, consequent on a firm gaining

market power, can reduce a firm‟s reluctance to pay dividends or increase them, consistent with

recent survey findings.

This paper is closely related to a recent paper by Grullon and Michaely (2008) -- it asks a

similar question but from a different perspective. Grullon and Michaely (2008) focus on the

corporate governance role of product market competition from an agency-based perspective.

They conjecture that firms operating in highly competitive industrial sectors have higher payout

ratios (either dividend or total payout), since competition constrains firms from wasting

corporate resources through perquisite consumption and managerially based investment decision.

In contrast, and with all else constant, firms with market power have an agency-based incentive

to spend corporate resources for the benefit of managers, rather than shareholders. If this

reasoning is correct, increasing competition over the past two decades should lead to better

governance, more firms paying dividends, and larger dividend payments: a conjecture which

contradicts the extant evidence In contrast, the risk-based explanation advanced in this paper is

consistent with existing evidence both over time and across firms.

This paper contributes to the literature in several ways. First, it helps understand recent changes

in dividend policy as documented in Fama and French (2001), DeAngelo et al. (2004), Avazian

et al. (2006) and Skinner (2008), by suggesting that market power and industrial structure is a

fundamental determinant of dividend policy through its impact on business risk. Second, this

paper complements the classic life-cycle theory of dividend policy articulated by DeAngelo,

DeAngelo, and Stultz (2006). The process of a firm moving to the mature stage in its life-cycle

5

is also the process of gaining market power, where the speed of this progress is also a function

of consolidation within the industry as the number of firms operating in the industry declines.

Finally, this paper contributes to the literature that links industrial organization and finance,

such as Booth (1981) on equity costs, Xu (2008) on capital structure, and Gaspar and Massa

(2006) and Irvine and Pontiff (2009) on idiosyncratic volatility.

The remainder of this paper is organized as follows. Section 1.1 reviews the literature. Sample

selection criteria and variable definitions are presented in Section 1.2. Section 1.3 reports the

results of the impact of market power on dividend policy and the risk-based explanation. Section

1.4 concludes.

1.1 Literature Review

In an imperfectly competitive economy, a firm faces a downward-sloping demand curve for its

own product, indicating that the market price of its product will increase when the firm reduces

supply. This firm is regarded as having market power since at the profit-maximizing output

level it can profitably charge a product price above its marginal cost.3 Further, given this ability

to partially influence its price, the firm is able to smooth the fluctuation of its operating income

in the face of exogenous economy-wide and/or firm-specific shocks. The reduced business risk

due to market power will be reflected in capital markets, in the form of lower systematic and/or

idiosyncratic volatility of the firm‟s stock returns.

The impact of a firm‟s market power on its risk has been investigated in several empirical and

theoretical studies.4

An early empirical study by Sullivan (1978) documented a negative

3 From an antitrust perspective, market power indicates the potential that consumers can be exploited and resources

may be misallocated due to the absence of competition. However, if market power results from owning superior

products or business acumen, it is justified by intellectual property and does not offend the antitrust laws (Federal

Trade Commission, Protecting America‟s Consumers, part III, http://www.ftc.gov/opp/jointvent/classic3.shtm).

Instead of investigating the social welfare implications of market power, this paper focuses on the impact of market

power on a firm‟s corporate financial decisions, more specifically, dividend policy.

4 The measures of market power are different in empirical studies. Sullivan (1978) defines a firm with market

power as a larger firm and/or a firm in a concentrated industry. Gaspar and Massa (2006) measures market power

by the Lerner index of the firm and market concentration of the industry that the firm belongs to. In Irvine and

Pontiff (2009), a firm with higher market power is measured by belonging to industries with lower industry

turnover or import penetration, or having higher return on assets.

6

correlation between the CAPM beta and market power. He speculated that a firm with market

power will be able to “influence or more successfully react to major changes in social, economic

and political events” and hence is less subject to systematic risk. This idea was formalized by

both Subrahmanyam and Thomadakis (1980) and Booth (1980 and 1981) who both looked at

the impact of price uncertainty on a firm‟s cost of capital. Due to the non-infinite price-elasticity

of demand faced by a firm with market power, the economic rents generated from the optimal

output decision allow the firm to mitigate the impact of economy-wide shocks. Therefore, the

firm‟s systematic risk and cost of equity capital will decrease accordingly.

Using the more sophisticated techniques of asset-pricing tests, Hou and Robinson (2006) find

that firms operating in more concentrated industries earn lower returns after controlling for

generally-accepted risk factors. They propose a risk-based explanation that firms in concentrated

industries face fewer distress risks or engage in less innovation and thus have a lower cost of

capital, which is consistent with the earlier theoretical work. Further, several recent papers have

investigated the potential impact of market power on a firm‟s idiosyncratic risk. Gaspar and

Massa (2006) find that firms with higher market power have lower idiosyncratic volatility. Their

explanation is that market power can help a firm hedge the firm-specific shock from its product

market and/or reduce the information uncertainty faced by its investors. Irvine and Pontiff (2009)

document that the increase in idiosyncratic return volatility over the past four decades is

associated with a concurrent increase in the idiosyncratic volatility of fundamental cash flows,

while the latter is closely related to intensive competition caused by deregulation and

globalization. Peress (2010) sets up a rational expectation model under asymmetric information

to formalize these recent empirical findings, where firms with more market power are able to

transfer exogenous shocks in their product markets to customers and reduce the impact on their

profits. Despite focusing on different aspects of a firm‟s risk and using different measures of

market power, these papers have used more modern statistical techniques to confirm the classic

result that firms with more market power are less risky.

Risk has always been an important determinant of dividend policy. In their influential book on

value-investing, Security Analysis, Graham and Dodd (1951, p.596) give possibly the earliest

discussion of the link between risk and dividend policy by pointing out that

7

“a sound and realistic procedure for management would be to select some conservative dividend

rate as a base which is expected to be adhered to under all but unexpected adverse developments”

The field survey conducted by Lintner (1956) shows that conservative managers usually are

reluctant to increase dividends that will subsequently have to be reversed due to negative cash

flow shocks. More recently Brav et al. (2005) find that more than two-thirds of the CFOs of

dividend-paying firms regard the stability of future earnings as a very important factor to their

dividend policies, just second to maintaining consistency with historic dividend policy. Further,

Brav et al (2008) find in a survey on the impact of the May 2003 dividend tax cut that the

stability of future cash flows is regarded as the most important factor for dividend initiators.

While field work is unambiguous that business risk affects dividend policy, several recent

empirical studies have investigated the link between dividend policy and both the systematic

and idiosyncratic risk of its stock returns. In reexamining the information content of dividend

changes, Grullon et al. (2002) find that risk decreases following a dividend increase. They

interpret the result as evidence that the firm has moved to the mature stage of its life-cycle,

where it has fewer growth opportunities and assets in place play a bigger role in determining the

firm‟s value. Further, they argue that dividend changes convey information regarding the change

in the future riskiness of a firm‟s cash flow. Hoberg and Prabhala (2009) investigate the link

between a firm‟s propensity to pay dividends and its systematic and idiosyncratic risks. They

find that firms with higher risk, systematic or idiosyncratic, are less likely to pay dividends.

They claim that risk can explain almost 40% of the Fama French disappearing dividends puzzle.

Similarly Booth and Xu (2007) show that firms with more idiosyncratic risk are more likely to

smooth their dividends, due to the associated information asymmetries. Moreover, the

explanatory power of idiosyncratic risk is economically more important than that of systematic

risk. Besides, Chay and Suh (2009) find international evidence that a firm‟s cash flow

uncertainty, measured by its current stock return volatility, affects both its decision to pay a

dividend and the amount of dividend payment.

It is clear from the existing literature that there is a strong link between risk, however defined,

and a firm‟s dividend policy. It remains to relate these results to the fundamental, which

determines this risk, which we conjecture is in part the existence of market power.

8

1.2 Sample Selection and Variable Definitions

Taking the CRSP/Compustat Merged Database (Industrial Annual) as the base sample, we

follow the method described in Fama and French (2001), with some minor modifications, to get

the raw accounting variables and screen the sample for outliers (see Appendix 1.A for details).

The CRSP historical SIC codes are used to identify the four-digit SIC code for each firm-year

observation. Due to the data availability of industry-level measures of market power from the

Census of Manufacturers and the NBER Trade Database, the sample is limited to firms

operating in manufacturing industries (SIC 2011-3999) during the period 1972-2002. The final

sample is an unbalanced panel comprising 27,520 firm-year observations.

Two aspects of dividend policy are considered: the likelihood to pay a dividend and the level of

the dividend payment. Following Fama and French (2001), the likelihood to pay a dividend is

measured by a payer dummy, which takes the value one for any year t if the dividend per share

(Compustat data 26) of the firm in that year is positive and zero otherwise. Following Fama and

French (2002), the level of dividend payments for any year t is the common dividends

(Compustat data 21) scaled by total assets for year t.5

Market power is defined as the ability of a firm to raise the price of its product above its

marginal cost. However, there is no consensus as to the proper empirical proxy for market

power due to the complexity of the real world. Therefore, three proxies that have been applied in

the industrial organization literature are used: the Herfindahl-Hirschman index, the level of

import penetration, and the Lerner index.

Domestic competition is measured by the Herfindahl-Hirschman index (HHI), a measure widely

applied in empirical studies of industrial organization (Schmalensee, 1989). It is defined as the

sum of the squared market shares of all the firms in a given industry (Herfindahl 1950). An

industry with a higher HHI is generally regarded to be less competitive, indicating that firms in

5 Two other popular measures are dividend yield (defined as DPS/P) and dividend payout ratio (defined as

DPS/EPS). However, the noise added by price variations reduces the soundness of dividend yield as a stable

measure. Due to dividend smoothing found by Lintner (1956), many firms that experience temporary negative

earnings avoid changing their dividend payment. Therefore, these two measures are not used in this paper.

9

the industry have greater market power relative to firms in more competitive industries with a

lower HHI. The HHI data used in this research are collected from the 1982, 1987 and 1992

Census of Manufacturers. Compared with the Compustat-based HHI, the Census-based HHI is

more accurate since it is based on information from both private and public firms in a four-digit

SIC industry.6 Regulators such as the Department of Justice use the Census-based HHI to

measure market power in designing antitrust policies. However, one limitation of using the

Census-based HHI is the lack of time series variations, since it is only updated every five years

and hence at most three values are available for each industry in our study. Following MacKay

and Phillips (2005) and Akdogu and MacKay (2008), the same HHI is assigned to each industry

until the next census year, i.e. the 1987 HHI is assigned to observations in 1987, 1988, 1989,

1990, and 1991.

Due to increasing globalization, the market power of a firm is affected not only by other

domestic firms in the same industry, but also by foreign rivals that provide similar products. It

has been widely accepted that international trade increases competitive pressure for domestic

firms, since foreign supply pushes down the product price in the domestic market and hence

reduces the market power of domestic firms (Helpman and Krugman, 1985, 1989).7 Therefore,

competitive pressure from foreign rivals should be an indispensable aspect in the analysis of

market power.8 One widely-accepted measure of industry-level competitive pressure from

6 Starting from 1997, the North American Industry Classification System (NAICS) replaced the SIC system to

categorize industries in the Census. Ali et al. (2009) recently propose a method to translate the NAICS-based HHI

into the SIC-based HHI. However, they neglect the fact that these two categorizing systems are not one-to-one

matching, i.e. one NAICS code may be matched to several SIC codes, and vice versa. To avoid potential

measurement errors associated with this translation, our paper only considers the time period when the SIC-based

HHI is available.

7 This idea has been formalized by many models developed in international trade under imperfectly competitive

product markets.

8 For an industry like Motor Vehicles and Passenger Car bodies (4-digit SIC code 3711), four-firm concentration

ratio has always been above 90% and HHI was 2676 in year 1992 as documented by the Census of Manufacturers.

Both indicators imply that this industry is highly concentrated. Further investigation on the Compustat shows that

three major manufacturers have dominated this industry, i.e. General Motors, Chrysler, and Ford. However, it is

improper to conclude that these firms can enjoy the market power since news on employee layoffs and branch

shutdowns have been frequently reported in major journals and newspapers. Actually, increasing import has

intensified competition and survival threat faced by these giant American car manufacturers. Import penetration in

this industry has steadily increased from 14.68% in 1972 to 32.74% in 1996; while the average (median) import

penetration across all manufacturing industries has increased from 6.63% (3.54%) to 20.97% (14.94%) during the

10

foreign rivals is import penetration, defined as the proportion of domestic demand that is

satisfied by imports (Tybout, 2003). More specifically, import penetration of industry i at year t

is defined as follows:

Import Penetration (IPit) = it

it it it

imports

shipments exports imports

where the denominator measures the domestic absorption of the products from industry i at year

t. Higher import penetration implies a lower proportion of domestic product consumed in the

domestic country, which indicates lower market power for domestic firms.

Import penetration data for four-digit SIC-based manufacturing industries come from two

datasets compiled by Professor Peter Schott. 9

The first dataset contains import penetration data

during the period 1972-1996, and it has been used by Bernard et al. (2006). The second dataset

contains the data of multilateral imports and exports for the period 1989-2005. Along with the

values of shipments documented in the NBER-CES Manufacturing Industry Database (1958-

2002), we can calculate import penetration for the period 1989-2002.

Two datasets are not perfectly comparable since the second trade dataset uses a revised version

of HS-SIC4 concordance, which provides a more complete match of the imports and exports for

the relevant domestic industries.10

As a result, these two datasets may provide different values

same period. Therefore, without considering competitive pressure brought by foreign rivals, GM, Chrysler, and

Ford may be incorrectly classified as firms with high market power.

9 The first dataset is available under section “Import Penetration by SIC4 (1987 Revision) Manufacturing Industry,

1972 to 1999” at http://www.som.yale.edu/faculty/pks4/sub_international.htm. The data are actually available for

year 1972-1996. Prof. Schott kindly provided us the second dataset, containing multilateral imports and exports

during 1989-2005. 10 Before 1988, imports were classified according to the Tariff Schedule of the United States Annotated (TSUSA),

while U.S. exports were classified according to the “Schedule B” system. The Ominus Trade and Competitiveness

Act of 1988 mandates that both imports and exports are classified under the Harmonized System (HS) afterwards

(Feenstra, 1996 and 1997). The Center for International Data at UC Davis (http://www.internationaldata.org/)

provides the annual data of U.S. imports and exports between 1972 and 2006. Besides TSUSA, Schedule B, and HS

classification according to reporting years, these data also contain import- and export-based SIC number. However,

the import- and export-based SIC number are different from domestic-based SIC codes, which are defined

according to the method of processing for a good but this information is unknown for import and export (Feenstra

et al. 2002). Since the value of shipments required in import penetration formula is for industries classified

11

for the import penetration of a given industry during the overlapping period 1989-1996. To

obtain import penetration for the entire 1972-2002 period while minimize the potential

inconsistency between two datasets, we consider the following two approaches. The first

approach is to keep intact the 1989-2002 import penetration calculated from the second dataset,

and decide whether to extend the time series back to 1972 by using the first dataset. For each

industry, we compare the values of its import penetration in 1989 from two datasets. If the

absolute value of their difference ratio ((IP from 1989-2002 dataset)/(IP from 1972-1996

dataset)-1) is lower than 25%, 1972-1988 values will be included in the study. The alternative

approach that combines the two datasets is to take all the values of import penetration from the

1972-1996 dataset, and then decide whether to extend the time series to 2002 with data from the

1989-2002 dataset. This alternative approach requires the comparison of the values of import

penetration in 1996, by following the decision criterion mentioned above.

Appendix 1.B reports the annual mean and median of import penetration across all industries in

the manufacturing sector for the period 1972-2002 according to these two approaches. In any

year the distribution of import penetration across industries is positively skewed, which

indicates that some industries are heavily exposed to foreign competition. Moreover importantly,

on average import penetration has steadily increased and more than tripled during the sample

period, with mean (median) import penetration increasing from 7.88% (3.81%) in 1972 to 26.33%

(19.73%) in 2002 for the first approach; and increasing from 6.63% (3.54%) in 1972 to 27.20%

(20.49%) in 2002 for the second approach. This trend implies that on average U.S.

manufacturing firms have become more exposed to competitive pressures from foreign rivals as

a result of globalization.

Our preference is to use the updated dataset as much as possible. Apparently, the first approach

uses all the updated data, while the second approach only selectively uses the updated data after

1996. Thus, for the main test, we take the 1972-2002 value of import penetration constructed by

the first approach. Nevertheless, we will also consider the values of import penetration

constructed by the second approach in order to perform a robustness check on our results.

according to domestic-based SIC, we need import and export data identified based on domestic SIC codes.

Therefore, we use the data provided by Peter Schott.

12

When the above two industry-level measures of market power are applied to a firm-level study,

an implicit and unrealistic assumption is that every firm in a given industry has the same level of

market power. To capture the cross-firm difference in market power, the Lerner index is used,

defined as (P-MC)/P, where P is a firm‟s product price and MC is the marginal cost of

production (Lerner, 1934). Theoretically, a firm‟s Lerner index is equal to the inverse of the

price elasticity of the demand curve for its product. By definition, the Lerner index directly

captures the property of market power, i.e. the ability of a firm to charge price above marginal

cost. One challenge to use the Lerner index in empirical studies is that marginal costs are

unobservable. Researchers generally approximate the Lerner index by the price-cost margin,

where marginal cost is substituted by a measure of average variable cost. Following Gaspar and

Massa (2006), the Lerner index is defined as operating profits (before depreciation, interest,

special items, and taxes) over sales. The Lerner index is estimated for each firm using data from

COMPUSTAT.

Table 1.1 shows the pairwise correlations among the three measures of market power at the

industry level. Since the Lerner index is calculated at the firm level, it is averaged across firms

within each four-digit SIC industry for each year, using either equal weights (Panel B) or value

weights based on sales (Panel C). The industry median value for the Lerner index is used (Panel

D) to minimize the impact of outliers in all three panels. The Lerner index is negatively

correlated with import penetration and positively correlated with the HHI. These findings are

consistent with previous analysis that high market power can be measured by low import

penetration, high market concentration, and high values for the Lerner index. Hence, the

findings confirm that all three measures provide a consistent description of market power.

Further, the correlation between import penetration and the HHI is negative but insignificant,

indicating that competition among domestic firms and competition caused by foreign rivals are

two distinct aspects of market structure.

Previous studies have identified several firm characteristics that can explain a firm‟s dividend

policy. Fama and French (2001) emphasize that size, profitability, and growth opportunities are

the three most important determinants of a firm‟s propensity to pay a dividend. Firm size is

measured by its NYSE decile for a given year (NYP) so as to neutralize any effect of the growth

13

in firm size over time. Profitability is measured as the ratio of a firm‟s earnings before interest

and tax to its total assets. Investment opportunities are measured by two variables: the firm‟s

growth rate of assets (AGR) and its market-to-book ratio (M/B). Fama and French (2001) find

that larger, more profitable firms with fewer growth opportunities are more likely to pay

dividends. In testing for the level of dividend payments, the Fama and French (2002)

specialization is used, where the natural logarithm of a firm‟s total assets is used as the proxy for

size.

DeAngelo et al. (2006) suggest using the ratio of retained earnings over total assets (RE/TA) as

a measure of a firm‟s stage in its life cycle and find that it is an important determinant of the

propensity to pay a dividend. Firms with low RE/TA tend to be in the early stage of their life-

cycle where capital accumulation is still underway, so they are less likely to pay a dividend.

Conversely, firms with high RE/TA tend to be in the mature stage, where they have

accumulated profits and become largely self-financing, so they are likely to pay a dividend.

Altman et al (1977) have also used this measure as a proxy for risk consistent with the

conjecture that market power is a determinant of risk and the stage in the firm‟s life cycle.

Hoberg and Prabhala (2009) find that a firm‟s systematic and idiosyncratic risk can explain its

propensity to pay a dividend. Chay and Suh (2009) find international evidence that a firm‟s cash

flow uncertainty, measured by its current stock return volatility, affects both the likelihood to

pay a dividend and the amount of the dividend. To control for current stock return volatility

(RETVOL), the standard deviation of a firm‟s monthly returns during a one-year period is used.

The details for constructing these variables are given in Appendix 1.C. All the control variables

are winsorized at 1 and 99 percentile to avoid the impact of extreme values.

Table 1.2 presents the descriptive statistics and correlation matrix of firm-level variables. About

61% of firm-year observations in the sample pay dividends and most firms are quite large. The

correlation between profitability and RE/TA is 0.657. The high correlation is interpreted by

DeAngelo et al. (2006) as RE/TA partially capturing current profitability. The Lerner index is

also highly correlated with profitability, with a correlation equal to 0.750. This high correlation

is unsurprising since the Lerner index is approximated by operating income over sales and

14

profitability is defined according to Fama and French (2001) as the earnings before interest and

tax over total assets. Since high correlations may indicate potential multicollinearity among

these variables, variance inflation factors (VIF) are calculated for each variable. Results reported

in Panel C show that the VIFs of all the variables are far below critical the value of 10. As a

result, multicollinearity should not be a significant concern.

1.3 Empirical Results

The main focus of this research is on the impact of market power on two aspects of a firm‟s

dividend policy: the decision to pay a dividend and the level of the dividend payment. To give a

complete picture, all three measures of market power are considered both with and without

controlling for other firm characteristics. The reason for this is that firm characteristics are

themselves endogenous to the market structure in which the firm operates. After establishing the

positive impact of market power on a firm‟s dividend policy, our paper shows that this relation

can be further explained by the effect of a firm‟s market power on its future business risk.

Following Fama and French (2001), a multivariate logit model is estimated, where the

dependent variable is a payer dummy, which equals one if firm i pays dividends in year t and

zero otherwise. Other firm characteristics documented in the literature, such as firm size,

profitability, growth, retained earnings over total assets, and current risks are also used as

controls. The Fama and MacBeth (1973) procedure is applied for the estimation: the cross-

sectional logit regression is estimated year-by-year and the reported coefficients are the time

series averages of the annual estimates. The standard errors of the coefficients are constructed

from the time series of the standard deviations of the annual estimated coefficients and are

adjusted for autocorrelation using a Newey-West (1987) adjustment of two lags.

The results from the analysis are presented in Table 1.3. In more detail, the models presented in

Panel A of Table 1.3 replicate regression specifications already discussed in the literature. In

particular, Models A.1 and A.2 follow the specifications given in Fama and French (2001).

Models A.3 and A.4 add RE/TA, the ratio of cumulative retained earnings to total assets, as an

explanatory variable according to the specification given by DeAngelo et al. (2006). Models A.5

15

and A.6 are similar to Hoberg and Prabhala (2009), which considers current stock return

volatility, and Model A.7 pools all these variables together. The results are consistent with

previous findings that larger firms, more profitable firms, firms with fewer growth opportunities,

firms with higher accumulated earnings, and firms with lower current risk are more likely to pay

dividends.

Models in Panel B investigate the explanatory power of various market power measures without

controlling for other firm characteristics. All three measures of market power have significant

explanatory power for a firm‟s decision to pay a dividend. Their signs are consistent with the

prediction that a firm with higher market power, demonstrated either by being a part of a more

concentrated industry or by being a part of an industry with a smaller influx of foreign products

or by being able to charge the price of its product above the marginal cost, is more likely to pay

a dividend. Moreover, the test of the joint significance of the import penetration and the HHI

variables in Model B.4 shows that the explanatory power of both measures remains significant,

which is consistent with the fact that these two measures capture different aspects of market

power.

To assess the robustness of these results, Panel C shows the results where control variables are

structured in the basic form given by Fama and French (2001). These specifications are a

particularly hard test for the model, since the control variables themselves are endogenous, that

is, they are a function of the firm‟s market power. However, the results regarding the effect of

import penetration remain robust. The coefficient is significant and it has the correct sign in both

specifications of the control variables. When attention is turned to the Lerner index, even though

its explanatory power decreases, it still retains significant with the correct sign, even when all

the control variables are pooled together. Moreover, considering the high correlation between

the Lerner index and profitability, if the profitability variable is not included in the regression,

the explanatory power of the Lerner index increases notably. On the other hand, the coefficient

estimates of the HHI are insignificant, but with the correct sign. This result may be partially

attributed to the lack of variation of the Census-based HHI; data are available only for the years

1982, 1987, and 1992.

16

To summarize, the results of the logit models presented in Table 1.3 exhibit a positive relation

between market power and the decision to pay a dividend. Furthermore, among the three

measures of market power, the explanatory powers of import penetration and the Lerner index

are found to be significant and robust to the different specification of control variables, whereas

the performance of the HHI is found to be sensitive to the inclusion of control variables.

The results tabulated in Table 1.4 present tests of the relationship between a firm‟s market

power and the level of its dividend payment. The dependent variable is the common dividend

scaled by total assets. Time fixed-effects are used to control for common trends across all firms.

Since import penetration and the HHI are industry-level measures, the industry fixed-effects are

only applied for the regressions where market power is measured by the Lerner index. As

suggested by Petersen (2009), clustered robust standard errors are used to account for within-

firm correlation of the error terms, i.e. the observations are assumed to be independent across

firms, but not within firms. Since common dividends cannot be negative, estimating the model

by ordinary least squares (OLS) will lead to biased estimates of the coefficients. In order to

circumvent this problem, the research design of Fenn and Liang (2001) is followed using a one-

sided Tobit model for dividend payments (censored at zero). The marginal effects are reported

in Table 1.4.11

The models in Panel A examine the explanatory power of the three measures of market power

without controlling for other firm characteristics. The results are consistent with the prediction

that firms with higher market power, as defined by any of the three measures, pay more

dividends.

The models in Panel B include other firm characteristics that are relevant to dividend policy.

Model [1] only examines standard firm characteristics. It shows that firms that are either larger

or more profitable or with higher RE/TA pay more dividends; firms with higher asset growth

rates and currently riskier firms pay lower dividends.12

Models [2]-[5] of Panel B investigate the

11

We apply the OLS regression for all the specifications and get qualitatively similar results.

12 The positive coefficient estimate of the market-to-book ratio seems contrary to the implication of the expected

growth opportunities as found in Table 3. However, this positive coefficient on market-to-book has also been

17

impact of the various market power measures on dividend payments after controlling for other

firm characteristics. Although the explanatory power of the HHI becomes unstable and

insignificant, coefficient estimates of import penetration and the Lerner index do remain

statistically significant with the expected sign even after including all control variables in the

specification. Considering the high correlation between profitability and the Lerner index, the

explanatory power of the Lerner index with all control variables except profitability is examined

in Model [6]. Although the coefficient estimates of other variables are unchanged, the

explanatory power of the Lerner index almost doubles.

The results presented in Table 1.3 and Table 1.4 show that the impact of market power on a

firm‟s dividend policy is sensitive to the measure of market power used in the analysis. The

explanatory power of the HHI is the least robust, which can be explained either by theoretical or

empirical arguments. Theoretically, the classic „high HHI – high market power‟ argument is

challenged by the contestable market theory proposed by Baumol (1982). This theory argues

that the threat of entry ensures that market power is constrained or eliminated; hence an industry

with few firms may still be competitive due to the existence of potential entrants. In recent years,

both globalization and deregulation have increased the threat of entry, which weakens the link

between the HHI and market power. Nevertheless, the HHI still has strong explanatory power

when control variables are not considered, making this reasoning less relevant. Empirically, the

Census-based HHI used in our study lacks time variation since the data is available every five

years during a short sample period. Thus, the impact of market power measured by the HHI may

have been absorbed into other dividend determinants such as profitability, size, and lower

current risk, which in turn weakens the direct explanatory power of the HHI.

The empirical evidence suggests that market power has a positive impact on a firm‟s dividend

policy – a firm with higher market power is both more likely to pay a dividend and pay more.

documented in Fama and French (2002) and Li and Zhao (2008). A firm‟s market value measures two parts: the

value of growth opportunities and the value of assets in place. The second part implies that market-to-book ratio

also contains information about current profitability. Fama and French (2002) argue that the positive coefficient of

market-to-book may indicate that M/B contains more information about current profitability and less information

about future growth options for dividend payers. This interpretation is consistent with the view given by Grullon et

al. (2002) that a dividend payer is usually in its mature stage, where assets in place are the dominant determinant of

its market value.

18

However, what is left unanswered is how? Microeconomic theory shows that a firm with market

power is able to smooth the fluctuation of its operating performance and sustain superior

profitability when facing unexpected economy-wide and/or firm-specific future shocks, i.e. high

market power helps reduce a firm‟s business risk. With the expectation of more stable and

stronger future cash flow, managers would be more confident about their ability to pay and

maintain a dividend payment. This leads to the hypothesis that a firm with higher market power

has lower business risk.

The models in Panels A and B of Table 1.5 test the validity of this hypothesis by investigating

the impact of a firm‟s market power at year t on the stability and level of its operating

performance during the next five years [t+1, t+5]. Two measures of operating performance are

chosen consistent with Barber and Lyon (1996). The first measure is the return on assets (ROA),

defined as the operating income before depreciation scaled by the average of the beginning- and

ending- period book value of total assets. However, operating income is an accrual-based

measure, which is subject to potential earnings manipulations. In order to avoid the bias

introduced by potential earnings manipulations, Barber and Lyon (1996) suggest using the cash-

flow return on assets (CFROA), defined as the operating cash flow scaled by the average of

beginning- and ending-period book value of total assets (see Appendix 1.C for details).

Regression estimates tabulated in Panels A and B support the predictions that a firm with higher

market power will have more stable and better operating performance in the future, regardless of

which measure of operating performance is used. The predictive power of both import

penetration and the Lerner index is robust to the inclusion of controls. It can also be noticed that

coefficient estimates of the HHI suggest an opposite effect on risk, when other firm controls are

used. To some extent, the relative performance of the three market power measures on

predicting future business risks appears to be consistent with their relative performance in the

dividend policy tests. Coefficient estimates of the control variables show that more profitable,

larger firms with more retained earnings have more stable and better performance in the future,

whereas firms with high growth rates or currently riskier firms have higher business risk. The

market-to-book variable is excluded from the regression specifications, since it contains the

19

mixture of information on assets in place and future growth opportunities, which have opposite

impacts on a firm‟s business risk in the future.

It is of interest to examine whether the aforementioned results are stable over time or sensitive

to different ways of constructing the variables.

Table 1.6 reports subsample analysis. As mentioned in Variable Definitions section, the values

of import penetration used in the above regressions are constructed from combining two datasets

according to the first approach, which takes all the values of import penetration calculated with

recently updated 1989-2002 dataset and requires a comparison at 1989 to decide whether we can

take 1972-1988 values from the [1972, 1996] dataset. Appendix 1.B clearly shows that many

industries have been excluded from the 1972-1988 period due to this requirement. To ensure

that our previous results are not driven by the sample selection bias caused by the requirement

used to construct import penetration for the entire sample period, we relax this constrain in the

subsample analysis. We divide the sample into two subsamples: [1972, 1988] and [1989, 2002].

In the 1972-1988 subsample, we will consider all the observations from industries with available

values of import penetration in 1972-1996 dataset. The general results are consistent with the

findings in Table 1.4. Moreover, it is interesting to find that the explanatory power of import

penetration is stronger in the second period, which is consistent with the argument that the

increased levels of competition originating from globalization have become more important.

To check the robustness of our results with respect to the construction of import penetration for

the entire sample period, the regressions presented in Table 1.3 and Table 1.4 were repeated by

using import penetration constructed according to the second option, which requires comparison

at 1996 to decide whether use the new dataset for the period 1997-2002. The results, not

presented in the paper, are similar.

In Table 1.3, the approach of Fama and French (2001) was followed using the dividend per

share (Compustat DATA26) to identify the dividend payer. However, when investigating the

level of dividend payments in Table 1.4, Fama and French (2002) are followed using common

dividend (Compustat DATA21), scaled by total assets. The analysis presented in Table 1.3, was

20

repeated by defining dividend payer dummy according to common dividends (Compustat

DATA21). The results, not presented in the paper, are similar.

Finally, other combinations of control variables were used when testing for the impact of market

power on a firm‟s decision to pay dividends. In particular, all the specifications of firm

characteristics given in Panel A of Table 1.3 are considered separately for each of the measures

of market power. The results, not reported in the paper, are qualitatively consistent with those

reported in Panel C, where import penetration offers the most robust explanation and the HHI is

not robust to the inclusion of controls.

1.4 Conclusion

This paper investigates whether a firm‟s market power in its product market affects its dividend

policy, and if so, how this is achieved. The study is based on a sample of manufacturing firms

from 1972-2002 and considers three measures of market power suggested in the literature of

Industrial Organization and International Trade. The empirical findings generally support the

hypothesis that a firm with higher market power is more likely to make a dividend payment and

pay more when it does. Moreover, this positive impact can be explained by the fact that firms

with higher market power will expect better and more stable operating performance in the near

future, which implies lower business risk. The three measures of market power differ in their

performance in the tests, where import penetration and the Lerner index offer the most robust

explanation while the HHI is not robust to the inclusion of controls.

The findings of this paper are also consistent with recent empirical observations. The recent

evidence on various changes in dividend policy can be related to the fundamental changes that

have occurred in the competitive environment during the past two decades. Furthermore, the

risk-based explanation suggested in this paper also proposes a potential link between two

contemporaneous trends – the increasing conservatism in dividend policy and the increasing

idiosyncratic volatility.

21

Table 1.1: Market Power Measures

These tables show descriptive statistics and pairwise correlations of three market power measures for four-digit

SIC-based industries. Import penetration (IP) and Herfindahl-Hirschman Index (HHI) are industry-level measures

of market power. Import penetration (IP) is defined as the proportion of domestic demand that is satisfied by

imports. The Herfindahl-Hirschman Index (HHI) measures the domestic market concentration. Firm-level measure

of market power, the Lerner index, is aggregated at industry-level using equal-weight average (Lerner(EW)) and

sales-based value-weight average (Lerner(VW)) in each industry each year. Lerner(Median) is defined as the

median of the Lerner index across all firms in each industry each year. P-value is reported in the bracket.

Panel A: Descriptive Statistics

variable Mean Median

Standard

Deviation 25th

Percentile 75th

Percentile N

IP 0.169 0.104 0.189 0.030 0.239 6445

HHI 722.70 498 657.58 243 988 3081

Lerner(EW) 0.094 0.105 0.117 0.072 0.141 6445

Lerner(Median) 0.100 0.106 0.113 0.075 0.140 6445

Lerner(VW) 0.116 0.117 0.102 0.084 0.156 6445

Panel B: Industry EW Average

IP HHI Lerner(EW)

IP 1

HHI -0.024 1

[0.187]

Lerner(EW) -0.050 0.057 1

[0.000] [0.002]

Panel C: Industry VW Average

IP HHI Lerner(VW)

IP 1

HHI -0.024 1

[0.187]

Lerner(VW) -0.027 0.073 1

[0.037] [0.000]

Panel D: Industry Median

IP HHI Lerner(Median)

IP 1

HHI -0.024 1

[0.187]

Lerner(Median) -0.038 0.043 1

[0.002] [0.016]

22

Table 1.2: Summary Statistics

These tables present summary statistics and correlations matrix of the firm-level variables used in the regressions.

The sample includes all manufacturing firms with available import penetration or the HHI data during 1972-2002

after screening according to Fama and French (2001) criteria. Variables definitions are given in Appendix 1.C.

Panel A presents summary statistics of firm characteristics. Panel B presents the correlations between all the firm-

level explanatory variables. P-value is reported in the bracket. Panel C presents the result of multicollinearity test.

Panel A: Summary Statistics

Variable Mean Median

Standard

Deviation

25th

Percentile

75

Percentile N

Dividend Payer 0.609 1.000 0.488 0.000 1.000 27520

Div/TA 0.015 0.009 0.030 0.000 0.022 27503

Profitability 0.056 0.077 0.110 0.040 0.108 27520

LnAsset 5.148 5.029 2.078 3.592 6.606 27520

NYP 4.029 3.000 3.206 1.000 7.000 27520

M/B 1.460 1.139 1.036 0.903 1.592 27520

AGR 0.112 0.071 0.257 -0.012 0.171 27520

RE/TA 0.221 0.306 0.472 0.129 0.449 27520

RETVOL 0.124 0.107 0.074 0.078 0.149 27457

Lerner 0.087 0.108 0.201 0.064 0.156 27467

Panel B: The Correlation Matrix

Profitability LnAsset NYP M/B AGR RE/TA RETVOL Lerner

LnAsset 0.237 1

[<0.001]

NYP 0.195 0.818 1

[<0.001] [<0.001]

M/B -0.081 -0.022 0.196 1

[<0.001] [<0.001] [<0.001]

AGR 0.239 0.034 0.051 0.208 1

[<0.001] [<0.001] [<0.001] [<0.001]

RE/TA 0.657 0.270 0.183 -0.260 0.058 1

[<0.001] [<0.001] [<0.001] [<0.001] [<0.001]

RETVOL -0.348 -0.356 -0.277 0.107 0.008 -0.442 1

[<0.001] [<0.001] [<0.001] [<0.001] [0.183] [<0.001]

Lerner 0.750 0.291 0.210 -0.158 0.133 0.605 -0.315 1

[<0.001] [<0.001] [<0.001] [<0.001] [<0.001] [<0.001] [<0.001]

Panel C: Multicollinearity Diagnostic

Variable Profitability Lerner RE/TA LnAsset RETVOL M/B AGR

VIF 3.03 2.60 2.21 1.58 1.53 1.31 1.16

1/VIF 0.330 0.384 0.453 0.633 0.653 0.762 0.862

23

Table 1.3: The Decision to Pay Dividends as a Function of Market Power

This table reports the coefficient estimates of multivariate logit regressions that investigate the impacts of market

power and firm characteristics on a firm‟s decision to pay dividends. The dependent variable is a dividend payer

dummy, equal to one if firm i pays dividend (DATA 26) in year t and zero otherwise. Three measures of market

power are tested – import penetration (IP), the Herfindahl-Hirschman index (HHI), and the accounting approximate

of the Lerner index (Lerner). Other explanatory variables include: firm size measured by its NYSE decile (NYP),

Profitability, Growth rate of assets (AGR), market-to-book ratio (M/B), the ratio of retained earnings over total

assets (RE/TA), and current stock return volatility (RETVOL). Details on variable construction are given in

Appendix 1.C. Panel A reports the replications of related studies in the literature. Models in Panel B investigate the

impact of market power, without any controls, on a firm‟s decision to pay dividends. Panel C presents the results

about the impact of market power on a firm‟s decision to pay dividends, with some control variables. Fama and

MacBeth (1973) methodology is applied for the estimation, where cross-sectional logit regression is conducted

year-by-year and the reported coefficients are the time series averages of the annual estimates. The standard errors

of the coefficients are constructed from time series standard deviations of annual estimated coefficients and are

adjusted for autocorrelation using a Newey-West (1987) adjustment to two lags. t-statistics are presented in italic.

Panel A. Replicating Regressions Documented in the Literature

Model NYP Profitability AGR M/B RE/TA RETVOL Intercept N

A.1 0.465 7.662 -1.351 -1.257 27520

11.87 7.87 -6.24 -14.28

A.2 0.541 9.923 -0.825 -0.889 -0.586 27520

13.86 7.64 -3.29 -6.99 -4.62

A.3 0.439 2.662 -0.894 4.401 -2.122 27520

12.59 3.39 -4.1 12.78 -17.24

A.4 0.499 4.707 -0.814 4.480 -1.557 27520

15.82 4.54 -12.35 15.41 -15.14

A.5 0.381 6.468 -1.040 -18.542 1.257 27457

10.95 6 -5.55 -19.89 10.2

A.6 0.446 8.335 -0.635 -0.692 -17.391 1.609 27457

12.65 6.15 -2.68 -6.42 -18.72 11.05

A.7 0.436 4.366 -0.337 -0.664 3.793 -13.944 0.366 27457

14.55 4.1 -1.59 -9.1 13.22 -16.77 2.38

24

Panel B. The Decision to Pay Dividends as a Function of Market Power (No Controls)

Model IP HHI Lerner Intercept N

B.1 -1.756 0.765 27520

-21.18 7.03

B.2 3.064 0.095 12962

5.33 1.03

B.3 8.446 -0.363 27467

9.92 -4.75

B.4 -1.831 3.617 0.385 12962

-13.88 5.88 4.99

Panel C. The Decision to Pay Dividends as a Function of Market Power (With Controls)

Model IP HHI Lerner NYP Profitability AGR M/B RE/TA RETVOL Intercept N

C.1 -1.171 0.462 7.650 -1.362 -1.029 27520

-6.96 11.83 7.76 -6.33 -9.83

C.2 -1.169 0.436 4.303 -0.332 -0.681 3.846 -13.580 0.556 27457

-6.4 14.8 3.98 -1.56 -9.3 13.91 -16.45 3.46

C.3 0.294 0.515 6.166 -1.608 -1.454 12962

0.51 12.48 12.22 -9.71 -11.8

C.4 0.364 0.476 2.460 -0.581 -0.554 3.853 -14.860 0.149 12944

0.39 11.71 5.94 -4.18 -28.35 21.82 -17.23 1.02

C.5 1.271 0.458 6.527 -1.379 -1.280 27467

3.91 11.8 6.61 -6.35 -15.33

C.6 0.786 0.433 3.822 -0.356 -0.680 3.773 -14.045 0.375 27404

2.16 14.58 3.48 -1.66 -9.49 13.53 -16.85 2.47

C.7 2.531 0.428 -0.247 -0.637 3.889 -14.047 0.391 27404

3.93 14.52 -1.09 -9.85 13.44 -16.52 2.5

C.8 -1.087 0.713 0.510 6.132 -1.612 -1.253 12962

-4.82 1.13 12.26 11.64 -9.53 -9.16

C.9 -1.268 0.842 0.474 2.340 -0.565 -0.571 3.919 -14.515 0.344 12944

-6.18 0.83 11.81 5.33 -3.92 -25.67 23.34 -17.69 2.33

25

Table 1.4: Level of Dividend Payment as a Function of Market Power

This table presents the results of investigating the impact of a firm‟s market power, without and with controls, on a

firm‟s dividend payment level. The dependent variable is dividend payment, defined as common dividend

(Compustat data 21) at year t scaled by total assets (Compustat data 6) at year t. Three measures of market power

are tested – import penetration (IP), the Herfindahl-Hirschman index (HHI), and the accounting approximate of the

Lerner index (Lerner). Other explanatory variables include: firm size measured by the logarithm of total assets

(lnAsset), Profitability, Growth rate of assets (AGR), market-to-book ratio (M/B), the ratio of retained earnings

over total assets (RE/TA) and current stock return volatility (RETVOL). Details on variable construction are given

in Appendix 1.C. Coefficients are estimated by one-sided Tobit regression (censored at zero). Marginal Effects

(ME) evaluated at the means of the independent variables are reported. Clustered robust standard errors are used to

account for within-firm correlation of the error terms. Year fixed-effects are included for all the regressions.

Industry fixed-effects, based on two-digit SIC, are excluded in regressions that use import penetration or/and the

HHI to measure market power.

Panel A: One-sided Tobit Model (no control variables)

Dependent Variable: Div(t)/Total Assets(t)

[1] [2] [3] [4]

ME t ME t ME t ME t

IP -0.020 -5.95 -0.022 -4.93

HHI 0.027 2.98 0.032 3.55

Lerner 0.078 12.29

Year dummy yes yes yes yes

Industry dummy no no yes no

log likelihood 22367.09 7390.94 26340.34 7507.31

N 27503 12955 27450 12955

#cluster 2859 1862 2846 1862

Panel B: One-sided Tobit Model (with control variables)

Dependent Variable: Div(t)/Total Assets(t)

[1] [2] [3] [4] [5] [6]

ME t ME t ME t ME t ME t ME t

IP -0.006 -4.38 -0.005 -3.13

HHI -0.001 -0.15 0.001 0.25

Lerner 0.013 3.44 0.023 6.41

profitability 0.040 5.09 0.032 5.22 0.035 3.07 0.025 4.85 0.028 3.04

lnAsset 0.002 9.92 0.002 10.38 0.002 8.02 0.002 9.70 0.002 8.01 0.002 9.45

M/B 0.002 4.23 0.001 3.31 0.002 3.38 0.002 4.46 0.001 3.18 0.002 4.95

AGR -0.009 -5.68 -0.007 -5.74 -0.011 -3.99 -0.008 -6.48 -0.009 -3.98 -0.008 -6.31

RE/TA 0.023 14.99 0.017 15.14 0.023 10.94 0.023 14.18 0.019 10.93 0.024 14.46

RETVOL -0.066 -8.30 -0.053 -8.59 -0.089 -5.88 -0.062 -10.06 -0.071 -5.86 -0.064 -10.02

Year dummy yes yes yes yes yes yes Industry

dummy yes no no yes no yes

log

likelihood 27686.41 27479.18 9765.59 29758.16 9782.38 29698.24

N 27440 27440 12937 27387 12937 27387

#cluster 2854 2854 1858 2841 1854 2841

26

Table 1.5: Testing the Risk-Based Explanation

This table presents the results of investigating the impact of a firm‟s market power and other characteristics at year t on the volatility and mean of its operating

performance during the next five years [t+1, t+5]. Following Barber and Lyon (1996), operating performance is measured by two variables: return on assets

(ROA) and cash-flow return on assets (CFROA). ROA is defined as the operating income before depreciation (DATA13) scaled by the average of beginning-

and ending- period book value of total assets (DATA6). CFROA is defined as the operating cash flow scaled by the average of beginning- and ending-period

book value of total assets. The operating cash flow is equal to the operating income before depreciation (item 13) plus the decrease in receivables (item 2), the

decrease in inventory (item 3), the increase in accounts payable (item 70), the increase in other current liabilities (item 72), and the decrease in other current

assets (item 68). In Panel A, stability of future operating performance is measured by the standard deviation of ROA (σ(ROA)) and the standard deviation of

CFROA (σ(CFROA)) during year [t+1, t+5]. In Panel B, level of future operating performance is measured as the mean of ROA and mean of CFROA during

year [t+1, t+5]. All explanatory variables are calculated at year t. Three measures of market power are import penetration (IP), the Herfindahl-Hirschman index

(HHI), and the accounting approximate of the Lerner index (Lerner). Other explanatory variables include: firm size measured by the logarithm of total assets

(lnAsset), Profitability, Growth rate of assets (AGR), the ratio of retained earnings over total assets (RE/TA) and current stock return volatility (RETVOL).

Coefficients are estimated by OLS regression. Clustered robust standard errors are used to account for within-firm correlation of the error terms. Year fixed-

effects are included for all the regressions. Industry fixed-effects, based on two-digit SIC, are excluded in regressions that use import penetration or/and the HHI

to measure market power.

Panel A: the Impact of Market Power on the Stability of Future Operating Performance

σ(ROA)

[1] [2] [3] [4] [5] [6] [7]

Coef. t Coef. t Coef. t Coef. t Coef. t Coef. t Coef. t

IP 0.028 5.10 0.012 2.72

HHI -0.027 -1.67 0.032 2.58

Lerner -0.086 -17.59 -0.027 -5.17

Profitability -0.056 -6.63 -0.057 -6.62 -0.063 -5.96 -0.024 -2.67

lnAsset -0.006 -18.13 -0.006 -18.38 -0.007 -14.25 -0.006 -17.00

AGR 0.008 4.27 0.008 4.45 0.009 3.39 0.007 4.04

RE/TA -0.018 -7.07 -0.019 -7.30 -0.017 -5.00 -0.016 -6.19

RETVOL 0.090 8.43 0.096 9.05 0.114 6.82 0.090 8.50

year dummy yes yes yes yes yes yes yes

industry dummy no no yes yes no no yes

R2 0.0203 0.0052 0.1723 0.3095 0.297 0.3137 0.3099

N 19805 9498 19787 19766 19766 9488 19748

#cluster 2028 1328 2021 2022 2022 1324 2015

27

σ(CFROA)

[1] [2] [3] [4] [5] [6] [7]

Coef. t Coef. t Coef. t Coef. t Coef. t Coef. t Coef. t

IP 0.045 5.83 0.026 4.09

HHI -0.055 -2.70 0.014 0.85

Lerner -0.111 -17.26 -0.047 -6.37

Profitability -0.076 -7.08 -0.075 -6.84 -0.088 -6.54 -0.023 -2.02

lnAsset -0.008 -15.71 -0.008 -15.87 -0.008 -13.3 -0.007 -14.53

AGR 0.007 2.87 0.007 2.88 0.010 3.07 0.006 2.77

RE/TA -0.023 -6.67 -0.023 -6.71 -0.019 -4.48 -0.019 -5.51

RETVOL 0.082 7.06 0.088 7.59 0.107 5.57 0.083 7.14

Year dummy yes yes yes yes yes yes yes

Industry dummy no no yes yes no no yes

R2 0.0287 0.0133 0.1695 0.2607 0.2496 0.254 0.2631

N 19644 9409 19626 19605 19605 9399 19587

#cluster 2014 1316 2007 2008 2008 1312 2001

28

Panel B: the Impact of Market Power on the Level of Future Operating Performance mean(ROA)

[1] [2] [3] [4] [5] [6] [7]

Coef. t Coef. t Coef. t Coef. t Coef. t Coef. t Coef. t

IP -0.043 -3.46 -0.021 -2.41

HHI 0.023 0.57 -0.046 -1.50

Lerner 0.276 27.54 0.131 12.58

profitability 0.405 25.15 0.413 24.86 0.377 17.98 0.256 14.02

lnAsset 0.004 5.72 0.004 5.46 0.006 6.56 0.003 4.01

AGR -0.016 -5.00 -0.014 -4.41 -0.017 -3.54 -0.013 -4.31

RE/TA 0.030 7.03 0.030 6.99 0.027 4.95 0.022 5.10

RETVOL -0.063 -4.08 -0.065 -4.18 -0.068 -2.61 -0.065 -4.27

year dummy yes yes yes yes yes yes yes

industry dummy no no yes yes no no yes

R2 0.0506 0.0103 0.3334 0.3706 0.3524 0.3283 0.39

N 19805 9498 19787 19766 19766 9488 19748

#cluster 2028 1328 2021 2022 2022 1324 2015

mean(CFROA)

[1] [2] [3] [4] [5] [6] [7]

Coef. t Coef. t Coef. t Coef. t Coef. t Coef. t Coef. t

IP -0.049 -4.07 -0.024 -2.82

HHI 0.078 2.00 -0.011 -0.36

Lerner 0.255 27.71 0.127 13.06

profitability 0.324 21.6 0.332 21.37 0.310 16.3 0.181 10.73

lnAsset 0.008 9.92 0.007 10.01 0.009 9.31 0.006 8.22

AGR -0.025 -8.96 -0.025 -8.56 -0.030 -7.08 -0.024 -8.61

RE/TA 0.033 8.38 0.033 8.41 0.029 5.87 0.023 6.22

RETVOL -0.055 -3.85 -0.063 -4.27 -0.074 -2.99 -0.059 -4.13

year dummy yes yes yes yes yes yes yes

industry dummy no no yes yes no no yes

R2 0.0331 0.0114 0.3215 0.367 0.3453 0.3431 0.3862

N 19644 9409 19626 19605 19605 9399 19587

#cluster 2014 1316 2007 2008 2008 1312 2001

29

Table 1.6: Subsample Analysis

This table presents the results of investigating the impact of a firm‟s market power and other characteristics on a

firm‟s dividend payment level for two subsamples, [1972, 1988] and [1989, 2002]. The dependent variable is

dividend payment, defined as common dividend (Compustat data 21) at year t scaled by total assets (Compustat

data 6) at year t. Three measures of market power are tested – import penetration (IP), the Herfindahl-Hirschman

index (HHI), and the accounting approximate of the Lerner index (Lerner). For [1972, 1988] subsample, import

penetration comes from the 1972-1996 import penetration dataset posted on Peter Schott‟s website. For [1989,

2002] subsample, import penetration is calculated based on recently updated 1989-2005 multilateral imports and

exports dataset provided by Peter Schott and the NBER-CES Manufacturing Industry Database. Other explanatory

variables include: firm size measured by the logarithm of total assets (lnAsset), Profitability, Growth rate of assets

(AGR), market-to-book ratio (M/B), the ratio of retained earnings over total assets (RE/TA) and current stock

return volatility (RETVOL). Details on variable construction are given in Appendix 1.C. I estimate the coefficients

by one-sided Tobit regression (censored at zero). Marginal Effects evaluated at the means of the independent

variables are reported. Clustered robust standard errors are used to account for within-firm correlation of the error

terms. Year fixed-effects are included for all the regressions. Industry fixed-effects, based on two-digit SIC, are

excluded in regressions that use import penetration or/and the HHI to measure market power

Panel A: 1972 to 1988

Dependent Variable: Div(t)/Total Assets(t)

[1] [2] [3] [4] [5] [6]

ME t ME t ME t ME t ME t ME t

IP -0.014 -4.27 -0.003 -1.58

HHI 0.020 2.66 -0.003 -0.64

Lerner 0.089 17.75 0.016 3.95

profitability 0.048 10.32 0.027 5.67 0.035 6.41

lnAsset 0.002 14.48 0.002 11.20 0.002 13.08

M/B 0.003 6.88 0.002 5.47 0.003 6.58

AGR -0.008 -8.80 -0.008 -5.74 -0.009 -9.06

RE/TA 0.032 19.79 0.024 11.90 0.031 19.04

RETVOL -0.063 -15.33 -0.065 -8.20 -0.061 -14.57

Year dummy yes yes yes yes yes yes

Industry dummy no no yes no no yes

Log likelihood 26426.17 6622.07 28869.44 32551.43 8708.10 32718.61

N 21354 7475 21328 21285 7450 21259

#cluster 2504 1668 2497 2498 1663 2491

30

Panel B: 1989 to 2002

Dependent Variable: Div(t)/Total Assets(t)

[1] [2] [3] [4] [5] [6]

ME t ME t ME t ME t ME t ME t

IP -0.023 -5.31 -0.008 -4.04

HHI 0.038 3.31 0.002 0.39

Lerner 0.045 7.82 0.008 1.92

profitability 0.027 2.92 0.034 2.10 0.011 2.39

lnAsset 0.002 7.19 0.002 6.57 0.001 6.20

M/B 0.001 1.64 0.001 2.04 0.001 2.98

AGR -0.008 -4.01 -0.010 -2.92 -0.006 -4.44

RE/TA 0.017 9.76 0.021 8.31 0.016 9.00

RETVOL -0.070 -6.03 -0.106 -4.82 -0.054 -6.62

Year dummy yes yes yes yes yes yes

Industry dummy no no yes no no yes

Log likelihood 5913.16 3119.04 7750.39 7888.78 4307.09 9073.78

N 12525 7449 12484 12501 7441 12460

#cluster 1825 1370 1814 1822 1367 1811

31

Appendix 1.A: Sample Selection

We follow Fama and French (2001) for sample screening. The CRSP-Compustat sample for

calendar year t includes those firms with fiscal year-ends in t that have the following data

(COMPUSTAT data items in parentheses): total assets (6), stock price (199) and shares

outstanding (25) at the end of the fiscal year, income before extraordinary items (18), interest

expense (15), dividends per share by ex date (26), retained earnings13

(36),and (a) preferred

stock liquidating value (10), (b) preferred stock redemption value (56), or (c) preferred stock

carrying value (130). Firms must also have (a) stockholder's equity (216), (b) liabilities (181), or

(c) common equity (60) and preferred stock par value (130). Total assets must be available in

year t and t-1. The other items must be available in t. Balance sheet deferred taxes and

investment tax credit (35), income statement deferred taxes (50), purchases of common and

preferred stock (115), sales of common and preferred stock (108), and common treasury stock

(226) are used, but not required. Firms with book equity below $250,000 or assets (current or

lagged) below $500,000 are excluded. To ensure that firms are publicly traded, the securities

have to have CRSP share codes of 10 or 11.

13

Fama and French (2001) requirement on dividends for preferred stocks (19) is ignored, while the existence of

retained earnings (36) is needed to construct RE/TA ratio as in DeAngelo et al. (2006).

32

Appendix 1.B: Import Penetration in Manufacturing Industries

Mean (Median) is mean (median) import penetration across four-digit SIC (1987 version)

manufacturing industries for a given year.

Approach 1) Compare at 1989 Approach 2) Compare at 1996

Year Number of Industries Mean Median Number of Industries Mean Median

1972 237 7.88% 3.81% 386 6.63% 3.54%

1973 237 8.37% 4.10% 386 7.03% 3.72%

1974 237 8.75% 4.98% 386 7.39% 4.04%

1975 237 8.36% 4.71% 386 7.15% 4.21%

1976 237 8.98% 5.23% 386 7.70% 4.69%

1977 237 9.35% 5.53% 386 7.90% 4.91%

1978 237 10.61% 5.91% 386 9.01% 5.37%

1979 237 10.95% 6.39% 386 9.27% 5.67%

1980 237 11.30% 6.76% 386 9.51% 5.76%

1981 237 11.78% 7.32% 386 9.88% 6.38%

1982 237 11.91% 7.42% 386 9.96% 6.27%

1983 237 12.71% 8.30% 386 10.73% 6.60%

1984 237 14.70% 10.41% 386 12.47% 7.77%

1985 237 15.97% 11.26% 386 13.56% 9.05%

1986 237 17.44% 12.17% 386 14.94% 9.78%

1987 237 17.95% 12.37% 386 15.26% 10.15%

1988 237 18.79% 12.84% 386 15.94% 10.54%

1989 405 16.26% 10.18% 405 15.84% 11.04%

1990 405 16.86% 10.03% 405 16.38% 11.14%

1991 405 17.44% 10.97% 405 16.88% 11.41%

1992 402 18.03% 11.36% 404 17.77% 11.36%

1993 402 18.51% 11.84% 404 18.06% 11.85%

1994 400 19.75% 12.69% 402 19.18% 13.31%

1995 398 20.98% 13.03% 400 21.23% 13.60%

1996 397 21.31% 13.58% 399 21.51% 14.65%

1997 383 21.39% 15.54% 317 22.32% 16.26%

1998 383 22.32% 15.77% 317 23.33% 16.47%

1999 383 23.19% 16.36% 317 24.16% 17.30%

2000 381 24.46% 17.76% 316 25.55% 18.14%

2001 379 24.57% 18.01% 313 25.43% 19.91%

2002 378 26.33% 19.73% 313 27.20% 20.49%

Note:

33

Two sets of import penetration data based on the NBER International Trade Database are available for four-digit

SIC-based manufacturing industries. Import penetration data for the period 1972-1996 are directly available on

Professor Peter Schott‟s website. Import penetration for the period 1989-2002 can be constructed from the updated

multilateral imports and exports database (1989-2005) and the NBER-CES Manufacturing Industry Database

(1958-2002). Two datasets are not perfectly comparable since the second trade dataset uses a revised version of

HS-SIC4 concordance, which provides a more complete match of the imports and exports for the relevant domestic

industries. Two approaches can be used to combine two datasets of import penetration while minimize the

inconsistency. In approach 1), we compare the values of import penetration in 1989 from both the 1989-2002

dataset and the 1972-1996 dataset, for a given industry. If the absolute value of their difference ratio ((IP from

1989-2002 dataset)/(IP from 1972-1996 dataset)-1) is found to be lower than 25%, we will select the values of

import penetration during 1972-1988 to come from the earlier dataset, and the values during 1989-2002 from new

dataset; otherwise, only the data from 1989-2002 dataset will be considered in our study. In approach 2), we take

all the values of import penetration from the 1972-1996 dataset, and then decide whether to extend the time series

to 2002 with data from the 1989-2002 dataset. This approach requires the comparison of the values of import

penetration during the year 1996, which is done by following the same decision criteria mentioned above.

34

Appendix 1.C: Variable Definitions

Variable Definition

Dividend variables

Dividend Payer

For any year t, dividend payer dummy is equal to 1 if dividend per share values (26) is positive,

0 otherwise.

Dividend payment (Div/TA)

For any year t, dividend payment is common dividends (21) scaled by total assets (6) for year t.

Market power measures

IP Import Penetration (IPit) = it

it it it

imports

shipments exports imports

HHI The Herfindahl-Hirschman index (HHI), defined as the sum of squared market shares of all the

firms in a given industry.

Lerner The Lerner index is approximated by the price-cost margin, defined as operating profits (before

depreciation, special items, interest, and taxes) over sales. The numerator is calculated as sales

(12) minus cost of goods sold (41) minus selling, general, and administrative expenses (189).

Whenever this calculation is not possible, operating income (178) is used.

Control variables

Profitability

(Interest Expense (15) + Income Before Extraordinary Items (18)+ Deferred Taxes (DATA50 if

available))/ Total Assets(6)

lnAsset Size measure used in the regression about the level of a firm‟s dividend payments

log(Total Assets (6))

NYP Size measure used in the regression about a firm‟s decision to pay dividends

The NYSE market capitalization decile that a firm belongs to, where 1 stands for the smallest

decile and 10 stands for the largest decile.

M/B Measure for growth opportunities

(Assets (6)-Book Equity(60) + Market Equity)/Assets (6), where Market Equity= Stock Price

(199) times Shares Outstanding (25)

AGR Asset growth rate AGR(t)=Asset(t)/Asset(t-1) -1

RE/TA Retained Earnings(36)/ Total Assets(6)

RETVOL Measure for current risk

the standard deviation of a firm‟s monthly returns during one-year period

Variables used in testing risk-based explanation

ROA Return on asset (ROA) is defined as the operating income before depreciation (13) scaled by the

average of beginning- and ending- period book value of total assets (6).

CFROA

Cash-flow return on assets (CFROA) is defined as the operating cash flow scaled by the average

of beginning- and ending-period book value of total assets. The operating cash flow is equal to

the operating income before depreciation (13) plus the decrease in receivables (2), the decrease

in inventory (3), the increase in accounts payable (70), the increase in other current liabilities

(72), and the decrease in other current assets (68).

All numbers in parentheses refer to the Compustat (Industrial Annual) item names.

35

Chapter 2 Increase in Cash Holdings: Pervasive or Sector-Specific

An intriguing phenomenon, widely documented in the business press over the past few years, is

the gradual stockpiling of cash by large U.S. firms in the aftermath of the economic downturn in

the early 2000s.14

Furthermore, high-tech firms, those operating in information technology and

pharmaceutical industries, attracted special attention due to a more rapid speed of cash

accumulation despite the „growth‟ nature of their business.15

In this paper I examine the time trend in corporate cash holdings from 1974 to 2007, with a

primary focus on the difference between high-tech and non-high-tech sectors. The motivation

for this is twofold. First, the high-tech sector has predominantly been in the growth stage of its

industry life-cycle over this period, reflected by a huge influx of young public firms during the

1980s and 1990s. This change has been driven by the development of capital markets,

particularly venture capital that has allowed more immature firms to go public (Fama and

French, 2004; Hall and Lerner, 2009). Besides increasing the weight of high-tech sector in the

universe of publicly listed firms, these new listings may also contribute to more significant

changes in the population characteristics of high-tech sector, compared to that of non-high-tech

sector. Second and more important, the distinguishing characteristic of high-tech firms,

compared to non-high-tech ones, is the importance of research and development (R&D), which

is necessary for them to survive competition and gain market power. The availability of

financial slack, cash as a vital part, is important for high-tech firms, especially junior ones since

they are more exposed to capital market frictions. This is due to the high information asymmetry

concerning their growth opportunities as well as their lack of collateral for secured, particularly

bank, financing (Myers, 1984; Hall, 2002). Recent research has shown that large cash balances

14

“Corporate cash balances and economic activity”, by Governor Kevin M Warsh of the US Federal Reserve, 18

July 2006; “Behind Those Stock Piles of Corporate Cash”, by Mark Hulbert, The New York Times, October 22,

2006; “Corporate America sits on its cash”, by Justin Baer, Financial Times, September 23 2008.

15 “Too much cash, too little innovation”, Business Week, July 18, 2005; “The tech sector is hogging the green

blanket”, by Jesse Eisinger, The Wall Street Journal, April 05, 2006.

36

increase the likelihood of winning patent races in the pharmaceutical industry (Schroth and

Szalay, 2009) and that young firms do rely more heavily on cash reserves to smooth their R&D

expenditures (Brown and Petersen, 2009). Hence, difference in new listing effect and the

importance of cash reserve for R&D spending may imply a potential difference in the

evolutionary paths of corporate cash holdings between these two sectors.

I investigate the trend in cash holdings by examining the annual average cash-to-assets ratio

(cash ratio) over time. Like Bates et al. (2009), I find that U.S. firms on average have increased

their cash ratio since 1980. Using the official definition of high-tech industries offered by the

U.S. Department of Commerce, I split the sample of publicly listed firms into high-tech and

non-high-tech sectors, and find that the average cash ratio has increasingly differed across the

two sectors since 1980. The average cash ratio for the high-tech sector has more than tripled;

increasing from 11.2% in 1980 to 39.1% by 2007. In contrast, over the same period, the average

cash ratio of non-high-tech firms has remained relatively constant at around 11%; similar to

their level during the 1970s. This difference in the cash ratios of high-tech versus non-high-tech

firms is robust to alternative industry classification methodologies, such as the use of the Fama-

French industry classification or the Global Industry Classification Standard (GICS).

What causes this difference in the cash ratio over time? The literature on corporate cash

holdings shows that the level of a firm‟s cash holdings is a function of fundamental

characteristics related to the costs and benefits of holding cash (Kim, et al., 1998; Opler et al.,

1999). Although it is easy to justify high-tech firms having larger cash holdings than non-high-

tech firms, the growing difference in the cash ratio between these two sectors is puzzling. A clue

to deciphering this puzzle is the new listings effect. As equity markets developed, many firms

with weaker fundamentals went public in the 1980s and 1990s (Fama and French, 2004).

However, the nature and impact of new listings are different between the high-tech and non-

high-tech sectors. The high-tech sector has experienced a rapid expansion due to new listings in

the 1980s and 1990s. Moreover, these new listings were different from firms listed earlier. The

combined effect of these two aspects has led the population of public firms in the high-tech

sector to shift gradually toward the characteristics that, according to the literature on corporate

cash holdings, are typical of firms that hold more cash. On the other hand, the population

37

characteristics of the non-high-tech sector were not materially affected by new listings since

they were more similar to those existing firms.

To test whether the difference in changing firm characteristics can explain the difference in the

cash ratio over time, I follow the framework proposed by Fama and French (2001). After

estimating a regression model of corporate cash holdings using the first ten years of available

data, I then calculate out-of-sample forecasts using the observed firm characteristics and the

estimates from the regression model. The regression model is based on the Opler et al. (1999)

model, but augmented with additional variables to account for external debt and equity

financing and macroeconomic factors. These modifications are based on recent findings in the

literature and improve the explanatory power of the regression model. As emphasized by the

literature on R&D financing recently reviewed by Hall (2002), the nature of investment and

operation of firms in the high-tech sector is distinct from that in the non-high-tech sector, so it is

to be expected that the impact of various firm characteristics on corporate cash holdings may

differ between these two sectors. To address this potential difference, the modified cash model

is estimated separately for both the high-tech and non-high-tech sectors during the estimation

period. The out-of-sample forecasts on average justify the observed difference in the cash ratio

over time between the two sectors.

This paper is linked to several recent strands of research. First, this paper contributes to the

literature on new listings and equity market development (Fama and French, 2004; Brown and

Kapadia, 2007) as the first to provide a detailed comparison of the firm characteristics of new

listings in the high-tech and non-high-tech sectors. Existing research usually focuses on the

pervasive impact of new listings, and implicitly understates the cross-industry differences. In

contrast, I emphasize the significant difference between high-tech and non-high-tech new

listings and its importance for understanding changes in cash holdings. Furthermore, this paper

is indirectly linked to a recent literature that investigates the increasing conservatism of

corporate debt policy (Strebulaev and Yang, 2007; Byoun, et al., 2008) and a growing

preference for financial flexibility (Graham and Harvey, 2001; DeAngelo and DeAngelo, 2007).

This paper shows that new listings in the high-tech sector tend to hold a larger proportion of

their book assets in the form of cash while they seldom issue debt. This implies negative net

leverage if the cash is offset against any external debt. Given the increasing proportion of high-

38

tech firms in the overall sample of public firms, high-tech new listings may have contributed to

this observed conservatism in debt policy and preference for financial flexibility. Finally, this

paper provides a link to the literature on R&D financing. This literature argues that R&D-

intensive firms, particularly immature ones, are more likely to suffer from capital market

frictions; while they lack proper financial hedging instruments due to the nature of their

operations and investment (Hall, 2002; Carpenter and Petersen, 2002; Passov, 2003). This

implies that cash holdings of high-tech firms, especially for new listings, are different from non-

high-tech ones; an implication that is directly supported by this paper.

The remainder of the paper is organized as follows. Section 2.1 reviews current studies on

corporate cash policy, new listings, and R&D financing. Section 2.2 provides the evidence on

the difference in cash trends between the high-tech and non-high-tech sectors. Section 2.3

provides the explanation for this difference, and Section 2.4 concludes.

2.1 Literature Review

This paper is linked to existing research in three areas: corporate cash holdings, new listing

effects, and R&D financing. In what follows, I provide a brief discussion of the related studies.

If capital markets were perfect, i.e. external financing was frictionless, holding cash and cash

equivalents would be irrelevant since firms could always raise external financing at no cost

when internal funds were insufficient (Modigliani and Miller, 1958). Thus, maintaining zero

cash would be the optimal choice for any firm. The existence of frictions in capital markets

provides the rationale for firms to hold cash. As Keynes (1936) pointed out there are two

primary motives to justify firms‟ cash holdings. First, holding cash can help a firm avoid the

transaction costs associated with either liquidating a non-cash asset or using external financing

to make cash payments. The second is a precautionary motive: the desire to hold cash as a

cushion to hedge the risk of future cash shortfalls, which may be caused by either adverse

business shocks or new investment opportunities.

Building on Keynes‟ insights, recent empirical studies on corporate cash holdings by Kim et al.

(1998) and Opler et al. (1999) group the theories into benefits and costs of holding one more

dollar of cash. They find that some relevant firm characteristics, such as business risk, growth

39

opportunities, and size among others, can help explain the observed levels of cash held by firms.

More specifically, firms with more growth opportunities and higher business risk usually hold

larger cash reserves as a percentage of their total assets, whereas larger firms, firms with higher

net working capital, and highly levered firms tend to hold less cash.16

This empirical model has

been used in many recent empirical studies, either to explore additional determinants of

corporate cash holdings or to compute the „optimal‟ level of cash that a firm should hold given

its characteristics.17

Over the past three decades, the development of equity markets and the growth of mutual funds

have led to a reduction in the cost of equity capital, which allowed many firms with weaker

fundamentals to enter into market (Fama and French, 2004). Ritter and Welch (2002) show that

the proportion of IPOs with negative earnings in the year before listing has increased over the

period from 1980 to 2001. Fama and French (2004) find that the new listings in the 1980s and

1990s are less profitable, have more growth opportunities, and have lower survival rates. Brown

and Kapadia (2007) find that these new listings are riskier and shift the overall characteristics of

the population, thus being a fundamental cause of the „increasing idiosyncratic risk‟ puzzle.

Fama and French (2001) discover that the shift caused by the new listings partly contributes to

the steady decline in the percentage of dividend payers among public firms since 1978. Bates et

al. (2009) find that changing firm characteristics can explain the phenomenon of rising cash

holdings.

However, only a few papers in this literature have considered the differences in the evolutionary

paths of firms in different industries. Fama and French (2004) briefly compare the distribution

of profitability and growth opportunities (measured by changes in total assets) across five

industries over the years, but understate these cross-industry differences due to their focus on

the universe of public firms. Brown and Kapadia (2007) find that tech-intensive industries have

higher idiosyncratic risk due to the new listings effect, but do not investigate this result in detail.

However, the evolution of the cross-sectional differences in cash-relevant firm characteristics in

16

Dittmar (2008) provides a recent survey of this literaure. Opler et al. (1999) and Bates et al. (2009) provide

detailed descriptions on the theories and evidence that support the empirical model on cash holdings.

17 An incomplete list of papers includes Harford (1999), Dittmar et al.(2003), Pinkowitz et al. (2006), Foley et al.

(2007), Dittmar and Mahrt-Smith (2007), Harford et al. (2008), Kecskes (2008), Duchin (2008), Bates et al. (2009),

and Duchin et al. (2009).

40

the high-tech and non-high-tech sectors may play an important role in deciphering the different

trends in their cash holdings. Although many firms went public in these two sectors, the impact

of these listings on the two populations of firms may be different.

It is widely accepted that the central feature of high-tech firms is their intensive investment in

research and development (R&D). Hall (2002) summarizes the two distinct features of R&D:

first, the major portion of R&D spending is on human capital, which requires smooth investment

and generates intangible assets; second, the output of the R&D investment is uncertain,

particularly at the early stage.

These features imply that high-tech firms suffer more from capital market imperfections. On the

one hand, it is hard for them to get debt financing since their intangible assets can barely be used

as collateral. Bradley et al. (1984) identify R&D intensity being inversely related to leverage.

Carpenter and Petersen (2002) find that debt financing is rarely used by small high-tech firms.

On the other hand, information asymmetry is severe for R&D projects, because it is difficult for

outside investors to assess the value and likelihood of success for these projects. Further, this

information gap cannot be easily reduced by voluntary disclosure due to strategic concerns.

High information asymmetry leads to costly equity financing according to the pecking order

theory (Myers and Majluf, 1984) or a greater tendency to time the market (Baker et al., 2004).

In practice, Brown et al. (2009) find that internal funds and external equity are the two major

sources of finance for high-tech firms. Hence, high-tech firms may typically be more likely to

time the market to issue equity and then spend the proceeds on R&D gradually over time.

Brown and Petersen (2009) find that R&D-positive firms, particularly financially-constrained

ones, tend to use their cash reserves to smooth their R&D investment.

From a different but related perspective, high-tech firms are more inclined to use cash holdings

for hedging. Froot, et al. (1993) point out that the hedging instrument used by a firm depends on

the nature of its investment and financing opportunities. Richard Passov, the treasurer of Pfizer,

argues that in practice R&D is usually regarded as a liability for high-tech firms since the

inability to consistently fund R&D could trigger financial distress (Passov, 2003). R&D

liabilities, coupled with the low correlation of R&D investment with a company‟s internal cash

flow, as well as costly external financing, indicate a strong hedging motive for high-tech firms.

41

However, the unique risks associated with R&D cannot be hedged in financial markets, making

cash the preferable hedging tool chosen by high-tech firms (Passov, 2003). Acharya, et al. (2007)

formalize the idea of using cash as a hedging tool, and find that a larger cash reserve is more

preferable than lower debt when a firm‟s hedging need is higher.

The above literature review leads to the testing hypothesis to explain the difference in cash

trends. Compared to the non-high-tech sector, the population of the publicly traded firms in the

high-tech sector has tilted toward the characteristics typical of firms that hold more cash. The

source of the tilt is the new lists: they dominate high-tech sector by number and they differ

notably from the senior firms. With a proper regression model of corporate cash holding, and

adjusting for the potential differences in the impact of various firm characteristics on corporate

cash holdings across these two sectors, the differences in the changing firm characteristics

across these two sectors can adequately explain the observed difference in their cash trends.

2.2 Time Trends in Corporate Cash Holdings

2.2.1 Sample

The base sample of my study contains all U.S. publicly traded firms in the CRSP-Compustat

merged database (Fundamental Annual) over the period from 1974 to 2007. The sample starts

in 1974 because CRSP expanded to include NASDAQ firms in 1973 and the U.S. GAAP was

changed in 1974 to require firms to immediately expense their R&D expenditures (Statement of

Financial Accounting Standards, SFAS, No. 2, 1974). Firms that incorporate outside the United

States are excluded. Financial firms (SIC codes 6000-6999) are excluded since they need to hold

cash and marketable securities in order to meet statutory capital requirements. I also exclude

utility firms (SIC codes 4900-4999) as their cash policy can be a by-product of regulation.

For a firm to be included in the sample in a given year, it must have equity traded on the NYSE,

AMEX, or NASDAQ with a share code of 10 or 11 (ordinary common shares). Furthermore,

firms in a given year are excluded if their assets or sales were non-positive or if their cash and

marketable securities were negative. The screening leaves an unbalanced panel of 138,193

observations for 14,948 unique firms during the period from 1974 to 2007.

42

Table 2.1 reports the annual number of firms in the whole sample, as well as in the high-tech

and non-high-tech sectors respectively. Despite general agreement on the characteristics of high-

tech firms, such as high intensity of research and development activities and large proportion of

employment on scientific, technical, and engineering personnel, there is less consensus on

precisely which industries should be classified as „high-tech‟. I follow Brown et al. (2009) and

use the official definition of high-tech industries offered by the United States Department of

Commerce.18

More specifically, the high-tech sector consists of firms from the following seven

industries defined by 3-digit SIC codes: drugs (SIC 283), office and computing equipment (SIC

357), communications equipment (SIC 366), electronic components (SIC 367), scientific

instruments (SIC 382), medical instruments (SIC 384), and software (SIC 737). The remaining

firms in the sample are classified as non-high-tech. Table 2.1 shows that there was a rapid

increase in the number of firms in the high-tech sector during the 1980s and 1990s, followed by

a decline in the 2000s. The proportion of public firms belonging to the high-tech sector

increased from 12.82% in 1974 to 38.1% in 2000, and remained around this level henceforth,

despite the decrease in the number of firms.19

It is clear that the high-tech sector has become

increasingly important in the universe of publicly listed US firms over the sample period.

2.2.2 Cash Trends

Figure 2.1 tracks the time trends in the cash-to-assets ratio and net leverage ratio of the whole

sample, as well as for the high-tech and non-high-tech sectors respectively, measured by the

annual mean, median, and value-weighted average. For each firm-year observation, the cash-to-

assets ratio is measured by cash and marketable securities divided by total assets; the net

leverage ratio is defined as total debt minus cash and marketable securities, and then divided by

total assets.

The trends in cash holdings and net leverage for the whole sample are plotted as scatter points in

Figure 2.1. The annual mean cash ratio increased from 7.5% in 1974 to 9.8% in 1980, and then

18

“An Assessment of United States Competitiveness in High-Technology Industries,” United States Department of

Commerce, February 1983. 19

A detailed check on the Fama-French 12 industry groups, not reported in the paper, shows that some industries in

the non-high-tech sector, such as Durables, Manufacturing, and Non-Durables, have gradually shrunk over time.

43

to 23% in 2007, reaching a peak of 23.7% in 2004.20

The value-weighted average of the cash

ratio, as well as the median cash ratio, also exhibit upward trends, but less steep than the trend

of the annual mean. The annual mean (median) net leverage ratio decreased from 20.2% (21.4%)

in 1974 to 16.6% (18.5%) in 1980 and then to -2.8% (0.7%) in 2007, implying that the average

firm could almost repay its debts with its own cash holdings by the end of the sample period.

The trends over the 1980-2006 period are close to those reported in Bates et al (2009).

When the differences between the two sectors are examined, the results are more dramatic. The

cash holdings of these two sectors are very similar over the 1974 –1979 period. However, the

average cash ratio in the high-tech sector started to increase since 1980; between 1980 and 2007,

it has more than tripled, from 11.2% to 39.1% (all in annual mean). In contrast, the average cash

ratio of non-high-tech firms remained stable at a level similar to that of the 1970s, and increased

only slightly in the 2000s. For the net leverage ratio, the annual average of high-tech firms

became negative in the early 1990s and continued to fall, while non-high-tech firms on average

are always net debtors.21

2.2.3 Robustness Tests

To show that the difference in the cash trends for these two sectors from 1980 to 2007 is not due

to the specific definition offered by the U.S. Department of Commerce, I use two alternative

definitions: the Fama-French industry classification and the Global Industry Classification

Standard (GICS).22

More specifically, the whole sample is split into different industry groups

20

The VW-cash ratio is equivalent to the aggregate cash ratio in Bates et al. (2009), which is defined as the sum of

the cash and marketable securities divided by the sum of book assets for all sample firms. The annual mean is

equivalent to the average cash ratio in Bates et al. (2009).

21 SIC code used here is from the Compustat file. The results remain virtually identical if I instead use the SIC code

from the CRSP file (SICCD) or the historical SIC code from the Compustat file (SICH).

22 The Fama-French industries are defined on Ken French‟s website (http://mba.tuck.dartmouth.edu/pages/faculty/

ken.french/datalibrary.html). The Global Industry Classification Standard (GICS), developed by Standard & Poor‟s

(S&P) and Morgan Stanley Capital International (MSCI), categorizes a firm according to its operational

characteristics as well as investors‟ perceptions of its principal business activity. The GICS data can be retrieved

from the CRSP-Compustat merged database (Fundamental Annual). It contains 10 economic sectors (according to

the first two digits of the GICS code), which can be further sub-divided into a hierarchy of 23 industry groups, 59

industries, and 123 sub-industries. Recent studies by Bhojraj et al. (2003) and Chan et al. (2007) compared the

GICS with the Fama-French classification in capital market research.

44

according to these two criteria respectively.23

Besides providing alternative categorization

schemes, these two classifications also provide a more detailed look on the component

industries in the high-tech and non-high-tech sectors.

After calculating the mean and median cash ratio, as well as the value-weighted average cash

ratio, for each industry group every year from 1980 to 2007, I investigate the significance of the

time trend in the average cash ratios for each industry by applying a linear trend model, i.e.

regressing the average cash ratio on a constant and a time index measured in years. Results for

the industry groups based on the Fama-French and the GICS schemes are reported in Table 2.2.

Regardless of the measure used to capture the industry average, the upward trend in cash

holdings is economically and statistically more significant for two industry groups: Business

Equipment (BusEq, including firms in computers, software, and electronic equipment) and

Healthcare (Hlth, including firms in drugs, medical equipment, and healthcare) by the Fama-

French standard; or Information Technology (including software, hardware, and electronic

equipment) and Health Care (including health care equipment and services, pharmaceuticals,

biotech and life sciences) by the GICS scheme.24

Although the Telecom group (Telcm,

including firms in telephone and television transmission) also exhibits a statistically significant

upward trend, its economic size is much smaller. The trends in the cash holdings of the other

industry groups are either statistically insignificant or even downward (e.g. Energy). Since

Business Equipment (or Information Technology by the GICS) and Healthcare industries are

generally regarded as high-tech industries, results in Table 2.2 verify the aforementioned

differences in the cash trends between the high-tech and non-high-tech sectors.

In the above analysis, I follow the standard approach in cash literature to use the book value of

total assets as the denominator of cash ratio. A potential concern for this approach is that the

upward trend in the high-tech sector might be artificial due to conventional accounting rule.

23

Since financials and utilities according to the SIC codes are excluded, only 10 of the Fama and French 12

industry groups are used in this study. According to the GICS, my sample contains small numbers of observations

from the financials (40) and the utilities (55). This is due to minor difference between the SIC and GICS schemes.

For example, Potlatch Corporation (NPERMNO= 49744) belongs to the financial sector by the GICS since it is

traded as a real estate investment trust. However, due to its business focus in pulp, paperboard, and wood products,

Potlatch is allocated in the paperboard mills industry by the SIC, so it belongs to manufacturing industry according

to the Fama-French 12-industry classification.

24 The R

2 of the linear trend model for these two industry groups are also much larger than others.

45

Current accounting rule (Statement of Financial Accounting Standards, SFAS, No. 2, 1974)

requires firms to immediately expense their R&D spending, so it is not reflected in the total

assets of a firm. Since R&D is an important part of investment for high-tech firms, especially

the young ones, this accounting rule can lead to a severe underestimate of assets and a

consequent overestimate of cash-to-assets ratios for high-tech firms.

To test whether this accounting issue drives my result, I follow Chan et al. (2001) to capitalize

R&D Expenses and add it to total assets. The R&D asset (RDA) for firm i in year t is calculated

as the weighted sum of its R&D expense over the past five years assuming an annual

amortization rate of 20%: 1 2 3 40.8 0.6 0.4 0.2it it it it it itRDC RD RD RD RD RD . The

R&D-adjusted book value is calculated by adding this R&D asset to the reported book value.

Figure 2.2 shows the trends in the cash ratio that uses the R&D-adjusted book assets as the

denominator.25

It is clear that incorporating R&D capital reduces the slope of cash trend in the

high-tech sector, from a 1.03% annual increase in Figure 2.1 to a 0.64% annual increase in

Figure 2.2 (in terms of annual mean). Nevertheless, the upward trend in cash holdings still holds

for the high-tech sector after this adjustment, indicating that it is not entirely driven by the

existing accounting standard. Moreover, Figure 2.2 also shows that incorporating R&D capital

on average has no significant impact on the cash trend in the non-high-tech sector.

In sum, the above analysis shows that the increase in corporate cash holdings is a phenomenon

specific to the high-tech sector. This finding is robust to alternative sector classifications and the

capitalization of R&D expense.26

25

The existing literature has no consensus on estimates for the duration and rate of amortizing R&D expense. Lev

and Sougiannis (1996) argue that they are mainly related to „the ability of innovators to appropriate the benefits of

innovations‟ and hence vary across industries by the effectiveness of patent protection. Damodaran (2006, p.82)

argues that the amortizable life may vary across firms to „reflect the difficulties associated with commercializing

research‟. I follow Chan et al. (2001) to use 5-year period uniform amortization for all firms and report the results

in Figure 2. Alternative specifications, such as 1) using 10 years for all firms, or 2) using 10 years for

pharmaceutical firms to address their long commercialization and patent process and using 5 years for firms in

information technology industries and non-high-tech sector, provide similar results.

26 Bates et al. (2009, p. 1997-1998) also examines the cash trends of high-tech firms vs. non-high-tech firms, and

find upward trends for both sectors. In their analysis, high-tech industries are defined according to Loughran and

Ritter (2004), which exclude pharmaceuticals (SIC 283); Non-high-tech industries are defined as “„old-economy‟

manufacturing firms as firms with SIC codes 2000-3999 that are not high technology firms”, which include

pharmaceuticals. These pharmaceuticals exhibit high cash ratios and occupy a large proportion of „old-economy

manufacturing firms‟, which jointly contribute to the upward cash trend of non-high-tech firms in their paper.

46

2.3 Explaining the Different Trends in Corporate Cash Holdings

Why are the trends in cash holdings so different between the high-tech and non-high-tech

sectors over these three decades? Considering the setting of such a long time period, the

explanation proposed in my paper is based on exploring some fundamental changes that have

transformed the population characteristics differently in these two sectors. One vital change is

associated with those firms that went public during this period.

Figure 2.3 shows annual numbers of IPO firms for 1974-2007.27

Overall, after a moderate

increase in the 1970s, stock markets experienced surges of new listings in the 1980s and 1990s,

before cooling down in the 2000s. Decomposing these IPO firms into the high-tech and non-

high-tech sectors, it is clear that both sectors experienced the same hot and cold markets.

Although there are fewer IPOs in the high-tech sector each year, except the Internet bubble

period of 1999 and 2000, they greatly contribute to the increasing weight of the high-tech sector

in the universe of public firms (Table 2.1).

More importantly, these new listings are closely linked to some distinct patterns of cash trends

in two sectors. As a preliminary check, Figure 2.4 depicts the cash holdings of newly IPO firms

and seasoned firms in the two sectors, with newly IPO firms being defined as those within five

years after their IPO dates and seasoned firms otherwise.28

The peaks in the annual mean and

median cash holdings only exist for the newly IPO firms, indicating that these firms tend to save

the proceeds from their IPOs (and/or the follow-up SEOs and debt financing). Furthermore, the

upward trend in cash holdings only exists for the high-tech sector, for both seasoned and newly

IPO subgroups. Hence, it is not the higher proportion of the newly IPO firms in the high-tech

sector that drives the results. Moreover, it also shows that when the newly IPO firms in the high-

tech sector become seasoned, on average they still hold more cash than those firms that went

public earlier. Overall, new listings seem to have transformed the high-tech and non-high-tech

sectors differently and potentially help explain their difference in cash trends.

27

The pattern on the whole sample (indicating by „All IPO‟) over 1974-2001 is close to Fama and French (2004,

Figure 2). The annual numbers of IPOs in my sample are always slightly smaller than those in Fama and French

(2004) since my sample is narrower by excluding financials and utilities.

28 Fama and French (2004) use this differentiation as well. Welch (1989, 1996) finds that firms usually make

sequential seasoned equity offerings (SEO) or issue debt right after their IPO.

47

In this section, I first examine the determinants of the cash ratio in a pooled sample, that is,

across all sectors, as well as in the high-tech and non-high-tech sectors separately, over the

entire sample period. I propose a modified version of the cash holding model by including

several variables identified recently in related literature. I also examine the consistency of my

results with that of previous studies and test whether the factors that affect the cash ratio differ

across the two sectors. Subsequently, I examine the evolution of the firm characteristics in the

two sectors to establish whether the population of the firms in the high-tech sector, compared to

the non-high-tech sector, drifted more toward those firms tending to hold more cash. The

contribution of the later listing cohorts to the evolution of the firm characteristics across the two

sectors is investigated separately. Finally, I directly test, using the basic framework proposed in

Fama and French (2001), whether the difference in the changing firm characteristics can provide

an adequate explanation for the observed difference in the cash trends.

2.3.1 The Determinants of Cash Holdings

Previous studies use the cash holdings model proposed by Opler et al. (1999). This model

attempts to explain firms‟ cash holdings using a number of important determinants, including

firm size, profitability, net working capital, growth opportunities (as measured by market to

book ratio, capital expenditure, and R&D expense), risk (as measured by industry cash flow

volatility), and leverage, as well as a dividend dummy. Based on some recent studies, as well as

the need of my analysis, I slightly extend the regression model of corporate cash holdings by

adding four more variables: net debt issuance, net equity issuance, and two macroeconomic

variables (T-bill yield and default spread) 29

.

The cash flow identity, OCF NetDiss NetEiss Investments Dividend NWC Cash ,

shows that net debt issuance (NetDiss) and net equity issuance (NetEiss) are two channels of

external financing and potentially contribute to the changes in cash holdings. Several recent

studies find that firms tend to save the proceeds from their external financing as cash, especially

when firms take advantage of potential mispricing or have strong precautionary motives

(Hertzel and Li, 2009; McLean, 2010). Moreover, decomposition in Figure 2.4 shows that the

29

These four variables are also used by Bates et al. (2009).

48

peaks in annual average cash ratio are associated with firms that went public within the past five

years, when firms prefer to make sequential seasoned equity offerings (SEO)

or issue debt

(Welch, 1989, 1996). Hence, it is necessary to add net equity issuance and net debt issuance into

the regression model to account for the impact of external financing on corporate cash holdings.

Cash holding models are usually estimated with year dummies to control for the potential

impact of common trends or business cycle effects since cash is held as a reserve against these

cyclical events. Because I will use changing firm characteristics to explain the difference in the

cash trends across the high-tech and non-high-tech sectors by using out-of-sample forecast,

accounting for common trends or business cycles with year dummies is not feasible. Instead, I

use the T-bill yield and the default spread. The risk-free rate represents the opportunity cost of

cash; a decrease in the risk-free rate will decrease the opportunity cost of cash, making it

cheaper to hold larger quantities of cash. The T-bill yield can be considered as a proxy for the

risk-free rate, and is expected to have a negative impact on corporate cash holdings. The default

spread, defined as the average yield on Baa less Aaa Moody‟s rated corporate bonds with a

maturity of approximately 20-25 years, usually increases when macroeconomic conditions

deteriorate and default risk intensifies. In such conditions, firms usually increase their cash

reserves due to precautionary motives. Hence, it is expected that corporate cash holdings and the

default spread are positively correlated.

The modified version of the regression model of corporate cash holdings is as follows. The

details on how the variables are constructed are provided in the Appendix 2.A.

0 1 2 3 4 5 6

7 8 9 10

11 12 13 14

( )

&

it it it itit it it

it it it it

it itit it

it it

it it t

Cash NWC CF CAPEXSize IndustrySigma MB

TA TA TA TA

R D ACQNLeverage DivDummy

Sales TA

NetDiss NetEiss TbillYield

t itDefaultSpread

Table 2.3 documents the descriptive statistics of firm characteristics for all firms, and for high-

tech and non-high-tech firms respectively. Z-statistics from the Wilcoxon rank-sum test reject

the null that high-tech and non-high-tech firms are from populations with the same distribution.

Tests for equal means are reported by t-statistics. Overall, firms from two sectors are different in

49

all aspects. Compared to non-high-tech firms, high-tech firms on average are smaller and less

profitable; have higher cash ratio, less net working capital, more business risk, higher market-to-

book ratios, lower leverage and less likely to be dividend payer; invest more on R&D but less on

capital expenditure; issue less debt but more equity.

Table 2.4 documents the correlation matrices of firm characteristics for all firms, and for high-

tech and non-high-tech firms respectively. The absolute values of most pair-wise correlations

are below 0.5. To investigate whether multicollinearity among these variables is a potential

concern, variance inflation factors (VIF) are calculated for each variable in the pooled sample

and two sectors separately. Results (unreported here) show that the VIFs of all the variables, in

the pooled sample as well as in high-tech and non-high-tech subsamples, are far below the

critical value of 10. Hence, multicollinearity should not be a significant concern.

Panel A of Table 2.5 presents the results of regressions estimated over the entire period of 1974

- 2007. There are 138,193 firm-year observations in the dataset. Requiring complete data for the

explanatory variables reduces the sample to 117,240 firm-year observations. The standard errors

are adjusted for firm-level clustering, assuming errors are independent across firms but not over

time (Petersen, 2009).

The first column in Panel A reports the coefficient estimates for the basic regression model

developed by Opler et al. (1999). The results are consistent with previous studies: small firms

with high growth opportunities and low net working capital in industries with high cash flow

volatility tend to hold more cash; while stable firms, such as those with dividend payments or

higher leverage, usually hold less cash. Moreover, higher levels of capital expenditures and

acquisitions are associated with lower cash holdings. The OLS estimates for the parameters of

the modified model, which incorporate net equity issuance, net debt issuance, and two

macroeconomic variables, are reported in column (2). The coefficients on net equity issuance

and net debt issuance are both statistically significant, consistent with the argument that firms

leave some of the proceeds from their external financing in the form of cash. The coefficients on

the T-bill yield and the default spread have the correct signs, that is, they are negative and

positive respectively. The estimated coefficients of the remaining variables are generally

consistent with the findings of the basic model in column (1). A major difference is that the

50

coefficient on the acquisitions variable doubles, indicating that it is closely related to two new

external financing variables. The modified model provides a better explanation of the variation

in cash holding as the adjusted R-squared increases by almost 4% to reach 55.8% compared to

51.9% for the basic model. Incorporating industry fixed effect does not change the results.

To examine whether the T-bill yield and the default spread properly capture the common trend

related to macroeconomic events, I replace these two variables with year dummies. The

coefficient estimates reported in column (4) are almost the same as those in column (2), and the

adjusted R-squared increases only by 0.4%. Therefore, using the T-bill yield and the default

spread gives results that are similar to those year dummy variables. Results are similar if the

model is estimated with year and industry fixed effects (Column 5). Finally, I use the Fama and

MacBeth (1973) method, estimating the coefficients in the cross-sectional regression each year

and reporting the time series averages of the annual estimates. The results obtained (Column 6)

are consistent with previous findings.

Since the business environment is different for firms in the high-tech versus non-high-tech

sectors, it is highly possible that the impacts of firm characteristics on corporate cash holdings

are different between these two sectors. I estimate the modified model of corporate cash

holdings for the high-tech and non-high-tech sectors separately over the period from 1974 to

2007, the results of which are reported in columns (7) and (8). The adjusted R-squared is 58.6%

for the high-tech sector, almost 40% higher than that of the non-high-tech sector (42.3%),

indicating that the cash model provides a much better fit for high-tech firms. Column (9) tests

whether the estimates of separate regressions are statistically different across two sectors.

Although all the coefficients, except dividend dummy, have the same signs across these two

sectors, there are statistically significant and economically large differences in all the other

explanatory variables between high-tech and non-high-tech firms, except for the market-to-book

ratio and the R&D-to-sales ratio. Furthermore, the F-test rejects the hypotheses that all the

coefficients are jointly equal across these two sectors. Overall, there exists preliminary evidence

in support of estimating the cash model for the two sectors separately.

Although the signs and statistical significance of the coefficients from the above regressions are

consistent with expectations, the results do not provide clear guidance on which determinants

51

are relatively more important in explaining corporate cash holdings.30

To examine the relative

importance of these explanatory variables, I apply a method proposed by Grömping (2007),

which provides estimates of the proportion of the variation of the dependant variable explained

by the variation of each of the explanatory variables by taking into account the pair-wise

correlation among the independent variables. Two values are reported for each determinant: the

percentage of the variation in the dependant variable that it explains (absolute value) and the

percentage of the variation explained within the regression model that it contributes

(standardized value). A higher absolute or standardized value indicates a more important

variable.

The results based on the OLS regressions of the modified cash model for the whole sample and

for the high-tech and non-high-tech sectors respectively are reported in Panel B of Table 2.5.31

According to the standardized values, the top five contributors to the overall variation in the

predicted cash holdings, for the pooled sample as well as for two separate sectors, are leverage,

R&D intensity, net working capital, size, and net equity issuance. In total, they contribute to

more than 80% of the total explanatory power of the model, irrespective of whether it is for the

whole sample or for the two separate sectors. However, the explanatory power of R&D and net

working capital is much stronger for high-tech than for non-high-tech firms. R&D and net

working capital jointly contribute to 44.3% of the total explanatory power of the model for the

high-tech sector, but only 19.3% for the non-high-tech sector. Overall the modified cash model

explains 58.7% of the variation in the cash holdings of high-tech firms, but just 42.3% for non-

high-tech firms. Hence, the absolute explanatory power of R&D and net working capital is even

higher in the high-tech sector. Moreover, the contribution from the industry-level cash flow

volatility (IndustrySigma) is relatively small, just around 4% of the total variation for the whole

sample and even smaller for the two separate sectors. This does not necessarily shake its role as

a key proxy for the precautionary motive to hold cash, as this industry-level measure has less

cross-sectional variation than those firm-level variables.

30

Several recent studies in corporate finance discussed the relative importance of explanatory variables, such as

Bekaert et al. (2008) on equity market segmentation, Frank and Goyal (2009) on capital structure decisions, and

Lemmon et al. (2008) on persistence in capital structures.

31 The relative importance results are obtained using the R package relaimpo, discussed in Grömping (2006).

52

2.3.2 Difference in Changing Firm Characteristics across Two Sectors

Previous studies establish that the characteristics of firms that went public during the 1980s and

1990s are different from those that went public earlier, in terms of lower profitability and more

growth opportunities (Fama and French, 2004). The shift in the characteristics of the population

of public firms, caused by these new listings, has been used to explain the puzzling phenomena

of disappearing dividends, increasing idiosyncratic risk, and increasing cash holdings (Fama and

French, 2001; Brown and Kapadia, 2007; Bates et al., 2009).

However, less is known about the differences in the attributes of new listings in the high-tech

versus non-high-tech sectors. According to Figure 2.4, there are many new listings in both

sectors, especially in the 1980s and 1990s. The analysis in the previous section identifies a set of

firm characteristics that are correlated with the cross-sectional variation in corporate cash

holdings. Furthermore, it is evident that five core variables contribute to more than 80% of the

total explanatory power of the model, for the whole sample as well as for the two sectors

separately. In this section, I investigate a) whether these cash-related characteristics of the listed

firms in the two sectors have changed over time, b) how the changes of these characteristics are

different between high-tech firms and non-high-tech firms, and c) to what extent the new

listings have contributed to these changes.

Table 2.6 reports the summary statistics of the cash-related characteristics of the listed firms in

the high-tech and non-high-tech sectors in sub-periods defined according to calendar decades,

with the 1980s and 1990s being the two sub-periods with huge increase in new listings.

The equally-weighted averages of each characteristic for the high-tech and non-high-tech

sectors during a given sub-period are reported respectively in Panel A. This presents a general

picture of the evolution of the characteristics of a typical firm in each sector over time. Since the

difference in the cash trends started at the beginning of the 1980s, I use the 1970s characteristics

as the benchmark for comparison purposes, and thus focus on changes in subsequent decades.

During the 1980-2007 period, both sectors experienced an increase in firm size, business risk

(measured by industry cash flow volatility), and R&D intensity, and a decrease in net working

capital, capital expenditure, leverage, and the proportion of dividend payers. Based on the

53

findings described in the previous section that the cash ratio increases with business risk and

R&D intensity, and decreases with size, net working capital, capital expenditure, leverage, and

dividend dummy, changes in these firm characteristics, except firm size, are all in the direction

of increasing cash holdings.

More importantly, the extent of changes is quite different across the two sectors. The high-tech

sector has gone through more significant changes relative to the non-high-tech sector. To

facilitate the comparison, I follow Brown and Kapadia (2007) to directly test the time trends in

the aforementioned firm characteristics of the two sectors over 1980-2007, by regressing each

firm characteristic on a constant and a time index measured in years. Slope coefficients and their

p-values are reported in Panel A of Table 2.6. Subsequently, I compare the estimates of the

slope coefficient between the high-tech and non-high-tech sectors, with the focus on the five

core characteristics.

The Leverage of high-tech firms declined much faster, at a rate of 0.4% per year (compared to

0.1% per year for non-high-tech firms). The average leverage of high-tech firms in the 2000s is

12.4%, only half of the level during the 1970s benchmark period. Compared with the 1970s,

R&D intensity has increased considerably in both sectors, but with a remarkable difference in

their rates. R&D/Sales of non-high-tech firms followed an upward trend with a slope equal to

0.1%, but the time trend is much steeper in the high-tech sector, increasing from 23.8% in 1980s

to 66.8% in 2000s with an annual rate of 2.4%. The ratio of net working capital to assets has

decreased in both sectors. Starting with a higher level of 29.1% (compared to 20.4% in non-

high-tech), NWC/TA of the high-tech firms, on average, ends up at a much lower level of 1.5%

in the 2000s (compared to 7.6% for the non-high-tech sector) since the rate of decrease in the

high-tech sector is three times faster than that in the non-high-tech sector. For both sectors, net

equity issuance has become more common after the 1970s and experienced an inverted-U shape

change with its peak in the 1990s. However, high-tech firms generally use much more equity

financing than firms in the non-high-tech sector and are more dependent on equity financing

than on debt financing. This is consistent with the findings of previous studies on the R&D

financing literature (Hall, 2002). Finally, compared to 1970s, the average size of firms in both

sectors has dropped in the 1980s and 1990s due to new listings. Over the entire period from

54

1980 to 2007, the average size has increased gradually, but high-tech firms are usually much

smaller than non-high-tech firms.

Besides these five core factors, Panel A also shows that the changes in several other factors are

also stronger in the high-tech sector. Business risk, measured by Industry Sigma, has increased

must faster in the high-tech sector, a result consistent with the findings on idiosyncratic

volatility in Brown and Kapadia (2007). The ratios of cash flow to assets are similar in both

sectors in the 1970s but drop afterwards, consistent with previous findings that new listings in

the 1980s and 1990s are less profitable (Fama and French, 2004). However, over 1980-2007, the

ratio of cash flow to assets gradually rises in the non-high-tech sector, but continues to decline

and becomes negative for the high-tech sector. The latter is consistent with the finding in Ritter

and Welch (2002) that over the past three decades the portion of technology IPOs and the

percentage of IPOs with negative earnings are positively correlated. Comparing the trends in

CF/TA, the source of internal funds, with the trends in the two channels of external financing, it

seems that on average high-tech firms have become more and more dependent on external

financing, especially equity, to fund their investments.

Panel A also shows that growth opportunities, as measured by the market-to-book ratio, are

higher in the high-tech sector. Furthermore, a difference deserving more attention is that the

growth opportunities in the high-tech sector are based on R&D expenditures, while the non-

high-tech sector‟s growth is built mostly on capital expenditure.32

This is important since capital

expenditures help build a firm‟s tangible assets, whereas R&D expenditures, treated as an

expense, generate intangible assets that are not reflected on the balance sheet. This partly

explains the difference in the market-to-book ratio across the two sectors. However, what is

more important is that growth opportunities based on R&D are more uncertain compared to

those based on capital expenditures, and hence they trigger a stronger precautionary motive for

holding cash.

Panel A provides evidence that the high-tech sector has shifted more significantly towards a set

of firms with characteristics consistent with higher and increasing cash holdings. In Panel B, I

32

The R&D-to-assets ratio is reported here to facilitate the direct comparison with the capital expenditure-to-assets

ratio.

55

decompose the firms in each sub-period according to the year they went IPO and investigate

whether the changes are driven by new listings or not.33

When compared to the non-high-tech sector, the high-tech sector has been progressively

dominated by new listings. During the 1980s, the 1980 IPO cohort constitutes 54% of the

observations in the high-tech sector, compared to 34% in the non-high-tech sector. During the

1990s, the 1980 and 1990 cohorts jointly contribute to 84% of the sample observations in the

high-tech sector, while the ratio is 70% for the non-high-tech sector. Even after the burst of the

Internet bubble, 79% of the observations in the 2000s are from the 1980s and 1990s cohorts in

the high-tech sector, whereas the ratio is 63% in the non-high-tech sector.

Moreover, firms in the later cohorts are different from those in the earlier cohorts, and these

differences are more significant in the high-tech as compared to the non-high-tech sector. In the

high-tech sector, although the firms that survived in each cohort have on average decreased their

leverage over time, a more important phenomenon is that the leverage of firms from the later

cohorts is always lower than that in an earlier cohort. A similar pattern is not obvious in the non-

high-tech sector. In both sectors, later cohorts have higher R&D intensity, but the increase takes

place much faster in the high-tech sector. The R&D-to-sales ratio of the Pre-1980 IPO cohorts is

around 10% in all the sub-periods, while the R&D-to-sales ratio of the 1980s cohort is between

35% and 38% over time, and the ratio of the 1990s cohort is even higher at 74%. Hence, the

higher R&D intensity of new listings has contributed to the increase in the average R&D

intensity in the high-tech sector. A similar analysis can be applied to other characteristics, for

example net working capital and net equity issuance.

In sum, the results tabulated in Table 2.6 show that when compared to the non-high-tech sector,

the population of publicly traded firms in the high-tech sector has tilted more towards

characteristics typical of firms that hold more cash. The source of this tilt is new listings: they

dominate the high-tech sector by number and they differ from more senior firms. What remains

to be answered is whether these differences in firm characteristics can explain the observed

difference in cash trends across the two sectors since the early 1980s.

33

Industry Sigma is excluded from this table since it is an industry measure and hence impossible to decompose the

contribution of each IPO cohort.

56

2.3.3 Can Difference in Changing Characteristics Explain Different Cash

Trends?

To test whether the difference in changing firm characteristics can explain the difference in the

observed cash trends across high-tech and non-high-tech sectors, I follow the basic empirical

strategy proposed in Fama and French (2001). First, I estimate the modified regression model of

corporate cash holdings over the period from 1974 to 1983 (estimation period). This provides a

model of how cash holdings depend on firm characteristics identified earlier. I then use this

estimated model of corporate cash holdings to predict the cash that firms should hold over the

1984-2007 period. By fixing the coefficients based on the 1974-1983 period but allowing for

changes in firm characteristics, the predictions capture the change in cash holdings explained by

changing firm characteristics.

Although the difference in the cash trends starts around 1980, I expand the estimation period to

1983 to increase the number of available observations for estimating the model coefficients.

Moreover, since this estimation period covers the recessions in the early 1980s and the

subsequent hot IPO markets, it contains adequate variations in the external financing variables

and the macroeconomic variables to ensure a proper estimate of their impacts on cash policy.34

Table 2.7 reports coefficient estimates of the modified model of cash holdings during the

estimation and forecast periods, for the pooled sample (all firms) and for the high-tech and non-

high-tech sectors respectively. Setting aside the dividend dummy, estimated slopes in both

estimation and forecast periods confirm the inferences drawn from regressions with the entire

sample period. In the pooled regression and separate regressions for high-tech and non-high-

tech firms, F-tests reject the hypothesis that all coefficients are jointly equal between estimation

and forecast periods. However, when the t-statistics of the estimates are considered, a smaller

number of variables is found to differ in the cases of the regressions conducted separately for

high-tech and non-high-tech firms, thus pointing to a lower extent of discrepancy.

34

If the estimation of a model takes place using only a segment of the available historical data, it is essential that

the estimation interval is representative of the entire period. An important lesson learnt from the recent credit

crunch is that including historical data on market crashes during the estimation period can improve forecasts for

future crises.

57

Given these parameter estimates from the estimation period, I can assess the cash-to-assets ratio

that a firm in the high-tech (non-high-tech) sector would have maintained had the firm chosen

its cash holdings in the same way as during the estimation period. More specifically, the

expected cash ratios are computed by applying the coefficient estimates from the estimation

period to the values of the explanatory variables for each firm-year observation during the

forecast period of 1984-2007. The finding that the actual cash ratio stays above the expected

value, in the spirit of Fama and French (2001), indicates that a firm is more inclined to hold cash

than in the estimation period.

Table 2.8 reports the annual mean of actual and expected cash holdings, in the high-tech sector

(Panel A) and in the non-high-tech sector (Panel B), with the coefficients being estimated

respectively from the pooled regression and the separate regressions during the estimation

period. Although the predicted cash holdings in the high-tech sector on average track the

upward trend, the results based on the coefficients from the separate regression are much better,

as evidenced from the smaller deviation from the annual means of actual values. More

specifically, if I ignore the difference between two sectors by applying coefficients estimates

from the pooled regression in the estimation period, the inference on the out-of-sample forecast

would be that high-tech firms on average hold more cash than needed for most years and the

excess cash increases faster over time, staying above 10% in the 2000s. Since the predicted level

of cash already takes into account the various motives of holding cash, persistently holding

excess cash at such a high level would be considered as a puzzle. Nevertheless, the results based

on the coefficients from the separate regression for the high-tech sector tells a different story:

high-tech firms on average seem to hold more cash than needed over the past decade, but the

excess cash is always below 5% of total assets. Compared to the remarkable differences

between the results from different estimation methods in the high-tech sector, the out-of-sample

predictions in the non-high-tech sector, as documented in Panel B, are only slightly affected,

and the expected cash ratios are closer to the actual values.

Overall, the analysis in this section shows that the difference in changing cash-related firm

characteristics can help explain the difference in cash trends across the high-tech and non-high-

tech sectors. However, the distinct results in the high-tech sector when using coefficient

estimates from the separate versus the pooled regressions imply that future studies should

58

properly account for the potential difference across these two sectors before concluding that

high-tech firms on average hold too much „excess‟ cash.

2.4 Conclusion

Corporate cash policy has triggered a lot of interest recently, but most of the attention has been

focused on the difference in cash holdings of large firms from the high-tech and non-high-tech

sectors. This paper studies the difference in the cash holdings of „typical‟ firms from these two

sectors over a longer period and attempts to provide an explanation for the difference.

Since 1980, the average cash holdings of publicly listed U.S. firms in the high-tech and non-

high-tech sectors have become more and more different. In contrast to the average cash-to-assets

ratio of non-high-tech firms that remained stable around 11%, which is similar to its level during

the 1970s, the average cash ratio in the high-tech sector has increased significantly, rising

gradually from 11.2% in 1980 to 39.1% in 2007. This upward trend in the cash holdings of the

high-tech sector, along with its growing weight in the universe of publicly listed U.S. firms,

contributed to the phenomenon of increasing cash holdings for the entire sample of firms, as

documented in Bates et al. (2009).

The identified difference in the cash trends between the high-tech and non-high-tech sectors

coincides with a growing difference in the cash-relevant firm characteristics across the two

sectors. When compared to the non-high-tech sector, the population of public firms in the high-

tech sector has tilted toward attaining the characteristics typical of firms that hold more cash.

This tilt is caused by the new listings: the high-tech sector has expanded considerably by the

new listings in the 1980s and 1990s, whose nature differ notably from existing firms. On the

other hand, the new listings in the non-high-tech sector are similar to existing ones and thus less

change is seen in the characteristics of the population.

In order to test whether the differences in changing firm characteristics provide an adequate

explanation for the observed difference in the cash trends across these two sectors, I expand the

basic regression model of corporate cash holdings developed in Opler et al. (1999) by adding

several variables to control for the impact of external financing, and macroeconomic events, and

I employ the basic framework proposed in Fama and French (2001). The out-of-sample

59

forecasts, based on coefficients estimated separately for the two sectors in the early sub-period

and changing firm characteristics in the forecast period, on average, adequately justify the

observed difference in cash trends across the high-tech and non-high-tech sectors over time.

Furthermore, for the high-tech sector, the average excess cash holdings is reduced to only

around 5% of the total assets when the coefficients from the separate estimation per sector are

used compared to an average of above 10% of the total assets, when the coefficients from a

pooled estimation are used. This result should generate some skepticism towards a “one-size-

for-all” cash holding model, as it may generate false claims that the high-tech sector holds too

much cash.

Over the past three decades, the high-tech sector has become more important among the

publicly listed US firms due to the disproportional influx of new listings into this sector.

Moreover, high-tech firms are different from traditional firms in terms of operation, investment,

and financial policies. The results in this paper document this difference and highlight the

importance of developing theoretical and empirical models in future studies to examine the

high-tech sector in particular, instead of relying on a generic model for all industries.

60

Table 2.1: The Distribution of Firms

This table reports the number of firms in the whole sample and in the high-tech and non-high-tech sectors

separately each year during the period from 1974 to 2007. The sample includes U.S. firms documented on the

Compustat-CRSP merged database (fundamental annual) that have positive total assets and sales and nonnegative

cash and marketable securities, and have common shares traded on the NYSE, AMEX, or Nasdaq. Financial firms

(SIC code 6000-6999) and utility firms (SIC codes 4900-4999) are excluded from the sample, leaving an

unbalanced panel of 138,193 observations for 14,948 unique firms. The high-tech and non-high-tech sectors are

defined according to the U.S. Department of Commerce. The proportions (in percentage) of the high-tech and non-

high-tech sectors in annual sample are reported respectively.

Whole Sample High-Tech Non-High-Tech

Year Number Number Percent Number Percent

1974 3400 434 12.8% 2966 87.2%

1975 3337 434 13.0% 2903 87.0%

1976 3328 442 13.3% 2886 86.7%

1977 3255 438 13.5% 2817 86.5%

1978 3202 447 14.0% 2755 86.0%

1979 3286 501 15.2% 2785 84.8%

1980 3318 526 15.9% 2792 84.1%

1981 3558 613 17.2% 2945 82.8%

1982 3668 677 18.5% 2991 81.5%

1983 3883 816 21.0% 3067 79.0%

1984 4127 960 23.3% 3167 76.7%

1985 4055 972 24.0% 3083 76.0%

1986 4116 1038 25.2% 3078 74.8%

1987 4301 1128 26.2% 3173 73.8%

1988 4211 1111 26.4% 3100 73.6%

1989 4043 1087 26.9% 2956 73.1%

1990 3998 1075 26.9% 2923 73.1%

1991 4025 1104 27.4% 2921 72.6%

1992 4216 1193 28.3% 3023 71.7%

1993 4586 1309 28.5% 3277 71.5%

1994 4888 1356 27.7% 3532 72.3%

1995 5097 1495 29.3% 3602 70.7%

1996 5535 1742 31.5% 3793 68.5%

1997 5614 1857 33.1% 3757 66.9%

1998 5304 1789 33.7% 3515 66.3%

1999 5036 1785 35.4% 3251 64.6%

2000 4874 1855 38.1% 3019 61.9%

2001 4328 1661 38.4% 2667 61.6%

2002 3954 1508 38.1% 2446 61.9%

2003 3655 1342 36.7% 2313 63.3%

2004 3628 1356 37.4% 2272 62.6%

2005 3548 1322 37.3% 2226 62.7%

2006 3455 1266 36.6% 2189 63.4%

2007 3364 1230 36.6% 2134 63.4%

138193 37869 100324

61

Table 2.2: Trends in Cash Holdings: Fama-French Industries and GICS

Economic Sectors

This table reports the trends in cash holdings of industry groups defined by the Fama-French 12 industry

classification (Panel A) and the Global Industry Classification Standard (GICS) (Panel B) over the period from

1980 to 2007. The sample includes U.S. firms documented on the Compustat-CRSP merged database (fundamental

annual) that have positive total assets and sales and nonnegative cash and marketable securities, and have common

shares traded on the NYSE, AMEX, or Nasdaq. Financial firms (SIC code 6000-6999) and utility firms (SIC codes

4900-4999) are excluded from the sample, leaving an unbalanced panel of 138,193 observations for 14,948 unique

firms. The cash-to-assets ratio (Cash/TA) is measured as cash plus marketable securities (CHE), divided by book

value of total assets (AT). The annual mean, median, and value-weighted average (based on annual book assets) of

Cash/TA in each subsample are regressed separately on a constant and a year index. Estimates of the slope

coefficient, p-value, and R-squared are reported for each industry group separately.

Panel A: the Fama-French 12 Industry Groups

Cash/TA_Mean Cash/TA_Median Cash/TA_VW

Description

(FF 12 industries) Slope P-value R-Sq Slope P-value R-Sq Slope P-value R-Sq

BusEq 0.85% <.0001 0.88 1.03% <.0001 0.88 0.72% <.0001 0.86

Chems 0.06% 0.144 0.08 -0.04% 0.191 0.06 0.03% 0.298 0.04

Durbl 0.05% 0.218 0.06 0.03% 0.502 0.02 0.00% 0.992 0.00

Enrgy -0.23% <.0001 0.57 -0.13% 0.000 0.42 -0.03% 0.453 0.02

Hlth 0.92% <.0001 0.88 1.11% <.0001 0.86 0.29% <.0001 0.53

Manuf 0.09% 0.002 0.31 0.04% 0.233 0.05 0.03% 0.313 0.04

NoDur -0.01% 0.842 0.00 -0.02% 0.362 0.03 -0.05% 0.031 0.17

Shops 0.03% 0.356 0.03 -0.02% 0.506 0.02 0.08% 0.020 0.19

Telcm 0.20% <.0001 0.45 0.15% <.0001 0.57 0.09% <.0001 0.51

other 0.15% 0.000 0.44 0.07% 0.027 0.17 0.01% 0.876 0.00

Panel B: the GICS Economic Sectors

Description

(GICS Economic Sectors)

Cash/TA_Mean Cash/TA_Median Cash/TA_VW

Slope P-value R-sq Slope P-value R-sq Slope P-value R-sq

Energy (10) -0.18% 0.000 0.42 -0.10% 0.005 0.26 -0.02% 0.648 0.01

Materials (15) -0.09% 0.004 0.28 -0.06% 0.016 0.20 0.03% 0.286 0.04

Industrial (20) 0.02% 0.573 0.01 -0.01% 0.739 0.00 -0.09% 0.002 0.31

Consumer discretionary (25) 0.08% 0.006 0.26 0.04% 0.181 0.07 -0.01% 0.830 0.00

Consumer staples (30) -0.06% 0.071 0.12 -0.06% 0.011 0.23 -0.08% 0.005 0.26

Health care (35) 0.82% <.0001 0.85 0.92% <.0001 0.86 0.25% <.0001 0.52

Financials (40) 0.04% 0.648 0.01 0.24% 0.006 0.25 -0.57% 0.118 0.09

Information technology (45) 0.86% <.0001 0.89 1.05% <.0001 0.90 0.80% <.0001 0.89

Telecommunications (50) 0.40% <.0001 0.72 0.31% <.0001 0.75 0.13% <.0001 0.51

Utilities (55) -0.07% 0.575 0.01 -0.09% 0.439 0.02 0.13% 0.150 0.08

62

Table 2.3: Descriptive Statistics

These tables present descriptive statistics of firm characteristics for all firms (Panel A), high-tech firms (Panel B),

and non-high-tech firms (Panel B) respectively. The sample includes U.S. firms documented on the Compustat-

CRSP merged database (fundamental annual) that have positive total assets and sales and nonnegative cash and

marketable securities, and have common shares traded on the NYSE, AMEX, or Nasdaq. Financial firms (SIC code

6000-6999) and utility firms (SIC codes 4900-4999) are excluded from the sample, leaving an unbalanced panel of

138,193 observations for 14,948 unique firms. The high-tech and non-high-tech sectors are defined according to the

U.S. Department of Commerce. High-tech sector contains firms from seven industries defined by 3-digit SIC codes

283, 357, 366, 367, 382, 384, and 737. The remaining firms in the sample are classified as non-high-tech. Size is

the logarithm of net assets. NWC is net working capital, equal to current assets minus current liabilities minus cash.

IndustrySigma is the mean of cash flow standard deviations of firms in the same industry, defined by 2-digit SIC

code. CF is operating income before depreciation, less interest and taxes. M/B is market value of equity plus total

assets minus book value of equity, and then divided by total assets. CAPEX is capital expenditure. R&D/Sales is

R&D expenditure to sales, where missing value of R&D expenditure is replaced by zero. Leverage is the ratio of

long-term debt plus debt in current liabilities to total assets. DivDummy is dividend dummy, set to one if common

dividend is positive. ACQN is acquisition expenditures. NetDiss is equal to long-term debt issuance minus long-

term debt reduction, scaled by total assets. NetEiss is equal to the sale of common and preferred stock minus the

purchase of common and preferred stock, scaled by total assets. Panel B presents summary statistics of firm

characteristics in two sectors. The t-statistics in Panel B are tests for equal means across two sectors. The z-

statistics in Panel B are from the Wilcoxon rank-sum test, which tests whether high-tech firms and non-high-tech

firms are from populations with the same distribution.

Panel A: All Firms

Variable N Mean

Standard

Deviation

25th

Percentile Median

75

Percentile

Cashta 138193 0.161 0.200 0.025 0.075 0.217

Size 138191 4.762 2.111 3.263 4.659 6.158

NWC/TA 134524 0.124 0.208 -0.011 0.120 0.269

Industry Sigma 138162 0.072 0.035 0.043 0.064 0.095

CF/TA 137890 0.017 0.207 0.008 0.070 0.116

MB 136324 1.918 1.705 1.001 1.342 2.084

Capex/TA 136592 0.073 0.076 0.024 0.049 0.091

RD/Sales 138193 0.151 0.716 0.000 0.000 0.040

Leverage 137784 0.239 0.208 0.054 0.209 0.364

DivDummy 138193 0.352 0.477 0.000 0.000 1.000

ACQN/TA 130292 0.019 0.055 0.000 0.000 0.002

NetDiss 133115 0.013 0.100 -0.018 0.000 0.030

NetEiss 134962 0.068 0.200 0.000 0.001 0.018

63

Panel B: High-Tech and Non-High-Tech Firms

Non-High-Tech Sector High-Tech Sector

Variable N Mean Median

Standard

Deviation

25th

Percentile

75th

Percentile N Mean Median

Standard

Deviation

25th

Percentile

75th

Percentile

t -

statistic

z -

statistic

Cashta 100324 0.113 0.054 0.149 0.020 0.142 37869 0.288 0.216 0.255 0.064 0.462 -125.9 -127.9

Size 100322 5.086 5.036 2.070 3.663 6.459 37869 3.904 3.698 1.974 2.528 5.053 98.0 96.7

NWC/TA 97017 0.127 0.120 0.206 -0.008 0.270 37507 0.116 0.118 0.213 -0.020 0.266 8.3 6.3

Industry Sigma 100293 0.061 0.055 0.029 0.039 0.079 37869 0.101 0.103 0.032 0.085 0.123 -207.9 -184.6

CF/TA 100142 0.044 0.074 0.160 0.026 0.116 37748 -0.054 0.052 0.287 -0.115 0.115 62.6 45.7

MB 98809 1.625 1.226 1.326 0.959 1.756 37515 2.688 1.886 2.260 1.244 3.208 -85.6 -110.2

Capex/TA 99171 0.078 0.053 0.081 0.026 0.099 37421 0.058 0.040 0.060 0.020 0.073 51.5 46.0

RD/Sales 100324 0.038 0.000 0.339 0.000 0.006 37869 0.450 0.088 1.201 0.032 0.201 -65.8 -233.4

Leverage 100042 0.269 0.247 0.206 0.103 0.393 37742 0.158 0.086 0.192 0.003 0.252 93.6 101.8

DivDummy 100324 0.426 0.000 0.494 0.000 1.000 37869 0.155 0.000 0.362 0.000 0.000 111.7 94.2

ACQN/TA 94893 0.020 0.000 0.056 0.000 0.003 35399 0.017 0.000 0.052 0.000 0.000 7.6 13.1

NetDiss 96535 0.014 0.000 0.102 -0.020 0.039 36580 0.009 0.000 0.094 -0.012 0.004 7.8 8.3

NetEiss 97993 0.044 0.000 0.159 0.000 0.008 36969 0.131 0.009 0.270 0.000 0.099 -58.4 -89.4

64

Table 2.4: Correlation Matrix

These tables present correlations matrices of firm characteristics of all firms (Panel A), high-tech firms (Panel B), and non-high-tech firms (Panel C)

respectively. The sample includes U.S. firms documented on the Compustat-CRSP merged database (fundamental annual) that have positive total assets and

sales and nonnegative cash and marketable securities, and have common shares traded on the NYSE, AMEX, or Nasdaq. Financial firms (SIC code 6000-6999)

and utility firms (SIC codes 4900-4999) are excluded from the sample, leaving an unbalanced panel of 138,193 observations for 14,948 unique firms. The high-

tech and non-high-tech sectors are defined according to the U.S. Department of Commerce. High-tech sector contains firms from seven industries defined by 3-

digit SIC codes 283, 357, 366, 367, 382, 384, and 737. The remaining firms in the sample are classified as non-high-tech. Size is the logarithm of net assets.

NWC is net working capital, equal to current assets minus current liabilities minus cash. IndustrySigma is the mean of cash flow standard deviations of firms in

the same industry, defined by 2-digit SIC code. CF is operating income before depreciation, less interest and taxes. M/B is market value of equity plus total

assets minus book value of equity, and then divided by total assets. CAPEX is capital expenditure. R&D/Sales is R&D expenditure to sales, where missing value

of R&D expenditure is replaced by zero. Leverage is the ratio of long-term debt plus debt in current liabilities to total assets. DivDummy is dividend dummy, set

to one if common dividend is positive. ACQN is acquisition expenditures. NetDiss is equal to long-term debt issuance minus long-term debt reduction, scaled by

total assets. NetEiss is equal to the sale of common and preferred stock minus the purchase of common and preferred stock, scaled by total assets. P-values are

presented in italic.

65

Panel A: All Firms

Cashta Size NWC/TA

Industry

Sigma CF/TA MB

Capex

/TA

RD

/Sales Leverage

Div

Dummy

ACQN

/TA NetDiss

Size -0.396

0.000

NWC/TA -0.259 0.052

0.000 0.000

Industry Sigma 0.386 -0.192 -0.218

0.000 0.000 0.000

CF/TA -0.304 0.425 0.339 -0.243

0.000 0.000 0.000 0.000

MB 0.388 -0.267 -0.222 0.287 -0.320

0.000 0.000 0.000 0.000 0.000

Capex/TA -0.139 0.030 -0.207 -0.125 0.063 0.049

0.000 0.000 0.000 0.000 0.000 0.000

RD/Sales 0.431 -0.234 -0.185 0.289 -0.482 0.290 -0.054

0.000 0.000 0.000 0.000 0.000 0.000 0.000

Leverage -0.422 0.197 -0.197 -0.184 -0.054 -0.181 0.101 -0.106

0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

DivDummy -0.224 0.461 0.191 -0.306 0.289 -0.183 0.014 -0.141 -0.053

0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

ACQN/TA -0.095 0.127 -0.058 0.064 0.052 -0.011 -0.075 -0.044 0.106 -0.017

0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

NetDiss -0.045 0.065 -0.003 -0.014 -0.039 0.006 0.215 0.013 0.244 0.017 0.291

0.000 0.000 0.354 0.000 0.000 0.020 0.000 0.000 0.000 0.000 0.000

NetEiss 0.425 -0.348 -0.180 0.199 -0.448 0.437 0.056 0.355 -0.157 -0.225 0.025 -0.086

0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

66

Panel B: High-Tech Firms

Cashta Size NWC/TA

Industry

Sigma CF/TA MB

Capex

/TA

RD

/Sales Leverage

Div

Dummy

ACQN

/TA NetDiss

Size -0.335

0.000

NWC/TA -0.426 0.116

0.000 0.000

Industry Sigma 0.381 -0.010 -0.423

0.000 0.051 0.000

CF/TA -0.290 0.479 0.460 -0.233

0.000 0.000 0.000 0.000

MB 0.333 -0.214 -0.264 0.180 -0.295

0.000 0.000 0.000 0.000 0.000

Capex/TA -0.193 0.080 0.012 -0.254 0.034 0.049

0.000 0.000 0.026 0.000 0.000 0.000

RD/Sales 0.450 -0.281 -0.314 0.317 -0.535 0.258 -0.055

0.000 0.000 0.000 0.000 0.000 0.000 0.000

Leverage -0.396 0.104 -0.111 -0.140 -0.100 -0.119 0.119 -0.054

0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

DivDummy -0.214 0.374 0.193 -0.210 0.246 -0.121 0.052 -0.139 -0.001

0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.811

ACQN/TA -0.121 0.188 -0.052 0.084 0.095 -0.049 -0.074 -0.071 0.066 0.012

0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.028

NetDiss -0.028 0.055 -0.010 0.000 -0.070 0.026 0.147 0.042 0.286 0.010 0.195

0.000 0.000 0.057 0.954 0.000 0.000 0.000 0.000 0.000 0.057 0.000

NetEiss 0.420 -0.361 -0.248 0.130 -0.452 0.417 0.017 0.382 -0.133 -0.183 -0.039 -0.083

0.000 0.000 0.000 0.000 0.000 0.000 0.001 0.000 0.000 0.000 0.000 0.000

67

Panel C: Non-High-Tech Firms

Cashta Size NWC/TA

Industry

Sigma CF/TA MB

Capex

/TA

RD

/Sales Leverage

Div

Dummy

ACQN

/TA NetDiss

Size -0.356

0.000

NWC/TA -0.180 0.021

0.000 0.000

Industry Sigma 0.144 -0.106 -0.158

0.000 0.000 0.000

CF/TA -0.198 0.371 0.279 -0.117

0.000 0.000 0.000 0.000

MB 0.297 -0.223 -0.207 0.181 -0.259

0.000 0.000 0.000 0.000 0.000

Capex/TA -0.068 -0.022 -0.280 -0.021 0.047 0.116

0.000 0.000 0.000 0.000 0.000 0.000

RD/Sales 0.262 -0.144 -0.088 0.084 -0.330 0.226 -0.009

0.000 0.000 0.000 0.000 0.000 0.000 0.007

Leverage -0.377 0.161 -0.245 -0.053 -0.125 -0.133 0.063 -0.065

0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

DivDummy -0.123 0.440 0.194 -0.219 0.286 -0.134 -0.031 -0.078 -0.152

0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

ACQN/TA -0.086 0.106 -0.061 0.088 0.023 0.021 -0.080 -0.024 0.116 -0.032

0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

NetDiss -0.050 0.065 0.000 -0.005 -0.032 0.006 0.233 0.001 0.233 0.013 0.322

0.000 0.000 0.894 0.120 0.000 0.074 0.000 0.767 0.000 0.000 0.000

NetEiss 0.350 -0.306 -0.141 0.115 -0.392 0.393 0.123 0.246 -0.113 -0.207 0.072 -0.090

0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

68

Table 2.5: Determinants of Corporate Cash Holdings: 1974-2007

This table analyzes the impact of firm characteristics on firms‟ cash holdings over the period from 1974 to 2007.

The sample includes U.S. firms documented on the Compustat-CRSP merged database (fundamental annual) that

have positive total assets and sales and nonnegative cash and marketable securities, and have common shares traded

on the NYSE, AMEX, or Nasdaq. Financial firms (SIC code 6000-6999) and utility firms (SIC codes 4900-4999)

are excluded from the sample, leaving an unbalanced panel of 138,193 observations for 14,948 unique firms. The

regression equation is:

0 1 2 3 4 5 6

7 8 9 10

11 12 13 14

( )

&

it it it itit it it

it it it it

it itit it

it it

it it t

Cash NWC CF CAPEXSize IndustrySigma MB

TA TA TA TA

R D ACQNLeverage DivDummy

Sales TA

NetDiss NetEiss TbillYield

t itDefaultSpread

Size is the logarithm of net assets. NWC is net working capital, equal to current assets minus current liabilities

minus cash. IndustrySigma is the mean of cash flow standard deviations of firms in the same industry, defined by

2-digit SIC code. CF is operating income before depreciation, less interest and taxes. M/B is market value of equity

plus total assets minus book value of equity, and then divided by total assets. CAPEX is capital expenditure.

R&D/Sales is R&D expenditure to sales, where missing value of R&D expenditure is replaced by zero. Leverage is

the ratio of long-term debt plus debt in current liabilities to total assets. DivDummy is dividend dummy, set to one

if common dividend is positive. ACQN is acquisition expenditures. NetDiss is equal to long-term debt issuance

minus long-term debt reduction, scaled by total assets. NetEiss is equal to the sale of common and preferred stock

minus the purchase of common and preferred stock, scaled by total assets. TbillYield is the average annual three-

month rates published by the Federal Reserve. DefaultSpread is the average yield on Baa less Aaa Moody‟s rated

corporate bonds with maturity of approximately 20-25 years. Missing explanatory values reduce the panel data used

here to 117,240 firm-year observations for 14,242 unique firms. Panel A reports the estimating results. Column (1)-

(6) are based on the pooled sample with all firms. Column (1) uses OLS regression to estimate the basic model

designed by Opler et al. (1999). Columns (2)-(5) use OLS regressions to estimate the modified model, without/with

dummy variables for industry and year. In Column (6), the coefficients and standard errors are estimated using the

Fama-MacBeth method (1973). Industries are defined according to the 2-digit SIC. Column (7) and (8) are separate

regressions for the high-tech and non-high-tech sectors. Column (9) tests whether the coefficients estimates for the

high-tech sector are different from the non-high-tech sector. P-values are reported. For each regression, adjusted R2

is reported. The standard errors are adjusted for clustering on firms. They are computed assuming observations are

independent across firms, but not across time. t-statistics are reported in the parentheses. *, **, and *** indicate

statistical significance at the 10%, 5%, and 1% levels, respectively. „F-test‟ indicates the joint test that all

coefficients are equal between the high-tech and non-high-tech sectors. Panel B reports the relative importance of

the explanatory variables in the regression with the pooled sample (Column (2) of Panel A), as well as the

regressions with high-tech firms (Column (7) of Panel A) and non-high-tech firms (Column (8) of Panel A)

respectively. Relative importance is obtained by means of variance decomposition as proposed by Grömping (2007).

For each determinant two values are reported: the percentage of the variation of the dependant variable that it

explains (absolute value) and the percentage of the variation explained within the regression model (standardized

value).

69

Panel A:

All Firms High-

Tech

Non-

High-Tech

Diff

(P-value)

OLS OLS

Industry

FE

Year

FE

Industry

and Year

FE

Fama-

MacBeth OLS OLS

[1] [2] [3] [4] [5] [6] [7] [8] [9]

Size -0.020*** -0.018*** -0.019*** -0.019*** -0.020*** -0.019*** -0.022*** -0.017*** 0.000

(-34.21) (-30.12) (-30.56) (-30.86) (-31.56) (-23.67) (-17.01) (-26.75)

NWC/TA -0.265*** -0.287*** -0.310*** -0.286*** -0.307*** -0.283*** -0.450*** -0.224*** 0.000

(-50.04) (-53.60) (-51.23) (-53.23) (-50.91) (-33.98) (-38.52) (-37.77)

Industry Sigma 0.651*** 0.674*** 0.565*** 0.704*** 0.624*** 0.416*** 0.580*** 0.161*** 0.000

(20.06) (20.48) (14.62) (18.72) (12.39) (7.72) (6.69) (4.85)

CF/TA 0.060*** 0.127*** 0.133*** 0.131*** 0.137*** 0.134*** 0.191*** 0.089*** 0.000

(11.44) (24.13) (25.11) (24.90) (25.87) (16.41) (23.94) (12.83)

MB 0.016*** 0.009*** 0.008*** 0.009*** 0.008*** 0.010*** 0.006*** 0.008*** 0.174

(25.86) (14.60) (13.28) (14.47) (13.17) (9.23) (7.71) (9.26)

Capex/TA -0.396*** -0.548*** -0.498*** -0.542*** -0.492*** -0.541*** -0.644*** -0.427*** 0.000

(-37.59) (-50.15) (-44.84) (-49.25) (-43.98) (-24.45) (-25.90) (-35.52)

RD/Sales 0.068*** 0.059*** 0.057*** 0.058*** 0.056*** 0.062*** 0.051*** 0.059*** 0.113

(42.57) (37.43) (34.72) (37.09) (34.43) (13.69) (29.64) (13.15)

Leverage -0.342*** -0.354*** -0.355*** -0.348*** -0.350*** -0.322*** -0.484*** -0.279*** 0.000

(-66.66) (-65.92) (-64.97) (-64.51) (-63.69) (-32.62) (-37.00) (-48.10)

DivDummy -0.008*** -0.008*** -0.009*** -0.008*** -0.008*** -0.009*** -0.022*** 0.002 0.000

(-4.03) (-4.18) (-4.29) (-3.97) (-4.01) (-2.94) (-3.85) (1.15)

ACQN/TA -0.221*** -0.444*** -0.431*** -0.446*** -0.433*** -0.394*** -0.617*** -0.346*** 0.000

(-27.67) (-48.58) (-47.28) (-48.14) (-46.99) (-18.85) (-30.03) (-37.24)

NetDiss 0.315*** 0.309*** 0.308*** 0.303*** 0.281*** 0.392*** 0.253*** 0.000

(48.44) (48.03) (47.48) (47.16) (23.78) (25.94) (36.72)

NetEiss 0.205*** 0.205*** 0.209*** 0.208*** 0.182*** 0.184*** 0.212*** 0.000

(53.25) (53.12) (53.88) (53.75) (12.80) (32.99) (39.27)

TbillYield -0.132*** -0.143*** -0.305*** -0.091*** 0.001

(-6.03) (-6.97) (-5.13) (-4.16)

Default Spread 1.893*** 1.816*** 2.069*** 1.305*** 0.043

(15.01) (14.26) (5.79) (10.87)

Constant 0.320*** 0.311*** 0.297*** 0.344*** 0.328*** 0.330*** 0.436*** 0.291*** 0.000

(62.14) (52.37) (20.67) (51.22) (21.80) (39.04) (29.64) (44.34)

Observations 117240 117240 117240 117240 117240 117240 32583 84657

Adj. R-squared 0.519 0.558 0.566 0.562 0.569 0.586 0.423

F-test for two sectors

F( 15, 117210) 538.26

Prob > F 0.0000

70

Panel B: Relative Importance of Determinants

Absolute Values Standardized Values

All Firms High-Tech

Non-

High-Tech All Firms High-Tech

Non-

High-Tech

Size 6.4% 3.4% 7.9% 11.5% 5.8% 18.6%

NWC/TA 6.6% 15.5% 5.4% 11.9% 26.4% 12.7%

Industry Sigma 3.8% 1.4% 0.1% 6.8% 2.3% 0.2%

CF/TA 1.0% 1.6% 0.6% 1.8% 2.7% 1.3%

MB 1.5% 0.5% 1.0% 2.6% 0.9% 2.4%

Capex/TA 2.2% 2.1% 2.0% 4.0% 3.6% 4.8%

RD/Sales 10.0% 10.5% 2.8% 17.9% 17.9% 6.6%

Leverage 15.5% 15.5% 14.0% 27.7% 26.4% 33.1%

DivDummy 0.1% 0.1% 0.0% 0.1% 0.2% 0.0%

ACQN/TA 0.8% 1.2% 0.9% 1.5% 2.0% 2.0%

NetDiss 1.1% 1.2% 1.1% 1.9% 2.0% 2.7%

NetEiss 6.8% 5.6% 6.5% 12.1% 9.5% 15.3%

TbillYield 0.0% 0.1% 0.0% 0.1% 0.2% 0.0%

Default Spread 0.1% 0.1% 0.1% 0.2% 0.1% 0.2%

Total Variation 55.8% 58.7% 42.3% 100% 100% 100%

By 5 Core Variables 45.3% 50.4% 36.5% 81.1% 85.9% 86.3%

By R&D and NWC 16.6% 26.0% 8.2% 29.7% 44.3% 19.3%

71

Table 2.6: Changes in Firm Characteristics

This table compares the characteristics of firms over the period from 1974 to 2007. The sample includes U.S. firms documented on the Compustat-CRSP

merged database (fundamental annual) that have positive total assets and sales and nonnegative cash and marketable securities, and have common shares traded

on the NYSE, AMEX, or Nasdaq. Financial firms (SIC code 6000-6999) and utility firms (SIC codes 4900-4999) are excluded from the sample, leaving an

unbalanced panel of 138,193 observations for 14,948 unique firms. Size is the logarithm of net assets. NWC is net working capital, equal to current assets minus

current liabilities minus cash. IndustrySigma is the mean of cash flow standard deviations of firms in the same industry, defined by 2-digit SIC code. CF is

operating income before depreciation, less interest and taxes. M/B is market value of equity plus total assets minus book value of equity, and then divided by

total assets. CAPEX is capital expenditure. R&D/Sales is R&D expenditure to sales, where missing value of R&D expenditure is replaced by zero. Leverage is

the ratio of long-term debt plus debt in current liabilities to total assets. DivDummy is dividend dummy, set to one if common dividend is positive. ACQN is

acquisition expenditures. NetDiss is equal to long-term debt issuance minus long-term debt reduction, scaled by total assets. NetEiss is equal to the sale of

common and preferred stock minus the purchase of common and preferred stock, scaled by total assets. In Panel A, the total sample is split into four sub-periods

for two sectors. Obs is the total number of observations of a given sector in the indicated sub-period. The results for each period are simple averages over all

firm-year observations locating in a given sub-sample. Firm characteristic of high-tech firms and non-high-tech firms over 1980-2007 are regressed separately

on a constant and a year index. Estimates of the slope coefficient are reported in the row titled Time Trends [1980, 2007]. P-values are also reported. In Panel B,

the observations in each sub-period are further divided according to listing cohorts. Based on the year of going public, firms are sorted into the following cohorts:

pre-1980 IPO (listed before 1980), 1980s IPO (listed from 1980 to 1989), 1990s IPO (listed from 1990 to 2000), and 2000s IPO (listed from 2001 to 2006). IPO

dates are identified first by using Jay Ritter‟s proprietary database of IPO dates. If the IPO date of a stock is unavailable from Ritter, the first trading date on the

CRSP is identified as the IPO date. Obs is the total number of observations of a given listing cohorts in a given sector in the indicated sub-period. The results for

each period are simple averages over all firm-year observations locating in a given sub-sample defined according to listing cohort and sub-period.

72

Panel A: Sub-Periods

Sector Sub-Period Obs Cash/TA Size NWC/TA

Industry

Sigma CF/TA MB

Capex

/TA

R&D

/TA

R&D

/Sales Leverage

Div

Dummy ACQN/TA NetDiss NetEiss

Non-

High-Tech

[1974, 1979] 17112 0.086 5.220 0.204 0.034 0.078 1.072 0.079 0.009 0.009 0.267 0.658 0.007 0.014 0.004

[1980, 1989] 30352 0.118 4.667 0.133 0.056 0.029 1.615 0.090 0.012 0.034 0.277 0.466 0.017 0.017 0.051

[1990, 2000] 36613 0.112 5.025 0.107 0.070 0.036 1.788 0.076 0.014 0.045 0.273 0.323 0.026 0.015 0.064

[2001, 2007] 16247 0.132 5.867 0.076 0.077 0.051 1.813 0.061 0.016 0.059 0.247 0.339 0.024 0.007 0.025

Time trend

[1980, 2007] 0.001 0.053 -0.003 0.001 0.001 0.013 -0.001 0.001 -0.001 0.001 0.000 -0.001

p-value <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 0.129 <.0001

High-Tech

[1974, 1979] 2696 0.096 4.262 0.291 0.043 0.079 1.497 0.080 0.055 0.050 0.250 0.414 0.005 0.017 0.023

[1980, 1989] 8928 0.205 3.591 0.198 0.074 -0.031 2.382 0.078 0.089 0.238 0.199 0.220 0.011 0.013 0.117

[1990, 2000] 16560 0.308 3.765 0.103 0.106 -0.064 3.029 0.056 0.131 0.502 0.141 0.115 0.018 0.007 0.172

[2001, 2007] 9685 0.384 4.332 0.015 0.131 -0.097 2.683 0.035 0.141 0.668 0.124 0.091 0.024 0.008 0.104

Time trend

[1980, 2007] 0.010 0.038 -0.010 0.003 -0.004 0.027 -0.002 0.024 -0.004 0.001 0.000 0.000

p-value <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 0.676 0.119

73

Panel B: Listing Cohorts in Each Sub-Period

Sector IPO Cohort Obs Cash/TA Size NWC/TA CF/TA MB Capex/TA R&D/TA R&D/Sales Leverage

Div

Dummy ACQN/TA NetDiss NetEiss

Sub-Period: [1980, 1989]

Non-High-Tech

Overall 30352 0.118 4.667 0.133 0.029 1.615 0.090 0.012 0.034 0.277 0.466 0.017 0.017 0.051

Pre-1980 IPO 20059 0.098 5.287 0.164 0.057 1.361 0.079 0.010 0.012 0.271 0.622 0.014 0.012 0.011

1980s IPO 10293 0.157 3.459 0.073 -0.026 2.111 0.113 0.017 0.079 0.287 0.161 0.022 0.027 0.128

High-Tech

Overall 8928 0.205 3.591 0.198 -0.031 2.382 0.078 0.089 0.238 0.199 0.220 0.011 0.013 0.117

Pre-1980 IPO 4099 0.133 4.488 0.245 0.047 1.875 0.077 0.067 0.072 0.217 0.402 0.010 0.012 0.036

1980s IPO 4829 0.267 2.829 0.159 -0.097 2.812 0.078 0.108 0.379 0.183 0.065 0.011 0.013 0.188

Sub-Period: [1990, 2000]

Non-High-Tech

Overall 36613 0.112 5.025 0.107 0.036 1.788 0.076 0.014 0.045 0.273 0.323 0.026 0.015 0.064

Pre-1980 IPO 11143 0.083 6.031 0.137 0.075 1.514 0.065 0.011 0.011 0.260 0.627 0.019 0.007 0.000

1980s IPO 10708 0.109 4.505 0.109 0.034 1.731 0.074 0.015 0.039 0.280 0.256 0.019 0.012 0.036

1990s IPO 14762 0.137 4.642 0.082 0.008 2.038 0.086 0.016 0.075 0.278 0.142 0.036 0.023 0.133

High-Tech

Overall 16560 0.308 3.765 0.103 -0.064 3.029 0.056 0.131 0.502 0.141 0.115 0.018 0.007 0.172

Pre-1980 IPO 2626 0.135 5.183 0.193 0.067 1.974 0.055 0.076 0.080 0.196 0.383 0.019 0.004 0.017

1980s IPO 5535 0.248 3.640 0.133 -0.026 2.738 0.055 0.125 0.348 0.160 0.106 0.015 0.008 0.084

1990s IPO 8399 0.402 3.404 0.055 -0.129 3.551 0.058 0.151 0.735 0.112 0.036 0.020 0.008 0.276

Sub-Period: [2001, 2007]

Non-High-Tech

Overall 16247 0.132 5.867 0.076 0.051 1.813 0.061 0.016 0.059 0.247 0.339 0.024 0.007 0.025

Pre-1980 IPO 4001 0.093 6.753 0.108 0.085 1.605 0.053 0.010 0.014 0.241 0.664 0.022 0.001 -0.005

1980s IPO 2586 0.122 5.548 0.115 0.060 1.725 0.058 0.017 0.039 0.234 0.383 0.020 0.005 0.010

1990s IPO 7627 0.146 5.562 0.060 0.037 1.814 0.061 0.019 0.083 0.251 0.190 0.024 0.009 0.022

2000s IPO 2033 0.164 5.669 0.027 0.026 2.329 0.079 0.013 0.086 0.259 0.207 0.036 0.012 0.112

High-Tech

Overall 9685 0.384 4.332 0.015 -0.097 2.683 0.035 0.141 0.668 0.124 0.091 0.024 0.008 0.104

Pre-1980 IPO 903 0.227 6.116 0.129 0.067 2.040 0.036 0.069 0.120 0.145 0.430 0.024 0.006 0.003

1980s IPO 1646 0.316 4.542 0.076 -0.027 2.491 0.034 0.116 0.382 0.133 0.140 0.023 0.004 0.051

1990s IPO 5971 0.407 4.102 -0.014 -0.129 2.691 0.035 0.155 0.744 0.125 0.038 0.023 0.010 0.098

2000s IPO 1165 0.484 3.829 -0.010 -0.161 3.415 0.035 0.161 1.108 0.093 0.035 0.028 0.006 0.281

74

Table 2.7: Determinants of Corporate Cash Holdings: Estimation vs.

Forecast Periods

This table presents the results of coefficient estimates for the high-tech and non-high-tech sectors, jointly and

separately, during the estimation period from 1974 to 1983 and the forecast period from 1984 to 2007. The sample

includes U.S. firms documented on the Compustat-CRSP merged database (fundamental annual) that have positive

total assets and sales and nonnegative cash and marketable securities, and have common shares traded on the NYSE,

AMEX, or Nasdaq. Financial firms (SIC code 6000-6999) and utility firms (SIC codes 4900-4999) are excluded

from the sample, leaving an unbalanced panel of 138,193 observations, 34,235 in estimation period and 103,958 in

forecast period. The regression equation is:

0 1 2 3 4 5 6

7 8 9 10

11 12 13 14

( )

&

it it it itit it it

it it it it

it itit it

it it

it it t

Cash NWC CF CAPEXSize IndustrySigma MB

TA TA TA TA

R D ACQNLeverage DivDummy

Sales TA

NetDiss NetEiss TbillYield

t itDefaultSpread

Size is the logarithm of net assets. NWC is net working capital, equal to current assets minus current liabilities

minus cash. IndustrySigma is the mean of cash flow standard deviations of firms in the same industry, defined by

2-digit SIC code. CF is operating income before depreciation, less interest and taxes. M/B is market value of equity

plus total assets minus book value of equity, and then divided by total assets. CAPEX is capital expenditure.

R&D/Sales is R&D expenditure to sales, where missing value of R&D expenditure is replaced by zero. Leverage is

the ratio of long-term debt plus debt in current liabilities to total assets. DivDummy is dividend dummy, set to one

if common dividend is positive. ACQN is acquisition expenditures. NetDiss is equal to long-term debt issuance

minus long-term debt reduction, scaled by total assets. NetEiss is equal to the sale of common and preferred stock

minus the purchase of common and preferred stock, scaled by total assets. TbillYield is the average annual three-

month rates published by the Federal Reserve. DefaultSpread is the average yield on Baa less Aaa Moody‟s rated

corporate bonds with maturity of approximately 20-25 years. Missing explanatory values reduce the panel data used

here to 29,004 firm-year observations in the regression for the estimation period and 88,236 firm-year observations

in the regression for the forecast period. Results from OLS regressions are reported for the pooled sample (Column

(1)-(2)), the high-tech sector (Column (3)-(4)), and the non-high-tech sector (Column (5)-(6)). The standard errors

are adjusted for clustering on firms. They are computed assuming observations are independent across firms, but

not across time. t-statistics are reported in the parentheses. *, **, and *** indicate statistical significance at the 10%,

5%, and 1% levels, respectively. #, ##, and ### indicate statistical significance at the 10%, 5%, and1% levels,

respectively, for a t-test that tests whether the coefficients are equal between the two sub-periods. „F-test‟ indicates

the joint test that all coefficients are equal between estimation and forecast periods.

75

All Firms High-Tech Non-High-Tech

1974-1983 1984-2007 1974-1983 1984-2007 1974-1983 1984-2007

[1] [2] [3] [4] [5] [6]

Size -0.016*** -0.019*** ###

-0.014*** -0.024*** ###

-0.016*** -0.017***

(-21.58) (-26.41)

(-7.12) (-16.64)

(-20.75) (-22.30)

NWC/TA -0.258*** -0.294*** ###

-0.441*** -0.451***

-0.234*** -0.221***

(-32.04) (-47.68)

(-20.18) (-35.59)

(-26.93) (-32.06)

Industry Sigma 0.371*** 0.709*** ###

0.612** 0.598***

0.288*** 0.178***

(4.30) (18.96)

(2.03) (5.43)

(3.18) (4.60)

CF/TA 0.129*** 0.131***

0.217*** 0.193***

0.094*** 0.090***

(10.12) (23.10)

(9.46) (22.88)

(6.26) (11.71)

MB 0.005*** 0.010*** ###

0.003 0.007***

0.005*** 0.009*** ##

(3.88) (15.36)

(1.34) (8.36)

(3.13) (9.30)

Capex/TA -0.445*** -0.574*** ###

-0.607*** -0.649***

-0.404*** -0.439***

(-26.03) (-44.57)

(-14.60) (-22.48)

(-21.47) (-30.98)

RD/Sales 0.043*** 0.058*** ##

0.036*** 0.051*** #

0.051*** 0.059***

(6.27) (36.21)

(4.16) (28.57)

(5.10) (12.50)

Leverage -0.270*** -0.370*** ###

-0.355*** -0.498*** ###

-0.258*** -0.284*** ###

(-33.62) (-59.96)

(-15.71) (-34.47)

(-30.17) (-42.54)

DivDummy 0.009*** -0.017*** ###

-0.001 -0.030*** ###

0.011*** -0.002 ###

(3.65) (-6.78)

(-0.10) (-4.33)

(4.47) (-0.88)

ACQN/TA -0.382*** -0.437*** ##

-0.553*** -0.607***

-0.350*** -0.343***

(-17.69) (-44.28)

(-9.27) (-28.33)

(-15.40) (-33.89)

NetDiss 0.245*** 0.324*** ###

0.346*** 0.398*** #

0.227*** 0.259*** ##

(21.49) (43.81)

(12.70) (23.83)

(18.07) (32.73)

NetEiss 0.279*** 0.197*** ###

0.290*** 0.175*** ###

0.261*** 0.204*** ###

(26.78) (47.55)

(16.99) (30.20)

(19.87) (34.42)

TbillYield -0.093*** -0.230*** ###

-0.142** -0.351*** ##

-0.081*** -0.143***

(-4.83) (-6.26)

(-2.46) (-4.09)

(-3.98) (-3.83)

Default Spread 0.420*** 2.994*** ###

1.020*** 2.809*** ###

0.284*** 2.623*** ###

(4.35) (12.44)

(3.78) (5.17)

(2.79) (10.78)

Constant 0.295*** 0.311***

0.360*** 0.437*** ###

0.288*** 0.282***

(39.77) (42.58) (15.72) (23.80)

(36.44) (34.00)

Observations 29004 88236 4492 28091 24512 60145

Adj. R-squared 0.466 0.558 0.596 0.561 0.424 0.421

F-test 36.36 9.19 10.49

p-value 0.00 0.00 0.00

76

Table 2.8: Actual and Predicted Cash Holdings: Forecast Period

This table reports the predicted cash ratios and the difference between actual and predicted cash ratios in the high-

tech sector (Panel A) and in the non-high-tech sector (Panel B) during the forecast period from 1984 to 2007. The

sample includes U.S. firms documented on the Compustat-CRSP merged database (fundamental annual) that have

positive total assets and sales and nonnegative cash and marketable securities, and have common shares traded on

the NYSE, AMEX, or Nasdaq. Financial firms (SIC code 6000-6999) and utility firms (SIC codes 4900-4999) are

excluded from the sample, leaving an unbalanced panel of 103,958 observations during the period from 1984 to

2007, including 32,541 observations in the high-tech sector and 71,417 observations in the non-high-tech sector.

Due to missing explanatory values, I can calculate expected cash holdings for 28,091 observations in the high-tech

sector and 60,145 observations in the non-high-tech sector. Predicted cash holdings are calculated for each firm-

year observation by fitting the firm characteristics in forecasting period into the cash holding model estimated in the

estimating period. Panel A reports the findings for the high-tech sector, where coefficients estimates during the

estimation period come from separate regression on high-tech firms in the estimation period and from the joint

estimation with pooled sample. Panel B reports the findings for the non-high-tech sector, where coefficients

estimates during the estimation period come from separate regression on high-tech firms in the estimation period

and from the joint estimation with pooled sample. Annual mean of actual cash, predicted cash, as well as the

deviations of the actual cash ratios from predicted cash holdings, are reported for two sectors separately. T-statistics

summarize the statistical significance of the deviations of the actual cash ratios from expected cash holdings in each

year.

77

Panel A: High-Tech Sector

Separate Estimation Pooled Estimation

Year Firms Actual Expected

Actual-

Expected t-stat Expected

Actual-

Expected t-stat

1984 832 0.225 0.195 0.030 6.07 0.181 0.044 8.48

1985 824 0.206 0.181 0.025 4.81 0.166 0.040 7.31

1986 868 0.225 0.211 0.014 2.72 0.189 0.036 6.68

1987 917 0.242 0.230 0.012 2.33 0.203 0.039 7.08

1988 933 0.221 0.214 0.007 1.29 0.186 0.035 6.37

1989 931 0.205 0.217 -0.012 -2.26 0.189 0.017 3.03

1990 907 0.207 0.228 -0.021 -3.99 0.194 0.013 2.43

1991 947 0.252 0.272 -0.020 -3.95 0.233 0.019 3.77

1992 1035 0.270 0.287 -0.017 -3.21 0.247 0.023 4.36

1993 1135 0.297 0.295 0.002 0.38 0.255 0.043 8.22

1994 1178 0.287 0.284 0.003 0.61 0.244 0.043 8.34

1995 1276 0.309 0.305 0.004 0.84 0.262 0.047 9.71

1996 1474 0.347 0.331 0.016 3.69 0.286 0.060 13.26

1997 1603 0.346 0.309 0.038 8.01 0.266 0.080 16.58

1998 1532 0.327 0.297 0.030 6.20 0.251 0.075 15.35

1999 1508 0.347 0.329 0.018 3.86 0.281 0.066 14.11

2000 1547 0.377 0.351 0.026 5.69 0.300 0.077 16.41

2001 1417 0.381 0.325 0.055 9.92 0.266 0.114 20.42

2002 1334 0.374 0.337 0.038 6.60 0.267 0.108 18.77

2003 1194 0.398 0.367 0.032 5.69 0.291 0.108 19.09

2004 1222 0.409 0.376 0.032 6.08 0.303 0.105 19.48

2005 1205 0.404 0.356 0.048 8.78 0.285 0.120 21.65

2006 1145 0.400 0.353 0.047 8.35 0.284 0.116 20.53

2007 1127 0.399 0.353 0.046 7.98 0.282 0.117 20.01

78

Panel B: Non-High-Tech Sector

Separate Estimation Pooled Estimation

Year Firms Actual Expected

Actual-

Expected t-stat Expected

Actual-

Expected t-stat

1984 2714 0.125 0.125 0.000 0.14 0.127 -0.002 -1.03

1985 2599 0.121 0.121 0.000 0.00 0.124 -0.003 -1.08

1986 2565 0.133 0.130 0.004 1.40 0.133 0.000 -0.02

1987 2648 0.129 0.132 -0.003 -1.01 0.136 -0.007 -2.64

1988 2602 0.118 0.122 -0.005 -1.85 0.126 -0.009 -3.43

1989 2452 0.114 0.123 -0.009 -3.59 0.127 -0.013 -5.12

1990 2424 0.108 0.124 -0.016 -6.38 0.129 -0.021 -8.19

1991 2477 0.117 0.137 -0.020 -8.01 0.143 -0.026 -10.27

1992 2566 0.120 0.144 -0.024 -9.89 0.150 -0.030 -12.37

1993 2810 0.127 0.149 -0.022 -9.94 0.155 -0.029 -12.65

1994 3026 0.112 0.140 -0.028 -13.25 0.145 -0.033 -15.80

1995 3074 0.108 0.134 -0.026 -12.36 0.139 -0.031 -14.82

1996 3174 0.122 0.141 -0.019 -8.99 0.146 -0.025 -11.58

1997 3130 0.117 0.128 -0.011 -5.06 0.133 -0.016 -7.18

1998 2931 0.108 0.115 -0.007 -3.07 0.119 -0.011 -5.06

1999 2686 0.111 0.120 -0.009 -3.86 0.125 -0.014 -6.10

2000 2450 0.119 0.125 -0.006 -2.43 0.131 -0.011 -4.44

2001 2179 0.122 0.125 -0.004 -1.32 0.132 -0.010 -3.47

2002 2026 0.127 0.131 -0.004 -1.25 0.138 -0.011 -3.77

2003 1954 0.133 0.136 -0.003 -1.11 0.145 -0.012 -4.07

2004 1941 0.144 0.142 0.002 0.73 0.150 -0.006 -2.23

2005 1923 0.145 0.140 0.005 1.82 0.148 -0.003 -1.04

2006 1921 0.145 0.134 0.010 3.60 0.142 0.002 0.86

2007 1873 0.142 0.132 0.010 3.55 0.139 0.003 0.87

79

Figure 2.1: Trends in Cash Holdings and Net Leverage: 1974-2007

These figures depict the annual mean, median, and value-weighted average (based on annual book assets) in the

cash ratio and net leverage ratio of the whole sample and of the high-tech and non-high-tech sectors over the period

1974 to 2007. The sample includes U.S. firms documented on the Compustat-CRSP merged database (fundamental

annual) that have positive total assets and sales and nonnegative cash and marketable securities, and have common

shares traded on the NYSE, AMEX, or Nasdaq. Financial firms (SIC code 6000-6999) and utility firms (SIC codes

4900-4999) are excluded from the sample, leaving an unbalanced panel of 138,193 observations for 14,948 unique

firms. The high-tech and non-high-tech sectors are defined according to the U.S. Department of Commerce. The

cash-to-assets ratio (Cash/TA) is measured as cash plus marketable securities (CHE), divided by book value of total

assets (AT). Net leverage is calculated as long-term debt (DLTT) plus debt in current liabilities (DLC) minus cash

and market securities, divided by book value of total assets. Figures in Panel A (B) depict the annual mean (median)

of cash ratio and net leverage of firms in two sectors separately. Figures in Panel C are the annual value-weighted

averages (based on annual book assets) in two sectors.

80

Panel A: Annual Mean

Panel B: Annual Median

Panel C: Annual Value-Weighted Average

0.1

.2.3

.4

1974 1978 1982 1986 1990 1994 1998 2002 2006Year

Non-High-Tech High-Tech

Whole Sample

Annual Mean Cash/TA

-.3-.2

-.10

.1.2

.3

1974 1978 1982 1986 1990 1994 1998 2002 2006Year

Non-High-Tech High-Tech

Whole Sample

Annual Mean Net Leverage

0.1

.2.3

.4

1974 1978 1982 1986 1990 1994 1998 2002 2006Year

Non-High-Tech High-Tech

Whole Sample

Annual Median Cash/TA

-.3-.2

-.10

.1.2

.3

1974 1978 1982 1986 1990 1994 1998 2002 2006Year

Non-High-Tech High-Tech

Whole Sample

Annual Median Net Leverage

0.1

.2.3

.4

1974 1978 1982 1986 1990 1994 1998 2002 2006Year

Non-High-Tech High-Tech

Whole Sample

Annual VW-average Cash/TA

-.3-.2

-.10

.1.2

.3

1974 1978 1982 1986 1990 1994 1998 2002 2006Year

Non-High-Tech High-Tech

Whole Sample

Annual VW-average Net Leverage

81

Figure 2.2: Trends in Cash Holdings – R&D Adjusted Assets as the

Denominator

These figures depict the annual mean and median of the cash ratio of the whole sample and of the high-tech and

non-high-tech sectors over the period 1974 to 2007. The sample includes U.S. firms documented on the Compustat-

CRSP merged database (fundamental annual) that have positive total assets and sales and nonnegative cash and

marketable securities, and have common shares traded on the NYSE, AMEX, or Nasdaq. Financial firms (SIC code

6000-6999) and utility firms (SIC codes 4900-4999) are excluded from the sample, leaving an unbalanced panel of

138,193 observations for 14,948 unique firms. The high-tech and non-high-tech sectors are defined according to the

U.S. Department of Commerce. The cash-to-R&D adjusted assets ratio is measured as cash plus marketable

securities (CHE), divided by the R&D-adjusted book assets (the sum of book assets (TA) and R&D asset). The

R&D asset (RDA) is defined as the weighted sum of its R&D expense over the past five years assuming an annual

amortization rate of 20% (1 2 3 40.8 0.6 0.4 0.2it it it it it itRDA RD RD RD RD RD ). Figures in Panel A (B)

depict the annual mean (median) of the cash-to-R&D adjusted assets ratio for firms in two sectors separately.

Panel A: Annual Mean Panel B: Annual Median

0.1

.2.3

.4

1974 1978 1982 1986 1990 1994 1998 2002 2006Year

Non-High-Tech High-Tech

Annual Mean -- Cash/(TA+RDA)0

.1.2

.3.4

1974 1978 1982 1986 1990 1994 1998 2002 2006Year

Non-High-Tech High-Tech

Annual Median -- Cash/(TA+RDA)

82

Figure 2.3: Annual Number of IPOs

This figure plots the number of IPOs each year over the period of 1974-2007, for the whole sample and for the

high-tech and non-high-tech sectors respectively. IPO dates are identified first by using Jay Ritter‟s proprietary

database of IPO dates (http://bear.cba.ufl.edu/ritter/FoundingDates.htm). If the IPO date of a stock is unavailable

from Ritter, the first trading date on the CRSP is identified as the IPO date. The high-tech and non-high-tech

sectors are defined according to the U.S. Department of Commerce.

0

100

200

300

400

500

600

700

800

900

19

74

19

75

19

76

19

77

19

78

19

79

19

80

19

81

19

82

19

83

19

84

19

85

19

86

19

87

19

88

19

89

19

90

19

91

19

92

19

93

19

94

19

95

19

96

19

97

19

98

19

99

20

00

20

01

20

02

20

03

20

04

20

05

20

06

20

07

Non-High-Tech IPOs High-Tech IPOs All IPOs

83

Figure 2.4: Trends in Cash Holdings of Newly IPO Firms and Seasoned

Firms

This figure plots the trends in cash holdings, annual mean and median, of IPO firms and seasoned firms in the high-

tech and non-high-tech sectors. The sample includes U.S. firms documented on the Compustat-CRSP merged

database (fundamental annual) that have positive total assets and sales and nonnegative cash and marketable

securities, and have common shares traded on the NYSE, AMEX, or Nasdaq. Financial firms (SIC code 6000-6999)

and utility firms (SIC codes 4900-4999) are excluded from the sample, leaving an unbalanced panel of 138,193

observations for 14,948 unique firms. The high-tech and non-high-tech sectors are defined according to the U.S.

Department of Commerce. Observations of Newly IPO firms are those within five years after their IPO dates.

Observations of seasoned firms are the ones beginning in the sixth year after the IPO dates. The cash-to-assets ratio

(Cash/TA) is measured as cash plus marketable securities (CHE), divided by book value of total assets (AT).

Panel A: Annual Mean

Panel B: Annual Median

.05

.15

.25

.35

.45

.55

Ca

sh-t

o-A

sse

ts R

atio

1974 1978 1982 1986 1990 1994 1998 2002 2006Year

Newly IPO Firms Seasoned Firms

High-Tech Sector

.05

.15

.25

.35

.45

.55

Ca

sh-t

o-A

sse

ts R

atio

1974 1978 1982 1986 1990 1994 1998 2002 2006Year

Newly IPO Firms Seasoned Firms

Non-High-Tech Sector

.05

.15

.25

.35

.45

.55

Ca

sh-t

o-A

sse

ts R

atio

1974 1978 1982 1986 1990 1994 1998 2002 2006Year

Newly IPO Firms Seasoned Firms

High-Tech Sector

.05

.15

.25

.35

.45

.55

Ca

sh-t

o-A

sse

ts R

atio

1974 1978 1982 1986 1990 1994 1998 2002 2006Year

Newly IPO Firms Seasoned Firms

Non-High-Tech Sector

84

Appendix 2.A: Variable Definitions

Variable Definition

Cash-to-assets

(Cash/TA)

Cash-to-assets ratio (Cash/TA) is measured as cash plus marketable securities (CHE),

divided by book value of total assets (AT).

Net Leverage Net leverage is calculated as long-term debt (DLTT) plus debt in current liabilities (DLC)

minus cash and market securities (CHE), divided by book value of total assets (AT).

Industry cash flow

volatility

(IndustrySigma)

For each firm-year, I compute the standard deviation of cash flow over assets for the

previous 10 years if there are at least 3 observations. Industry sigma is calculated as the

mean of cash flow standard deviations of firms in the same industry, defined by 2-digit SIC

code.

Market-to-Book

(MB)

MB is the ratio of market value of assets to book value of assets. The market value of assets

is equal to total assets (AT) minus book value of common equity (CEQ) plus the market

value of common equity (fiscal year end price (PRCC_F) times shares outstanding

(CSHO)).

Size Size is measured with the logarithm of net assets (AT-CHE) that is converted to 2006

dollars using the Consumer Price Index.

Cash flow over assets

(CF/TA)

Cash flow is defined as operating income before depreciation (OIBDP), less interest

(XINT) and taxes (TXT).

Net working Capital

over assets

(NWC/TA)

NWC/NA is the ratio of working capital (ACT-LCT) minus cash and marketable securities

(CHE) to total assets (AT).

Capital expenditures

(Capex/TA)

Capex/TA is the ratio of Capital expenditures (CAPX) to total assets (AT).

Leverage Leverage is the ratio of long-term debt (DLTT) plus debt in current liabilities (DLC) to

total assets (AT).

R&D Expense

(R&D/Sales)

R&D/Sales is the ratio of R&D expenditure (XRD) to Sales (SALE). If R&D expenditure

(XRD) is missing, I follow the tradition to set the missing value to zero.

Dividend payer

(DivDummy)

Dividend payer dummy is set to one if a common dividend (DVC) is positive; else equal to

zero.

Acquisition

(ACQN/ TA)

ACQN/NA is the ratio of Acquisitions (AQC) to total assets (AT).

Net debt issuance

(NetDiss)

Net debt issuance is equal to long-term debt issuance (DLTIS) minus long-term debt

reduction (DLTR), scaled by total assets (AT).

Net equity issuance

(NetEiss)

Net equity issuance is equal to the sale of common and preferred stock (SSTK) minus the

purchase of common and preferred stock (PRSTKC), scaled by total assets (AT).

T-bill yield The average annual three-month rates published by the Federal Reserve. Data are from

http://research.stlouisfed.org/fred2/ .

Default spread The average yield on Baa less Aaa Moody‟s rated corporate bonds with maturity of

approximately 20-25 years. Data are from http://research.stlouisfed.org/fred2/ .

IPO date Jay Ritter‟s proprietary database of IPO dates (http://bear.cba.ufl.edu/ritter/

FoundingDates.htm) is used. If the IPO date of a stock is unavailable from Ritter, the first

trading date on the CRSP is identified as the IPO date.

All names in parentheses refer to the Compustat (XPF version, Fundamental Annual) item names.

85

Chapter 3 The Value of Cash: Industry and Temporal Effect

There is evidence that corporate cash holdings of publicly listed U.S. firms have increased

substantially over time, which is attributed by Bates et al. (2009) to changes in firm

characteristics caused by new listings. Zhou (2009) finds that the increase in cash holdings is

specific to the high-tech sector, which includes pharmaceutical, biotechnology and information

technology industries, and the increase is caused mainly by changing firm characteristics due to

the influx of newly listed firms in this sector. These findings raise an important question. If

corporate actions are the result of firms‟ value maximization behavior, then it follows that this

increase in cash holdings for an average high-tech firm should also reflect an increase in the

value of an additional dollar of cash, compared to non-high-tech firms.

In this paper, I investigate whether there has been a change in the difference in the marginal

value of cash between high-tech and non-high-tech firms over time and if so why? This study is

important since high-tech firms have become an increasingly important part of the North

American economy as the importance of manufacturing industries declines and knowledge

intensive industries take their place. Moreover, the large influx of new listings over the past few

decades has shifted the population of high-tech firms toward more immature and knowledge-

based firms that are generally more financially constrained because they lack hard assets and

strong cash flow. Hence, despite the development of capital markets, there may still exist a

stronger precautionary demand for cash and hence a higher marginal value of cash among these

firms.

To examine the difference in the marginal value of cash between high-tech and non-high-tech

firms and its potential change over time, I apply an extended version of the empirical

methodology designed by Faulkender and Wang (2006) to a large sample of publicly listed U.S.

firms for the period 1972-2007. For the full sample period, I find that the value of an additional

dollar of cash is higher for an average high-tech firm, compared to that for an average non-high-

tech firm. Of more interest is that I show that for the two sub-periods, 1972-1989 and 1990-2007,

there is a significant rise in the difference between these two sectors: cash is valued much more

86

highly for high-tech firms than for non-high-tech firms in the later period. This finding is robust

to alternative definitions of the high-tech sector, divisions of sub-periods, as well as for how the

marginal value of cash is estimated.

I then investigate two potential explanations for this temporal change: fundamental explanation

and misvaluation explanation. According to the fundamental explanation, shareholders tend to

put higher value on the cash held by high-tech firms since these firms in general have stronger

precautionary demand for cash due to their firm characteristics and investment opportunities,

and the rapid growth of this sector due to new listings as documented in Zhou (2009) may imply

an increasing difference between high-tech and non-high-tech sectors. To examine this

explanation, I compare several empirical proxies for the precautionary demand of cash. I find

that on average the characteristics of high-tech firms, compared to non-high-tech firms, have

shifted more toward those with a stronger precautionary demand for cash. This is support for a

widening difference in the marginal value of cash across the two sectors in the second period.

On the other hand, the characteristics of high-tech firms also imply that they are more

speculative and harder-to-arbitrage and hence are more likely to be misvalued by shareholders.

The impact of misvaluation may be stronger in the second sub-period due to the change in

population characteristics driven by the influx of new listings. To investigate this alternative

explanation based on misvaluation, I compare the difference in the marginal value of cash

holdings across periods with large and small changes in market sentiment as measured by Baker

and Wurgler (2007). The results show that although misvaluation may have an impact, it cannot

fully explain the phenomenon.

This paper contributes to the literature from the following aspects. First, it contributes to our

understanding of corporate cash policy. Along with the growing difference in the level of cash

holdings by high-tech and non-high-tech firms, there is a pronounced difference in the marginal

value of cash across these two sectors and over time. This is consistent with Keynes‟

precautionary motive for holding cash; the significant changes that have occurred in the

industrial structure of the North American economy; and the increased ability of immature firms

to go public due to the development of capital markets. Furthermore, Brown and Petersen (2010)

87

find some empirical evidence that young firms with positive R&D expenses use their cash

holdings to smooth their R&D investment, and that this practice has become more common over

time. Given that R&D investment is usually much more intensive among high-tech firms, my

paper tests the value implication of these changes in investment policy.

The remainder of the paper is organized as follows. Section 3.1 reviews the related literature.

Section 3.2 describes the empirical methodology. Section 3.3 describes the sample used in this

paper. Empirical findings are presented in Section 3.4. Section 3.5 discusses the potential

explanations, and Section 3.6 concludes.

3.1 Related Literature

In a frictionless world, holding cash is irrelevant for corporate valuation because a firm can

finance any value-increasing investment opportunity with external financing if it lacks internal

funds (Modigliani and Miller, 1958). Hence, the market value of one dollar should be exactly

one dollar. However, the existence of market imperfections implies that a dollar of cash might

be worth more than a dollar for a firm with valuable investment opportunities and volatile cash

flow, since it can help avoid the adverse effects of underinvestment due to capital rationing.

This argument for the precautionary demand for cash can be traced back to Keynes (1936).

Nevertheless, only recently has research quantified the marginal value of corporate cash

holdings and examined its variation across firms. Applying a revised event study method that

links the cross-sectional variation in the changes in market valuation of a firm with the changes

in its financial characteristics, Faulkender and Wang (2006) find that the marginal value of cash

is higher for firms with lower liquidity, lower leverage, and higher external financial constraints.

Using a value-based model designed by Fama and French (1998), Pinkowitz and Williamson

(2007) show that firms with more volatile cash flow and more promising growth opportunities

generally have higher value for their cash holdings. Further, they find that computer software

and pharmaceutical firms have the highest marginal value of cash among the Fama-French 49

88

industries.35

As further evidence of the precautionary motive, Denis and Sibilkov (2009) find

empirical support for the argument that cash is more valuable for financially constrained firms

because it allows these firms to take value-increasing projects they might otherwise forgo.

Moreover, the positive impact of cash on investment, and hence valuation, is particularly strong

for constrained firms that appear to have valuable investment opportunities but low levels of

internal funds, i.e. ones with high hedging needs as defined by Acharya, et al. (2007).

Prior research focuses on cross-sectional variation in the value of cash holdings. To date no

studies have investigated whether the value of cash changes over time and whether the value

changes differently for different firms. This is particularly important since the industrial

composition of publicly listed U.S. firms has changed over time due to the influx of new listings.

Several recent studies have documented the impact of this change on corporate financial policies,

such as the disappearing dividend (Fama and French, 2001) and the dwindling investment to

cash flow sensitivity (Brown and Petersen, 2009) phenomena. Closely related to this paper,

Zhou (2009) finds that high-tech firms on average have increased their corporate cash holdings

since the 1980s, while non-high-tech firms did not, at least until the early 2000s.36

This

difference in cash trends can be attributed to differences in changing population characteristics

driven by the influx of new listings in both sectors. More specifically, new listings in the high-

tech sector tend to have R&D-based growth opportunities but low (even negative) and volatile

internal cash flows, and rely heavily on external equity financing, which is not always

available.37

Hence, they tend to fall into the category of financially constrained firms with high

hedging needs as described by Acharya et al. (2007) and Denis and Sibilkov (2009). As a result,

the population of high-tech firms has shifted more towards those firms with a high precautionary

demand for cash. If shareholders are aware of this secular shift and believe that firms

35

These valuation methodologies have also been applied to investigate the impact of agency problems on the value

of cash (Dittmar and Mahrt-Smith, 2007) and the impact of corporate diversification (Tong, 2010).

36 More specifically, the cash ratio of an average high-tech firm has more than tripled in the period since 1980,

whereas that for the average non-high-tech firm did not see any increase until the early 2000s; even then the

increase is not of a similar order of magnitude.

37 In an article in the Harvard Business Review, Richard Passov, the Treasurer of Pfizer, mentioned that “History

has shown, however, that in times of need, external financing can be exorbitantly expensive or simply unavailable”

and gave the funding crisis experienced by Intel as an example (Passov, 2003, p.120).

89

accumulate cash so as to maximize shareholders‟ value, it is expected that the marginal value of

cash should have increased much more over time for high-tech than for non-high-tech firms.38

Hence, the difference in the marginal value of cash between these two sectors should be larger

in the latter sub-period.

On the other hand, the marginal value of cash holdings, inferred from the market valuation of

equity, may suffer from shareholders‟ misvalution. This issue is of special concern in my study

since firms from these two sectors are exposed differently to the impact of market misvaluation.

According to De Long et al. (1990) and Shleifer and Vishny (1997), market misvaluation is the

result of the sentiment-based demand of noise traders and the limits to arbitrage by rational

investors. This implies that those more speculative and harder-to-arbitrage firms should be more

exposed to market misvaluation (Baker and Wurgler, 2006, 2007). Overall, high-tech firms are

more likely to belong to this category due to the opacity and uncertainty of their R&D

investments. Moreover, this concern may become stronger in the latter sub-period due to the

changing population characteristics of high-tech sector, driven by the large influx of new listing.

Overall, these may cause the biased estimate of the marginal value of cash and potentially

contribute to the increase in the difference across two sectors over time.

3.2 Empirical Methodology

To estimate the difference in the marginal value of cash across high-tech and non-high-tech

firms, I build on the methodology proposed by Faulkender and Wang (2006) with a slight

modification.39

38

From a different but related perspective, Brown and Petersen (2010) find evidence that young firms tend to use

their cash reserves to smooth R&D investment, a practice which has become more prevalent over time. This also

indicates that, compared to non-high-tech firms, cash might have become more and more valuable for high-tech

firms due to its direct impact on their investment policy.

39 An alternative method to estimate the marginal value of cash is based on Fama and French (1998) methodology,

which uses the market-to-book ratio as the dependent variable. However, this method does not control for the

variation in risk factors over time and the variation in exposures to those factors across firms, which affect firms‟

discount rates. Moreover, the market-to-book ratio may be biased by the accounting treatment for the book value of

assets (Faulkender and Wang, 2006).

90

, , , , , ,

, , 0 1 2 3 4 5 6

, 1 , 1 , 1 , 1 , 1 , 1

, 1 , , 1 , ,

7 8 , 9 10 11 ,

, 1 , 1 , 1 , 1 , 1

,

12

* *

*

i t i t i t i t i t i tB

i t i t

i t i t i t i t i t i t

i t i t i t i t i t

i t i t

i t i t i t i t i t

i t

C E NA RD I Dr R

M M M M M M

C NF C C CL L

M M M M M

CHTDummy

13 ,

, 1

i t

i t

HTDummyM

(3.1)

This methodology is essentially a long-run event study, investigating the impact of the

unexpected change in corporate cash holdings on a firm‟s valuation over the event window of a

fiscal year. The dependent variable is the excess return for firm i over fiscal year t, computed as

stock i‟s return minus the return of stock i‟s benchmark portfolio over fiscal year t. The

benchmark portfolios are the twenty-five Fama and French portfolios, formed by an independent

sort of stocks on their size and book-to-market characteristics. The benchmark returns are used

to control for the variation in the discount rate due to its sensitivity to time-varying risk factors.

Besides the change in cash holdings ( ,i tC ), the independent variables also include the changes in

the firm‟s profitability (measured by earnings before extraordinary items ( ,i tE )),changes in the

firm‟s investment policy (including net assets ( ,i tNA ) and R&D expenditures ( ,i tRD )), and

changes in the firm‟s financial policy (including interest expense ( ,i tI ), dividends ( ,i tD ),

leverage ( ,i tL ), and net financing ( ,i tNF )). Details on how these variables are constructed are

provided in Appendix 3.A. Except for leverage, all the independent variables are scaled by the

one-year lagged market value of equity ( , 1i tM ). Since stock i‟s return can be rewritten as the

change in the market valuation of equity divided by the lagged market value

( , , 1 , 1( ) /i t i t i tM M M ), the coefficient estimates can be interpreted directly as the value that

shareholders place on a one-dollar change in the related independent variables.

Besides the direct valuation of cash holdings given by shareholders ( 1 ), Faulkender and Wang

(2006) also include the interaction terms of the change in cash with the lagged cash position and

91

the level of leverage, since the marginal value of cash to shareholders may decrease with the

firm‟s cash position and the level of leverage ( 10 0 , 11 0 ).

To investigate the difference in the marginal value of cash between high-tech and non-high-tech

firms, I follow the logic of Dittmar and Mahrt-Smith (2007) to slightly modify this empirical

model by including an indicator variable (HTDummy) equal to one for high-tech firms and zero

for non-high-tech firms, as well as an interaction term between the change in cash and this

indicator variable. The coefficient estimate on this interaction term ( 12 ) directly represents the

difference in the marginal value of cash between high-tech and non-high-tech firms. HTDummy

itself is also included in the regression to ensure that the estimated coefficient on this newly-

added interaction term is not due to the property of being a high-tech firm itself.40

3.3 Sample and Summary Statistics

The base sample contains all U.S. publicly traded firms in the CRSP-Compustat merged

database (Fundamental Annual) for the period 1972 to 2007.41

Financial firms (SIC codes 6000-

6999) are excluded since they need to hold cash and marketable securities to meet their statutory

capital requirements. I also exclude utilities (SIC codes 4900-4999) as their cash policy can be a

by-product of regulation. Firm-year observations are deleted if there is a missing value for any

variable in equation (3.1). To reduce the effect of outliers, all the variables are winsorized at the

one percent tails using the whole sample. The screening leaves an unbalanced panel of 95,880

observations for 12,057 unique firms during the period from 1972 to 2007.

I follow Brown et al. (2009) to use the official definition of high-tech industries offered by the

United States Department of Commerce. More specifically, the high-tech sector consists of

firms from the following seven industries defined by 3-digit SIC codes: drugs (SIC 283), office

and computing equipment (SIC 357), communications equipment (SIC 366), electronic

components (SIC 367), scientific instruments (SIC 382), medical instruments (SIC 384), and

40

This logic has also been applied to compare the value of cash for constrained vs. unconstrained firms (Denis and

Sibilkov, 2009), diversified vs. single-segment firms (Tong, 2010).

41 My sample starts at 1972 so as to be comparable to Faulkender and Wang (2006).

92

software (SIC 737). The remaining firms in the sample are classified as non-high-tech. The

results are robust to alternative definitions, such as the Fama-French industry classification and

the Global Industry Classification Standard (GICS).

Table 3.1 provides the distribution of firm-year observations (Panel A), as well as the summary

statistics for the pooled sample, high-tech firms, and non-high-tech firms respectively for the

full period and two sub-periods of equal length (Panel B). It is clear that high-tech firms have

become increasingly important in the universe of publicly listed firms over the sample period;

from about 10% in the early 1970s, they have increased to comprise over 30% of the sample in

the 2000s. Although the summary statistics of the whole sample are similar to those reported by

Faulkender and Wang (2006), there are some differences between high-tech and non-high-tech

firms over time. Compared to non-high-tech firms, high-tech firms tend to have lower leverage

and hold more cash. While leverage has decreased over time in both sectors, cash holdings have

increased in the high-tech sector but decreased in the non-high-tech sector.

3.4 Empirical Results

Table 3.2 presents the primary results from estimating equation (3.1) for high-tech and non-

high-tech firms for the different time periods. Coefficient estimates are reported in Panel A.

Panel B reports the marginal value of cash for the average firm of each subgroup, as the sum of

the coefficient on the change in cash and the respective coefficients on the cash interactions

times the sample means of these interaction variables.

I first examine the marginal value of cash in the whole sample and report the results in column

[1]. Taking away the high-tech dummy and its interaction with the change in cash holding, the

regression model is that of Faulkender and Wang (2006). Coefficient estimates in column [1]

are close to their findings. Especially, the marginal value of cash decreases significantly when a

firm‟s cash position or leverage level increases. Ignoring the potential difference between high-

tech and non-high-tech firms, shareholders put a value of $1.03 (=$1.564 + (-$0.746*0.1790) +

(-$1.426*0.279)) on an extra dollar of cash held by an average firm during the period from 1972

to 2007. This is slightly higher than Faulkender and Wang‟s finding for the period from 1972 to

93

2001.42

However, it confirms that across all firms, not accounting for their individual

characteristics, a dollar of cash is close to being worth a dollar so that financial constraints in

aggregate and across long periods of time are of only moderate importance.

Column [2] investigates the impact of affiliation with the high-tech sector over the full sample

period by incorporating the high-tech dummy and its interaction with the change in cash

holdings. The coefficient estimate of the interaction term (0.398) is positive and statistically

significant, indicating that shareholders place a much higher value on an additional dollar of

cash held by a high-tech as compared to that held by a non-high-tech firm.43

The marginal value

of one dollar cash held by an average non-high-tech firm is about $0.93, while it is $1.33 for an

average high-tech firm.

An unanswered question in the literature is whether this difference in valuation has changed

over time. This is important since recent empirical studies find that the level of corporate cash

holdings has increased over the past three decades; a phenomenon that Zhou (2009) shows is

specific to the high-tech sector until the early 2000s. It is unclear how shareholders evaluate this

increasing difference in the level of cash. To investigate this, I divide the full sample period into

two sub-periods of equal length – namely, 1972-1989 and 1990-2007, and estimate equation

(3.1) for each sub-period. The coefficient estimates for the interaction term of the HTDummy

with the change in cash holdings from Column [3] and [4] show that the difference in the

marginal value of cash between high-tech and non-high-tech firms has widened over time. More

specifically, the marginal value of cash for an average high-tech firm was $0.209 higher than

that of an average non-high-tech firm in the first sub-period, but the difference has increased to

$0.417 during the second sub-period. This increase is statistically significant at 5% level. F-test

rejects the hypothesis that all coefficients are jointly equal between two sub-periods. Panel B

shows that the marginal value of cash has increased for both sectors over time: from $0.91 to

$1.60 for the high-tech sector and from $0.70 to $1.19 for the non-high-tech sector.

42

I replicate the main findings of Faulkender and Wang (2006), the details of which are presented in Appendix 3.B.

43 This is consistent with Pinkowitz and Williamson (2007), who find the value of cash in the computer software

and pharmaceuticals industries is higher than that of other industries as defined by the Fama-French 48 industry

classification.

94

To check whether this increasing difference in the marginal value of cash is not driven by the

specifics of industry definitions and sub-periods, Table 3.3 conducts several robustness checks.

First, I split the sample into high-tech and non-high-tech sectors according to two alternative

methods – the Fama-French industry classification and the Global Industry Classification

Standard (GICS), and calculate the difference in the marginal value of cash between the two

sectors for the same two sub-periods.44

The results documented in Panel A are consistent with

those in Table 3.2. According to the Fama-French 12 industry classification, the difference has

increased from $ 0.251 to $0.383 across the two sub-periods. For the GICS classification, the

increase is from $0.207 to $0.365. The increase is statistically significant at 10% level in both

settings. Under all three classification schemes, the marginal value of cash for high-tech firms

on average is significantly higher than for non-high-tech firms, and this difference has increased

significantly over time. Panel B of Table 3.3 documents the difference in the marginal value of

cash between the two sectors for alternative definitions for the sub-periods. I split the full

sample period into four sub-periods, each with a nine-year window, and estimate the regression

model in each sub-period. By and large, it supports previous findings, since the difference in the

marginal value of cash is much larger in the latter two sub-periods, which cover the 1990s and

2000s, as compared to the two earlier sub-periods that cover the 1970s and 1980s.

In regression equation (3.1), the coefficient of the interaction term between HTDummy and the

change in cash holdings directly measures the difference in the marginal value of cash between

high-tech and non-high-tech firms. One potential concern is that this method implicitly assumes

that firms from both sectors have the same sensitivity to all the other firm-specific factors, and

on average have the same cash level and leverage. As an additional test, I apply separate

regressions for high-tech and non-high-tech firms for the full sample period and the two sub-

periods respectively, explicitly addressing the differences in coefficients and incorporating

44

Ken French‟s website (http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/datalibrary.html) provides the

definition of the Fama-French 12 industries. According to this classification, Business Equipment (BusEq) and

Healthcare (Hlth) are regarded as high-tech. The Global Industry Classification Standard (GICS), developed by

Standard & Poor‟s (S&P) and Morgan Stanley Capital International (MSCI), categorizes a firm according to its

operational characteristics as well as investors‟ perceptions of its principal business activity. Information

Technology (including software, hardware, and electronic equipment) and Health Care (including health care

equipment and services, pharmaceuticals, biotech and life sciences) by the GICS scheme are regarded as high-tech.

95

different sample means for lagged cash and leverage across different subsamples. The

coefficient estimates are reported in Panel A of Table 3.4 and the marginal values of cash are

presented in Panel B.

The first two columns of Table 3.4 show that in the full sample period, for an average high-tech

firm the marginal value of cash is $1.58 (=$2.059 + (-$1.117*0.191) + (-$1.777*0.151)), while

for non-high-tech firms it is $0.84 (=$1.290 + (-$0.615*0.175) + (-$1.064 *0.319)). Using the

same methodology for the sub-periods, I find the marginal value of cash has increased for both

high-tech and non-high-tech firms, from $1.06 to $1.82 for the high-tech sector and $0.66 to

$1.06 for the non-high-tech sector. This indicates that the difference in the marginal value of

cash between these two sectors has almost doubled over these two periods, increasing from

$0.40 for 1972-1989 to $0.76 for 1990-2007. This supports the main findings in Table 3.2.

To check whether the results are not due to short run momentum effects, I also use the returns of

125 portfolios developed in Daniel, Grinblatt, Titman, and Wermers (1997, DGTW hereafter) as

benchmark returns. The DGTW portfolios are based on a triple-sort on each firm‟s size, book-

to-market, and momentum.45

The results, reported in Table 3.5, are qualitatively similar to the

earlier results whereby the coefficient on the interaction term between the high-tech dummy and

the change in cash has increased from 0.280 to 0.448 over the two sub-periods.

As a further robustness check I estimate equation (3.1) for three-year rolling windows. The

coefficient estimates for the interaction term between the high-tech dummy and the change in

corporate cash holdings are then plotted in Figure 3.1.46

Despite the variation in the estimates,

the difference in the value of cash between high-tech and non-high-tech firms was generally

lower in the 1970s and 1980s before peaking in the early 1990s and again in 2000 and then

dropping off. Noticeably these peaks broadly coincide with the biotechnology bubble in the

early 1990s and the internet bubble of the late 1990s respectively. The large influx of young

45

Stock assignments to the 125 DGTW benchmark portfolios and the monthly returns for each DGTW portfolio are

kindly provided by Russ Wermer (http://www.smith.umd.edu/faculty/rwermers/ftpsite/Dgtw/coverpage.htm). Data

are available from 1975.

46 The window length used here is three years, so as to smooth some fluctuations that may evolve in annual cross-

sectional regressions and avoid over-smoothing with long window.

96

firms with high growth opportunities around these periods may indicate that holding cash is a

value-increasing decision. However, this also may be consistent with a “mispricing” explanation,

since high-tech firms may have been overvalued at these times. I will discuss these two

alternative explanations in the next section.

In sum, this section shows that the difference in the marginal value of cash holdings between

high-tech and non-high-tech firms has become larger in the 1990s and 2000s. The results are

robust to alternative industry classifications schemes, division of sub-periods, and different

methods to calculate the marginal value of cash.

3.5 Explanations

This paper tests two explanations for why the difference in the marginal value of cash holdings

between high-tech firms and non-high-tech firms has increased. The fundamental explanation

suggests that the population characteristics of these two sectors have changed towards the types

of firms with a higher precautionary demand for cash. The second explanation is based on

mispricing argument, that is, high-tech firms are more likely to be misvalued than non-high-tech

firms and this misvaluation effect may be stronger during the second part of the sample period.

In this section, I will test which explanation plays a more important role to help understand my

empirical findings.

3.5.1 Fundamental Explanation

According to the precautionary motive for holding cash, accumulating an additional dollar of

cash is a value-increasing decision for a firm if this dollar can be used to finance a project that

may otherwise be forgone. Consistent with this idea, previous studies have shown that cash is

more valuable for firms with more growth opportunities, more unstable cash flows, and are

more financially constrained (Faulkender and Wang, 2006; Pinkowitz and Williamson, 2007).

Fama and French (2004) find that new listings in the 1980s and 1990s tend to have more growth

opportunities and are less profitable. Zhou (2009) documents that the characteristics of new

listings vary across the high-tech and non-high-tech sectors, which contributes to the difference

97

in observed cash trends between these two sectors since the 1980s. These differences in

changing population characteristics across high-tech and non-high-tech sectors over time, due to

the influx of newly listed firms, may also contribute to the widening difference in the marginal

value of cash. To test this hypothesis, I compare widely-used empirical proxies for the

precautionary demand for cash for both high-tech and non-high-tech firms during the two sub-

periods; these include the type and amount of growth opportunities, the level and stability of

cash flows, and the access to external financing. The mean and median of each subsample are

reported in Table 3.6. Tests for equal means and distributions are reported by t- and z-statistics

respectively.

R&D Expenditure (R&D/TA), measured as research and development expenditures scaled by

book assets, has been used as a proxy for growth opportunities in the cash literature (Opler et al.,

1999; Pinkowitz and Williamson, 2007). Bates et al. (2009) show that increasing R&D intensity

contributes to the upward cash trend among U.S. publicly listed firms. Based on the analysis of

variance decomposition, Zhou (2009) shows that R&D intensity is one of the key determinants

of the level of corporate cash holdings, especially for high-tech firms. In a related study, Brown

and Petersen (2010) find that firms tend to use their cash reserves to smooth their R&D

investment. Moreover, this action is more substantial for firms facing financial frictions and it

has become more significant over time. These imply that cash holdings should be more valuable

for a firm with more intensive R&D investment. Table 3.6 shows that the average R&D

expenditure has increased significantly from 7.75% to 12.4% in the high-tech sector, while it

only increased slightly from 1.02% to 1.31% in the non-high-tech sector.

Another commonly used proxy for growth opportunities is Capital Expenditures (CapEx/TA).

Table 3.6 shows that both sectors have decreased their capital expenditures over time. More

importantly, while capital expenditures always dominated R&D expenses for non-high-tech

firms, they have become subordinated to R&D expenses for an average high-tech firm during

the second sub-period (12.4% for the mean R&D/TA versus 4.8% for the mean CapEx/TA).

Compared to capital expenditures, R&D expenses are a form of investment that is subject to

more information asymmetries since its output tends to be more uncertain and usually cannot be

used as collateral (Hall, 2002). Hence, this increase in the relative importance of R&D-based

98

growth in the high-tech sector provides supporting evidence that the precautionary demand for

holding cash has become stronger for high-tech firms over time.

Industry cash flow volatility (Industry Sigma) measures the stability of a firm‟s cash flow and is

usually used as a proxy for the business risk determinant of the precautionary demand for cash

(Opler et al., 1999; Han and Qiu, 2007). Table 3.6 shows that the sample average industry cash

flow volatility has increased from 4.41% to 8.6% for non-high-tech firms and 6.35% to 13.92%

for high-tech firms over these two sub-periods. It is clear that business risk has increased

significantly in both sectors, but the economic scale of the increase is more pronounced for

high-tech firms and hence the difference between these two sectors has widened over time. This

is also consistent with a much stronger increase in the precautionary motive to hold cash for

high-tech firms, compared to non-high-tech firms.

Cash Flows (CF/TA), net equity issuance (NetEiss), and net debt issuance (NetDiss) are three

key sources of cash. Compared to non-high-tech firms, high-tech firms on average have become

less profitable, less dependent on debt financing, and more reliant on equity financing over time.

More specifically, during the second sub-period, the average ratio of cash flows-to-assets for

high-tech firms was as low as -6.89%, while the average ratio of net equity issuance-to-assets

was 7.03% and the average ratio of net debt issuance-to-assets was only 0.82%. All of these

factors reflect the growing immaturity of newly listed high-tech firms, which results in the

changing population characteristics. McLean (2010) finds that over the past three decades firms

have increasingly saved share issuance proceeds as cash and this is driven by the precautionary

motive. The lack of stable and sufficient internal funds and the increasing dependence on

volatile external financing sources indicates that holding cash is of higher value for high-tech

firms than for non-high-tech firms over time.

In sum, it is clear from this analysis that the population of high-tech firms has shifted more

towards those with a stronger precautionary demand for cash. Hence, holding an additional

dollar of cash has become more valuable for high-tech firms, as compared to earlier periods as

well as those contemporaneous firms in the non-high-tech sector. In support of this argument,

99

the last row of Table 3.6 shows that the level of corporate cash holdings, measured by the ratio

of cash-to-assets, has increased substantially for high-tech firms over time.

3.5.2 Potential Impact of Mispricing

By using the excess return as the dependent variable, regression model (3.1) controls for „both

the time-series variation in risk factors and the cross-sectional variation in exposures to those

factors‟ (Faulkender and Wang, 2006, p.1966). However, the excess return may still contain the

impact of market misvaluation. This may lead to a biased estimate of the marginal value of cash

and partly contribute to the larger difference between two sectors in the second sub-period.

One way to test the mispricing hypothesis is based on the impact of investor sentiment on stock

valuation. According to De Long et al. (1990) and Shleifer and Vishny (1997), the sentiment-

based demand of noise traders combined with limits on the arbitrage activity of rational

investors leads to the existence of mispricing in capital markets. Hence, overvaluation tends to

increase with investor sentiment and harder-to-arbitrage stocks are more sensitive to

sentiment.47

Consistent with these theoretical implications, Baker and Wurgler (2007) construct

a monthly investor sentiment index and report that, after controlling for market returns, changes

in this index are positively correlated with the contemporaneous returns of more speculative and

harder to arbitrage stocks, but are slightly negatively related to the returns of bond-like stocks.48

Compared to non-high-tech firms, high-tech firms tend to be more speculative and harder-to-

arbitrage due to their reliance on R&D-based growth opportunities, smaller size, and lack of

sufficient and stable internal cash flows. Table 3.6 shows that these differences in firm

characteristics between two sectors have become stronger in the second sub-period, as a result

47

More speculative and harder-to-arbitrage stocks usually include the stocks of low capitalization, unprofitable,

highly volatile, younger, non-dividend paying, and growth firms.

48 Baker and Wurgler‟s sentiment index is based on six widely accepted measures of investor sentiment in the

literature, including NYSE turnover, the dividend premium, the closed-end fund discount, the number of IPOs, the

first-day returns on IPOs, and the equity share in new issues. They construct the composite sentiment index as the

first principal component of the parts of these six proxies that are orthogonal to the fundamentals. The annual index

was used in Baker and Wurgler (2006), while the monthly index was used in Baker and Wurgler (2007). Both data

are available at Jeffrey Wurgler‟s website: www.stern.nyu.edu/~jwurgler. The negative coefficient on bond-like

stocks can be explained as the result of substantial sentiment-driven switch in demand for these two types of stocks.

100

of new listings. If mispricing explains the results that the value of cash has increased more for

high-tech firms in the second subperiod, then I expect to see that the impact of sentiment on

high-tech firms maybe even stronger in the latter sub-period. To test this, I first examine the

impact of sentiment on stock valuation in general and then check its potential influence on the

difference in the marginal value of cash.

For each firm-year observation, I identify Baker and Wurgler‟s monthly sentiment data for the

beginning and end of the fiscal year, which is supposed to affect the market valuation of equity

at these two points in time. The difference between these two sentiment levels (ΔBWSI)

measures the change in sentiment over the fiscal year. I first test whether and how the change in

sentiment affects contemporaneous returns, by adding the change in sentiment (ΔBWSI) into

Faulkender and Wang‟s original model. Table 3.7 shows that coefficient estimates on ΔBWSI

are always positive for high-tech firms and negative for non-high-tech firms. This finding is

consistent with Baker and Wurgler (2007) since high-tech stocks in general are more speculative

and harder-to-arbitrage than non-high-tech stocks. Moreover, although the coefficient estimates

of ΔBWSI are only statistically significant at 10% level during the first sub-period for two

sectors respectively, their statistical significance has increased to 1% level during the second

sub-period. This shows that the impact of the changes in investor sentiment on stock returns is

stronger for both sectors during the second sub-period, which may contribute to the increasing

difference in the marginal value of cash between these two sectors.49

To examine this, I sort the sample into periods with large changes in sentiment (Large |ΔBWSI|)

and periods with small changes in sentiment (Small |ΔBWSI|) as the two extreme quartiles and

two middle quartiles of the distribution of annual changes in sentiment. If investor sentiment

indeed affects the difference in the marginal value of cash between high-tech and non-high-tech

firms, its impact should be stronger in the Large |ΔBWSI| periods due to substantial sentiment

fluctuations, i.e. the coefficient on HTDummy*ΔCash/ME in regression equation (3.1) is

expected to be significantly higher in Large |ΔBWSI| periods, compared to in Small |ΔBWSI|

49

The t-test shows that the change over two sub-periods is statistically significant only for non-high-tech firms. For

high-tech firms, even though the coefficient estimate for ΔBWSI has almost doubled over time, the t-test shows that

it is not statistically significant.

101

periods. However, the results in Pane B of Table 3.8 show that during both sub-periods the

coefficients on this interaction term are not statistically different between the Large |ΔBWSI|

periods and the Small |ΔBWSI| periods.

In sum, according to these tests based on investor sentiment, the larger difference in the

marginal value of cash between high-tech and non-high-tech firms in the second sub-period

cannot be attributed solely to market misvaluation.

3.6 Conclusion

This paper investigates how the difference in the marginal value of cash between high-tech and

non-high-tech firms has changed over time. I find that on average the difference has become

larger in the sub-period that covers the 1990s and 2000s, compared to that of the 1970s and

1980s. This result is robust to alternative definitions of the high-tech sector, division of the sub-

periods, and the method of calculating the marginal value of cash. Further investigation shows

that this increasing difference in the marginal value of cash between these two sectors can be

explained by changing firm characteristics related to the precautionary demand for holding cash.

The dominant implication is that more immature high-tech firms have entered the capital

markets and have exactly the characteristics of firms with a strong precautionary demand for

cash due to the existence of financial constraints and highly uncertain growth opportunities. The

value of these higher cash holdings is then reflected in their stock prices. For high-tech firms,

this precautionary demand for cash has become stronger over time. Misvaluation due to

fluctuations in investor sentiment and the biotech and internet bubble periods may contribute to

this result, but cannot be the main cause.

102

Table 3.1: Sample Distribution and Descriptive Statistics

Panel A reports the number of firms in the whole sample and in the high-tech and non-high-tech sectors separately

each year over the period of 1972-2007. Tables in Panel B present descriptive statistics of firm characteristics for

all firms, high-tech firms, and non-high-tech firms in the full sample period (Panel B.1) and two sub-periods (Panel

B.2) respectively. Data are from the CRSP-Compustat merged database (Fundamental Annual) for the period of

1972–2007. I exclude companies from the financial (SIC 6000–6999) and utility (SIC 4900–4999) industries. I also

exclude all observations with missing data. The details on variable definitions are in Appendix 3.A. The high-tech

and non-high-tech sectors are defined according to the U.S. Department of Commerce. High-tech sector contains

firms from seven industries defined by 3-digit SIC codes 283, 357, 366, 367, 382, 384, and 737. The remaining

firms in the sample are classified as non-high-tech. The proportions (in percentage) of the high-tech and non-high-

tech sectors in annual sample are reported respectively.

103

Panel A: The Distribution of Firms

Whole Sample High-Tech Non-High-Tech

Year Number Number Percent Number Percent

1972 1095 103 9.41% 992 90.59%

1973 2117 212 10.01% 1905 89.99%

1974 2529 288 11.39% 2241 88.61%

1975 2605 312 11.98% 2293 88.02%

1976 2577 309 11.99% 2268 88.01%

1977 2484 289 11.63% 2195 88.37%

1978 2384 303 12.71% 2081 87.29%

1979 2431 322 13.25% 2109 86.75%

1980 2535 373 14.71% 2162 85.29%

1981 2527 387 15.31% 2140 84.69%

1982 2657 453 17.05% 2204 82.95%

1983 2682 500 18.64% 2182 81.36%

1984 2769 601 21.70% 2168 78.30%

1985 2742 648 23.63% 2094 76.37%

1986 2644 630 23.83% 2014 76.17%

1987 2732 647 23.68% 2085 76.32%

1988 2900 735 25.34% 2165 74.66%

1989 2825 729 25.81% 2096 74.19%

1990 2815 715 25.40% 2100 74.60%

1991 2852 737 25.84% 2115 74.16%

1992 2900 760 26.21% 2140 73.79%

1993 3082 847 27.48% 2235 72.52%

1994 3315 928 27.99% 2387 72.01%

1995 3392 906 26.71% 2486 73.29%

1996 3382 938 27.74% 2444 72.26%

1997 3395 979 28.84% 2416 71.16%

1998 3261 967 29.65% 2294 70.35%

1999 3017 879 29.13% 2138 70.87%

2000 2832 845 29.84% 1987 70.16%

2001 2622 857 32.68% 1765 67.32%

2002 2450 834 34.04% 1616 65.96%

2003 2339 758 32.41% 1581 67.59%

2004 2265 724 31.96% 1541 68.04%

2005 2277 758 33.29% 1519 66.71%

2006 2267 715 31.54% 1552 68.46%

2007 2182 688 31.53% 1494 68.47%

Total 95880 22676 23.65% 73204 76.35%

104

Panel B: Descriptive Statistics

Panel B.1. Full Sample Period [1972, 2007]

Total High-Tech Non-High-Tech

Mean Median Mean Median Mean Median

, ,

B

i t i tr R -0.0059 -0.0863 0.0331 -0.0981 -0.0180 -0.0833

ΔCash/ME 0.0022 -0.0004 -0.0003 -0.0008 0.0030 -0.0003

Lagged Cash/ME 0.1790 0.0956 0.1909 0.1132 0.1753 0.0903

Leverage 0.2794 0.2226 0.1508 0.0692 0.3192 0.2774

ΔEarnings/ME 0.0130 0.0064 0.0200 0.0052 0.0108 0.0068

ΔNet Assets/ME -0.0057 0.0220 0.0000 0.0166 -0.0075 0.0249

ΔR&D/ME 0.0001 0.0000 0.0007 0.0014 -0.0001 0.0000

ΔInterest/ME 0.0002 0.0000 -0.0003 0.0000 0.0003 0.0000

ΔDividends/ME -0.0002 0.0000 0.0000 0.0000 -0.0002 0.0000

Net Finance/ME 0.0476 0.0011 0.0522 0.0043 0.0461 0.0000

Obs 95880 22676 73204

105

Panel B.2. Two Sub-Periods

First Sub-Period [1972, 1989]

Total High-Tech Non-High-Tech

Mean Median Mean Median Mean Median

, ,

B

i t i tr R -0.0041 -0.0694 0.0128 -0.0787 -0.0077 -0.0677

ΔCash/ME 0.0035 -0.0012 0.0025 -0.0012 0.0037 -0.0012

Lagged Cash/ME 0.1905 0.1097 0.1516 0.0888 0.1987 0.1142

Leverage 0.3278 0.2925 0.2111 0.1468 0.3523 0.3274

ΔEarnings/ME 0.0137 0.0086 0.0165 0.0077 0.0131 0.0088

ΔNet Assets/ME -0.0281 0.0253 0.0019 0.0309 -0.0344 0.0233

ΔR&D/ME 0.0004 0.0000 0.0026 0.0018 -0.0001 0.0000

ΔInterest/ME 0.0007 0.0003 0.0004 0.0001 0.0008 0.0004

ΔDividends/ME 0.0001 0.0000 0.0003 0.0000 0.0001 0.0000

Net Finance/ME 0.0472 0.0005 0.0484 0.0037 0.0470 0.0000

Obs 45235 7841 37394

Second Sub-Period [1990, 2007]

Total High-Tech Non-High-Tech

Mean Median Mean Median Mean Median

, ,

B

i t i tr R -0.0075 -0.1034 0.0439 -0.1097 -0.0289 -0.1008

ΔCash/ME 0.0011 0.0000 -0.0017 -0.0006 0.0023 0.0000

Lagged Cash/ME 0.1687 0.0818 0.2117 0.1295 0.1509 0.0653

Leverage 0.2362 0.1629 0.1189 0.0368 0.2847 0.2286

ΔEarnings/ME 0.0124 0.0048 0.0219 0.0040 0.0085 0.0051

ΔNet Assets/ME 0.0142 0.0203 -0.0010 0.0114 0.0205 0.0259

ΔR&D/ME -0.0002 0.0000 -0.0003 0.0013 -0.0002 0.0000

ΔInterest/ME -0.0004 0.0000 -0.0007 0.0000 -0.0002 0.0000

ΔDividends/ME -0.0005 0.0000 -0.0001 0.0000 -0.0006 0.0000

Net Finance/ME 0.0479 0.0016 0.0542 0.0044 0.0453 0.0000

Obs 50645 14835 35810

106

Table 3.2: Value of Cash

This table shows the results of regressions for the full sample period of 1972-2007 and two sub-periods of equal

length (1972-1989 and 1990-2007). Panel A reports the regression results. In Panel B, I use the mean levels of

lagged cash and leverage of each subsample, along with high-tech dummy, to compute the marginal value of one

dollar cash for the average firm in the sample. The marginal value for the average firm is coefficient on the change

in cash plus the sample means for all variables that are interacted with the change in cash times the respective

regression coefficient from the model. Data are from the CRSP-Compustat merged database (Fundamental Annual)

for the period of 1972–2007. I exclude companies from the financial (SIC 6000–6999) and utility (SIC 4900–4999)

industries. I also exclude all observations with missing data.

, , , , , ,

, , 0 1 2 3 4 5 6

, 1 , 1 , 1 , 1 , 1 , 1

, 1 , , 1 , ,

7 8 , 9 10 11 ,

, 1 , 1 , 1 , 1 , 1

,

12

* *

*

i t i t i t i t i t i tB

i t i t

i t i t i t i t i t i t

i t i t i t i t i t

i t i t

i t i t i t i t i t

i t

C E NA RD I Dr R

M M M M M M

C NF C C CL L

M M M M M

CHTDummy

13 ,

, 1

i t

i t

HTDummyM

I follow the logic of Dittmar and Mahrt-Smith (2007) to modify Faulkender and Wang (2006) method by

incorporating an interaction term between changes in cash holdings and the high-tech indicator variable

(HTDummy). HTdummy takes a value of 1 if a firm belongs to the high-tech sector defined according to the U.S.

Department of Commerce, which includes seven industries defined by 3-digit SIC codes 283, 357, 366, 367, 382,

384, and 737. The details on variable definitions are in Appendix 3.A. Robust t-statistics are reported in parentheses.

For each regression, adjusted R2 is reported. The standard errors are adjusted for clustering on firms. They are

computed assuming observations are independent across firms, but not across time. *, **, and *** indicate

statistical significance at the 10%, 5%, and 1% levels, respectively. #, ##, and ### indicate statistical significance at

the 10%, 5%, and1% levels, respectively, for a t-test that tests whether the coefficients are equal between the two

sub-periods. “F-test” indicates the joint test that all coefficients are equal between two sub-periods.

107

Panel A: Regression Results

1972-2007 1972-1989 1990-2007

Independent Variables [1] [2] [3] [4]

ΔCash/ME 1.564*** 1.398*** 1.063*** 1.678*** ###

(43.12) (35.24) (20.56) (28.89)

HTDummy *ΔCash/ME 0.398*** 0.209*** 0.417*** ##

(8.81) (3.25) (6.95)

HTDummy -0.044*** -0.041*** -0.043***

(-9.26) (-6.33) (-6.74)

ΔEarnings/ME 0.412*** 0.414*** 0.404*** 0.415***

(36.33) (36.51) (26.52) (25.74)

ΔNet Assets/ME 0.169*** 0.166*** 0.149*** 0.186*** ###

(28.69) (28.07) (20.69) (19.97)

ΔR&D/ME 1.182*** 1.151*** 1.599*** 0.842*** ###

(9.75) (9.48) (9.06) (5.13)

ΔInterest/ME -1.295*** -1.269*** -1.124*** -1.537*** ##

(-16.65) (-16.30) (-12.32) (-11.21)

ΔDividends/ME 3.288*** 3.296*** 4.190*** 2.078*** ###

(16.50) (16.58) (16.55) (6.30)

Lagged Cash/ME 0.301*** 0.310*** 0.268*** 0.329*** ##

(22.85) (23.37) (16.98) (15.90)

Leverage -0.477*** -0.499*** -0.452*** -0.564*** ###

(-58.75) (-57.46) (-39.27) (-45.85)

Net Finance/ME 0.041*** 0.041*** 0.037*** 0.034*

(3.59) (3.59) (2.72) (1.81)

Lagged Cash/ME *ΔCash/ME -0.746*** -0.753*** -0.533*** -1.021*** ###

(-13.76) (-13.88) (-8.08) (-12.39)

Leverage*ΔCash/ME -1.426*** -1.175*** -0.803*** -1.349*** ###

(-20.79) (-16.30) (-8.55) (-12.23)

Constant 0.054*** 0.070*** 0.087*** 0.058*** ###

(17.16) (19.82) (17.69) (11.99)

Observations 95880 95880 45235 50645

Adj. R-squared 0.20 0.199 0.201 0.205

F-test (p-value) 37.25 (0.00)

Panel B: The Marginal Value of Cash for the Average Firms

Mean(Lagged Cash/ME) 0.179 0.179 0.191 0.169

Mean(Lt) 0.279 0.279 0.328 0.236

Mean (HTDummy) 0.237 0.173 0.293

Marginal value of $1 cash (HT) 1.33 0.91 1.60

Marginal value of $1 cash (NHT) 0.93 0.70 1.19

Marginal value of $1 cash (Average) 1.03 1.03 0.73 1.31

108

Table 3.3: Robustness Check – Alternative Industry Classifications and Sub-

periods

This table provides the robustness check for alternative definitions of high-tech sector and alternative division of

sub-periods. Panel A is about alternative industry classifications. According to the Fama-French 12 industry

classification, Business Equipment (BusEq) and Healthcare (Hlth) are regarded as the high-tech sector. According

to the Global Industry Classification Standard (GICS), Information Technology (including software, hardware, and

electronic equipment) and Health Care (including health care equipment and services, pharmaceuticals, biotech and

life sciences) sectors are regarded as high-tech. Panel B considers alternative division of sub-periods. In Panel B,

the high-tech sector is defined according to the U.S. Department of Commerce, which includes seven industries

defined by 3-digit SIC codes 283, 357, 366, 367, 382, 384, and 737. Data are from the CRSP-Compustat merged

database (Fundamental Annual) for the period of 1972–2007. I exclude companies from the financial (SIC 6000–

6999) and utility (SIC 4900–4999) industries. I also exclude all observations with missing data.

, , , , , ,

, , 0 1 2 3 4 5 6

, 1 , 1 , 1 , 1 , 1 , 1

, 1 , , 1 , ,

7 8 , 9 10 11 ,

, 1 , 1 , 1 , 1 , 1

,

12

* *

*

i t i t i t i t i t i tB

i t i t

i t i t i t i t i t i t

i t i t i t i t i t

i t i t

i t i t i t i t i t

i t

C E NA RD I Dr R

M M M M M M

C NF C C CL L

M M M M M

CHTDummy

13 ,

, 1

i t

i t

HTDummyM

I follow the logic of Dittmar and Mahrt-Smith (2007) to modify Faulkender and Wang (2006) method by

incorporating an interaction term between changes in cash holdings and the high-tech indicator variable

(HTDummy). The details on variable definitions are in Appendix 3.A. Coefficient estimates of some variables are

not reported here. Robust t-statistics are reported in parentheses. For each regression, adjusted R2 is reported. The

standard errors are adjusted for clustering on firms. They are computed assuming observations are independent

across firms, but not across time. *, **, and *** indicate statistical significance at the 10%, 5%, and 1% levels,

respectively. In Panel A, #, ##, and ### indicate statistical significance at the 10%, 5%, and1% levels, respectively,

for a t-test that tests whether the coefficients are equal between the two sub-periods. “F-test” indicates the joint test

that all coefficients are equal between two sub-periods.

109

Panel A: Alternative Industry Classification (FF12 and GICS)

FF12 Classification GICS Classification

1972-1989 1990-2007 1972-1989 1990-2007

Independent variables [1] [2] [3] [4]

ΔCash/ME 1.038*** 1.681*** ###

1.083*** 1.702*** ###

(20.30) (28.63)

(18.65) (28.56)

HTDummy *ΔCash/ME 0.251*** 0.383*** #

0.207*** 0.365*** #

(4.28) (6.77)

(3.24) (6.47)

HTDummy -0.025*** -0.036***

-0.037*** -0.041***

(-4.00) (-6.01)

(-5.38) (-6.89)

Lagged Cash/ME *ΔCash/ME -0.522*** -1.014*** ###

-0.515*** -1.052*** ###

(-7.94) (-12.31)

(-6.67) (-12.41)

Leverage*ΔCash/ME -0.782*** -1.388*** ###

-0.875*** -1.375*** ###

(-8.42) (-12.79)

(-8.08) (-12.37)

Constant 0.082*** 0.057*** ###

0.086*** 0.059*** ###

(16.78) (11.66)

(16.08) (11.95)

Observations 45235 50645 36827 49962

Adj. R-squared 0.201 0.205 0.203 0.207

F-test (p-value) 36.89 (0.00) 29.52 (0.00)

Panel B: Alternative Division of Sub-Periods

1972-1980 1981-1989 1990-1998 1999-2007

Independent Variables [1] [2] [3] [4]

ΔCash/ME 0.948*** 1.124*** 1.540*** 1.895***

(12.85) (16.00) (21.18) (20.40)

HTDummy *ΔCash/ME 0.274** 0.146* 0.436*** 0.369***

(2.34) (1.91) (5.31) (4.05)

HTDummy 0.018 -0.068*** -0.006 -0.091***

(1.60) (-8.49) (-0.77) (-9.44)

Lagged Cash/ME *ΔCash/ME -0.469*** -0.596*** -0.930*** -1.148***

(-5.48) (-6.05) (-8.43) (-9.41)

Leverage*ΔCash/ME -0.724*** -0.855*** -1.409*** -1.336***

(-5.53) (-6.64) (-9.83) (-7.82)

Constant 0.087*** 0.097*** 0.047*** 0.073***

(11.92) (14.71) (7.49) (9.90)

Observations 20757 24478 28394 22251

Adj. R-squared 0.210 0.207 0.202 0.213

110

Table 3.4: Robustness Check--- Faulkender and Wang (2006) Model

This table presents the results of regressions for the full sample period of 1972-2007 and two sub-periods of equal

length (1972-1989 and 1990-2007) by using the original methodology designed by Faulkender and Wang (2006).

Panel A reports the regression results. In Panel B, I use the mean levels of lagged cash and leverage of each

subsample to compute the marginal value of one dollar cash for the average firm in the sample. The marginal value

for the average firm is coefficient on the change in cash plus the sample means for all variables that are interacted

with the change in cash times the respective regression coefficient from the model.

, , , , , ,

, , 0 1 2 3 4 5 6

, 1 , 1 , 1 , 1 , 1 , 1

, 1 , , 1 , ,

7 8 , 9 10 11 , ,

, 1 , 1 , 1 , 1 , 1

* *

i t i t i t i t i t i tB

i t i t

i t i t i t i t i t i t

i t i t i t i t i t

i t i t i t

i t i t i t i t i t

C E NA RD I Dr R

M M M M M M

C NF C C CL L

M M M M M

The high-tech sector is defined according to the U.S. Department of Commerce, which includes seven industries

defined by 3-digit SIC codes 283, 357, 366, 367, 382, 384, and 737. All the remaining firms are categorized as non-

high-tech. Data are from the CRSP-Compustat merged database (Fundamental Annual) for the period of 1972–

2007. I exclude companies from the financial (SIC 6000–6999) and utility (SIC 4900–4999) industries. I also

exclude all observations with missing data. The details on variable definitions are in Appendix 3.A. Robust t-

statistics are reported in parentheses. For each regression, adjusted R2 is reported. The standard errors are adjusted

for clustering on firms. They are computed assuming observations are independent across firms, but not across time.

*, **, and *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively.

111

Panel A: Regression Results

1972-2007 1972-1989 1990-2007

High-Tech

Non-

High-Tech High-Tech

Non-

High-Tech High-Tech

Non-

High-Tech

Independent Variables [1] [2] [3] [4] [5] [6]

ΔCash/ME 2.059*** 1.290*** 1.425*** 1.024*** 2.318*** 1.549***

(28.16) (30.53) (12.96) (18.82) (25.03) (24.26)

ΔEarnings/ME 0.454*** 0.400*** 0.476*** 0.388*** 0.444*** 0.402***

(17.12) (32.14) (10.34) (24.20) (13.72) (21.71)

ΔNet Assets/ME 0.267*** 0.153*** 0.244*** 0.140*** 0.280*** 0.172***

(13.99) (24.89) (9.71) (18.65) (10.73) (17.43)

ΔR&D/ME 1.206*** 0.780*** 1.746*** 1.197*** 0.952*** 0.226

(7.25) (4.21) (6.26) (5.23) (4.60) (0.76)

ΔInterest/ME -1.695*** -1.199*** -2.033*** -1.036*** -1.457*** -1.524***

(-6.49) (-14.90) (-5.38) (-11.07) (-4.21) (-10.34)

ΔDividends/ME 3.548*** 3.410*** 4.702*** 4.262*** 2.950*** 2.038***

(4.49) (16.78) (4.53) (16.44) (2.65) (6.00)

Lagged Cash/ME 0.416*** 0.275*** 0.347*** 0.255*** 0.440*** 0.270***

(13.19) (19.17) (6.56) (15.56) (11.88) (11.23)

Leverage -0.538*** -0.480*** -0.474*** -0.443*** -0.620*** -0.538***

(-23.29) (-50.93) (-15.01) (-35.79) (-18.94) (-40.62)

Net Finance/ME 0.116*** 0.024** 0.078* 0.031** 0.125*** 0.007

(3.33) (1.98) (1.72) (2.22) (2.60) (0.34)

Lagged Cash/ME *ΔCash/ME -1.117*** -0.615*** -0.673*** -0.504*** -1.400*** -0.803***

(-8.54) (-10.68) (-2.84) (-7.55) (-8.98) (-8.51)

Leverage*ΔCash/ME -1.777*** -1.064*** -1.247*** -0.738*** -1.669*** -1.285***

(-9.86) (-13.70) (-4.57) (-7.43) (-6.73) (-10.49)

Constant 0.001 0.072*** 0.032*** 0.087*** -0.013 0.064***

(0.17) (20.03) (2.82) (17.09) (-1.45) (12.59)

Observations 22676 73204 7841 37394 14835 35810

Adj. R-squared 0.22 0.19 0.22 0.20 0.23 0.19

Panel B: The Marginal Value of Cash for the Average Firm

Mean(Lagged Cash/ME) 0.191 0.175 0.152 0.199 0.212 0.151

Mean(Lt) 0.151 0.319 0.211 0.352 0.119 0.285

The marginal value of $1 cash 1.58 0.84 1.06 0.66 1.82 1.06

Difference 0.74 0.40 0.76

112

Table 3.5: Robustness Check – Benchmark Returns based on the DGTW

portfolios

This table shows the regression results, whose benchmark returns are from the 125 DGTW portfolios based on a

triple-sort on size, book-to-market, and momentum. These portfolios are developed by Daniel, Grinblatt, Titman,

and Wermers (1997) and the data are available from 1975. The regression results are for the full sample period of

1975-2007 and two sub-periods (1975-1989 and 1990-2007). Data are from the CRSP-Compustat merged database

(Fundamental Annual) for the period of 1972–2007. I exclude companies from the financial (SIC 6000–6999) and

utility (SIC 4900–4999) industries. I also exclude all observations with missing data.

, , , , , ,

, , 0 1 2 3 4 5 6

, 1 , 1 , 1 , 1 , 1 , 1

, 1 , , 1 , ,

7 8 , 9 10 11 ,

, 1 , 1 , 1 , 1 , 1

,

12

* *

*

i t i t i t i t i t i tB

i t i t

i t i t i t i t i t i t

i t i t i t i t i t

i t i t

i t i t i t i t i t

i t

C E NA RD I Dr R

M M M M M M

C NF C C CL L

M M M M M

CHTDummy

13 ,

, 1

i t

i t

HTDummyM

I follow the logic of Dittmar and Mahrt-Smith (2007) to modify Faulkender and Wang (2006) method by

incorporating an interaction term between changes in cash holdings and the high-tech indicator variable

(HTDummy). HTdummy takes a value of 1 if a firm belongs to the high-tech sector defined according to the U.S.

Department of Commerce, which includes seven industries defined by 3-digit SIC codes 283, 357, 366, 367, 382,

384, and 737. The details on variable definitions are in Appendix 3.A. Robust t-statistics are reported in parentheses.

For each regression, adjusted R2 is reported. The standard errors are adjusted for clustering on firms. They are

computed assuming observations are independent across firms, but not across time. *, **, and *** indicate

statistical significance at the 10%, 5%, and 1% levels, respectively.

113

1975-2007 1975-1989 1990-2007

Independent variables [1] [2] [3]

ΔCash/ME 1.500*** 1.184*** 1.757***

(32.98) (19.33) (27.16)

HTDummy *ΔCash/ME 0.440*** 0.280*** 0.448***

(8.57) (3.76) (6.65)

HTDummy -0.059*** -0.053*** -0.061***

(-11.60) (-7.29) (-8.95)

ΔEarnings/ME 0.450*** 0.450*** 0.443***

(34.75) (24.89) (24.87)

ΔNet Assets/ME 0.164*** 0.155*** 0.176***

(24.00) (17.72) (16.94)

ΔR&D/ME 0.945*** 1.294*** 0.731***

(6.75) (6.12) (3.95)

ΔInterest/ME -1.960*** -1.939*** -2.021***

(-21.12) (-17.16) (-12.94)

ΔDividends/ME 3.709*** 4.477*** 2.428***

(14.11) (13.60) (5.50)

Lagged Cash/ME 0.354*** 0.313*** 0.367***

(24.30) (17.42) (16.76)

Leverage -0.499*** -0.434*** -0.571***

(-51.08) (-32.17) (-43.42)

Net Finance/ME 0.062*** 0.054*** 0.057***

(4.69) (3.32) (2.65)

Lagged Cash/ME *ΔCash/ME -0.877*** -0.654*** -1.133***

(-13.94) (-8.30) (-12.29)

Leverage*ΔCash/ME -1.176*** -0.835*** -1.332***

(-13.86) (-7.27) (-10.81)

Constant 0.081*** 0.095*** 0.074***

(20.82) (16.89) (14.47)

Observations 80895 35827 45068

Adj. R-squared 0.193 0.187 0.202

114

Table 3.6: Fundamental Explanation

This table compares the population characteristics related to the precautionary motive of holding cash between

high-tech (HT) and non-high-tech (NHT) firms over two sub-periods of equal length, [1972, 1989] and [1990,

2007]. High-tech firms are defined according to the U.S. Department of Commerce, which includes seven

industries defined by 3-digit SIC codes 283, 357, 366, 367, 382, 384, and 737. The details on variable

definitions are in Appendix 3.A. Mean and median of each variable are reported for each subsample. Column

[3] and column [7] (Column [4] and column [8]) contain statistical tests on the equal mean (equal distribution)

between high-tech and non-high-tech firms during first and second sub-period respectively. Column [9] and

column [10] (Column [11] and Column [12]) contains statistical tests on the equal mean and equal distribution

for observations in two sub-periods in the non-high-tech sector (the high-tech sector) respectively. The t-

statistics are tests for equal means across two groups. The z-statistics are from the Wilcoxon rank-sum test,

which tests whether observations in two groups are from populations with the same distribution.

1972-1989 1990-2007 NHT

over 2 sub-periods HT

over 2 sub-periods

NHT HT t-stat z-stat NHT HT t-stat z-stat t-stat z-stat t-stat z-stat

[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12]

R&D/TA Mean 0.0102 0.0775 69.0 108.5 0.0131 0.1240 107.7 145.7 10.8 -12.5 33.4 28.1

Median 0.0000 0.0570 0.0000 0.0893

CapEx/TA Mean 0.0826 0.0752 -8.6 -6.3 0.0695 0.0480 -37.6 -33.3 -23.7 -35.0 -31.3 -39.3

Median 0.0606 0.0561 0.0463 0.0334

Industry Sigma Mean 0.0441 0.0635 55.8 66.5 0.0860 0.1392 101.6 109.4 129.3 143.6 140.4 106.7

Median 0.0352 0.0560 0.0713 0.1231

CF/TA Mean 0.0509 0.0079 -15.0 2.9 0.0428 -0.0689 -39.3 -26.0 -7.0 11.3 -19.8 -14.3

Median 0.0690 0.0733 0.0744 0.0554

NetEiss Mean 0.0119 0.0434 20.7 36.8 0.0178 0.0703 33.3 50.2 9.1 7.4 12.8 14.8

Median 0.0000 0.0019 0.0000 0.0068

NetDiss Mean 0.0150 0.0131 -1.7 -2.5 0.0138 0.0082 -6.5 -5.1 -1.9 -5.6 -3.9 -4.1

Median 0.0000 0.0000 0.0000 0.0000

Cash/TA Mean 0.0941 0.1553 29.5 26.1 0.1042 0.2962 88.0 90.6 10.5 -14.3 49.4 42.7

Median 0.0533 0.0849 0.0469 0.2358

Obs 37394 7841 35810 14835

115

Table 3.7: Mispricing Explanation – Impact of Changes in Sentiment

This table reports the impact of changes in investor sentiment on market valuation of equity of high-tech (HT) and non-high-tech (NHT) firms during the full

sample period of 1972-2007 and two sub-periods of equal length (1972-1989 and 1990-2007). Monthly data of the sentiment index designed by Baker and

Wurgler (2007), available for the period of July 1965 - December 2007, are matched to the beginning and end of each given fiscal year, the difference of which

measures the annual change in investor sentiment (ΔBWSI). ΔBWSI is incorporate into the original regression model designed by Faulkender and Wang (2006).

, , , , , ,

, , 0 1 2 3 4 5 6

, 1 , 1 , 1 , 1 , 1 , 1

, 1 , , 1 , ,

7 8 , 9 10 11 ,

, 1 , 1 , 1 , 1 , 1

12 , ,

* *

BWSI

i t i t i t i t i t i tB

i t i t

i t i t i t i t i t i t

i t i t i t i t i t

i t i t

i t i t i t i t i t

i t i t

C E NA RD I Dr R

M M M M M M

C NF C C CL L

M M M M M

The high-tech sector is defined according to the U.S. Department of Commerce, which includes seven industries defined by 3-digit SIC codes 283, 357, 366,

367, 382, 384, and 737. All the remaining firms are categorized as non-high-tech. Data are from the CRSP-Compustat merged database (Fundamental Annual)

for the period of 1972–2007. I exclude companies from the financial (SIC 6000–6999) and utility (SIC 4900–4999) industries. I also exclude all observations

with missing data. The details on variable definitions are in Appendix 3.A. Robust t-statistics are reported in parentheses. For each regression, adjusted R2 is

reported. The standard errors are adjusted for clustering on firms. They are computed assuming observations are independent across firms, but not across time. *,

**, and *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively. The last three columns test the equality of coefficient estimates across

two subperiods, for all firms, high-tech firms, and non-high-tech firms respectively. #, ##, and ### indicate statistical significance at the 10%, 5%, and1% levels,

respectively, for a t-test that tests whether the coefficients are equal between the two sub-periods. “F-test” indicates the joint test that all coefficients are equal

between two sub-periods.

116

1972-2007 1972-1989 1990-2007

Compare Two

Subperiods

Independent Variables All firms HT NHT All firms HT NHT All firms HT NHT

All

firms HT NHT

[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12]

ΔBWSI -0.010*** 0.026*** -0.024*** -0.002 0.016* -0.006* -0.018*** 0.032*** -0.044*** ### ###

(-4.06) (4.15) (-8.84) (-0.52) (1.72) (-1.95) (-4.40) (3.90) (-9.48)

ΔCash/ME 1.564*** 2.054*** 1.286*** 1.125*** 1.424*** 1.023*** 1.884*** 2.310*** 1.544*** ### ### ###

(43.02) (28.09) (30.34) (23.36) (12.96) (18.81) (36.24) (24.88) (24.04)

ΔEarnings/ME 0.412*** 0.457*** 0.401*** 0.404*** 0.475*** 0.389*** 0.412*** 0.449*** 0.401***

(36.28) (17.25) (32.17) (26.48) (10.34) (24.21) (25.42) (13.89) (21.63)

ΔNet Assets/ME 0.169*** 0.265*** 0.153*** 0.150*** 0.243*** 0.140*** 0.191*** 0.275*** 0.173*** ### ###

(28.70) (13.82) (24.90) (20.91) (9.67) (18.65) (20.55) (10.52) (17.55)

ΔR&D/ME 1.193*** 1.198*** 0.803*** 1.566*** 1.738*** 1.204*** 0.907*** 0.945*** 0.254 ### ## ##

(9.82) (7.21) (4.33) (8.91) (6.24) (5.26) (5.52) (4.57) (0.86)

ΔInterest/ME -1.288*** -1.676*** -1.182*** -1.131*** -2.024*** -1.034*** -1.557*** -1.441*** -1.467*** ### ##

(-16.57) (-6.43) (-14.70) (-12.40) (-5.36) (-11.06) (-11.36) (-4.17) (-9.95)

ΔDividends/ME 3.301*** 3.554*** 3.421*** 4.210*** 4.643*** 4.264*** 2.034*** 2.998*** 2.050*** ### ###

(16.53) (4.47) (16.81) (16.60) (4.48) (16.46) (6.12) (2.66) (6.00)

Lagged Cash/ME 0.301*** 0.420*** 0.275*** 0.269*** 0.350*** 0.255*** 0.314*** 0.446*** 0.270*** #

(22.82) (13.21) (19.17) (17.04) (6.57) (15.56) (15.36) (11.93) (11.23)

Leverage -0.477*** -0.537*** -0.481*** -0.439*** -0.473*** -0.444*** -0.539*** -0.619*** -0.538*** ### ### ###

(-58.73) (-23.16) (-51.01) (-39.02) (-15.02) (-35.79) (-47.09) (-18.86) (-40.53)

Net Finance/ME 0.041*** 0.114*** 0.023** 0.036*** 0.078* 0.032** 0.033* 0.122** 0.006

(3.59) (3.27) (1.98) (2.67) (1.72) (2.22) (1.75) (2.52) (0.27)

Lagged Cash/ME *ΔCash/ME -0.743*** -1.116*** -0.609*** -0.544*** -0.673*** -0.503*** -0.998*** -1.397*** -0.792*** ### ## ##

(-13.69) (-8.54) (-10.56) (-8.31) (-2.85) (-7.53) (-12.10) (-8.95) (-8.40)

Leverage*ΔCash/ME -1.428*** -1.769*** -1.063*** -0.875*** -1.244*** -0.738*** -1.666*** -1.663*** -1.289*** ### ###

(-20.78) (-9.81) (-13.65) (-9.51) (-4.57) (-7.43) (-16.20) (-6.69) (-10.49)

Constant 0.055*** 0.001 0.074*** 0.075*** 0.033*** 0.087*** 0.042*** -0.015 0.066*** ### ### ###

(17.27) (0.08) (20.34) (16.44) (2.87) (17.12) (9.60) (-1.63) (13.00)

Observations 95615 22604 73011 45235 7841 37394 50380 14763 35617

Adj. R-squared 0.197 0.223 0.193 0.200 0.216 0.200 0.203 0.232 0.194

F-test

(p-value) 42.64

(0.00)

11.99

(0.00)

30.58

(0.00)

117

Table 3.8: Mispricing Explanation – Periods of Large and Small Changes in

Sentiment

This table shows the results of regressions for the periods with large and small absolute changes in investor

sentiment during the full sample period of 1972-2007 and two sub-periods of equal length (1972-1989 and

1990-2007). Data are from the CRSP-Compustat merged database (Fundamental Annual) for the period of

1972–2007. I exclude companies from the financial (SIC 6000–6999) and utility (SIC 4900–4999) industries. I

also exclude all observations with missing data. Monthly data of the sentiment index designed by Baker and

Wurgler (2007), available for the period of July 1965 - December 2007, are matched to the beginning and end

of each given fiscal year, the difference of which measures the annual change in investor sentiment (ΔBWSI).

The sample is sorted into quartiles according to the distribution of the unique values of ΔBWSI. Panel A

reports descriptive statistics of ΔBWSI for each quartile. The top and bottom quartiles of ΔBWSI are classified

into the periods with large absolute changes in sentiment (Large |ΔBWSI|), while the middle two quartiles are

labelled as the periods with small absolute changes in sentiment (Small |ΔBWSI|). Panel B reports the

regression results for different subsamples.

, , , , , ,

, , 0 1 2 3 4 5 6

, 1 , 1 , 1 , 1 , 1 , 1

, 1 , , 1 , ,

7 8 , 9 10 11 ,

, 1 , 1 , 1 , 1 , 1

,

12

* *

*

i t i t i t i t i t i tB

i t i t

i t i t i t i t i t i t

i t i t i t i t i t

i t i t

i t i t i t i t i t

i t

C E NA RD I Dr R

M M M M M M

C NF C C CL L

M M M M M

CHTDummy

13 ,

, 1

i t

i t

HTDummyM

I follow the logic of Dittmar and Mahrt-Smith (2007) to modify Faulkender and Wang (2006) method by

incorporating an interaction term between changes in cash holdings and the high-tech indicator variable

(HTDummy). HTdummy takes a value of 1 if a firm belongs to the high-tech sector defined according to the

U.S. Department of Commerce, which includes seven industries defined by 3-digit SIC codes 283, 357, 366,

367, 382, 384, and 737. The details on variable definitions are in Appendix 3.A. Robust t-statistics are

reported in parentheses. For each regression, adjusted R2 is reported. The standard errors are adjusted for

clustering on firms. They are computed assuming observations are independent across firms, but not across

time. *, **, and *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively. #, ##, and

### indicate statistical significance at the 10%, 5%, and1% levels, respectively, for a t-test that tests whether

the coefficients are equal between the two sub-periods. “F-test” indicates the joint test that all coefficients are

equal between two sub-periods.

Panel A: Quartiles Based on the Distribution of Annual Changes in Baker and Wurgler Sentiment Index (BWSI)

Variable: ΔBWSI

Quartile

based on ΔBWSI Obs Mean StD Min Max

1 22399 -0.849 0.403 -2.465 -0.380

2 19933 -0.178 0.107 -0.378 0.002

3 28774 0.142 0.117 0.003 0.438

4 24509 0.807 0.295 0.441 1.977

118

Panel B: Regressions in Periods of Large and Small Changes in Sentiment

1972-2007 1972-1989 1990-2007

Independent Variables

Large

|ΔBWSI|

Small

|ΔBWSI|

Large

|ΔBWSI|

Small

|ΔBWSI|

Large

|ΔBWSI|

Small

|ΔBWSI|

[1] [2] [3] [4] [5] [6]

ΔCash/ME 1.438*** 1.352***

1.116*** 0.998***

1.738*** 1.625***

(26.12) (24.19)

(15.97) (13.66)

(21.12) (19.73)

HTDummy *ΔCash/ME 0.396*** 0.402***

0.229** 0.191**

0.373*** 0.452***

(6.35) (6.32)

(2.46) (2.15)

(4.50) (5.28)

HTDummy -0.056*** -0.032***, ###

-0.034*** -0.049***

-0.065*** -0.025***, ###

(-8.38) (-5.10)

(-3.60) (-5.40)

(-6.88) (-2.96)

ΔEarnings/ME 0.390*** 0.442***, ##

0.426*** 0.383***

0.354*** 0.489***, ###

(25.35) (27.54)

(20.52) (18.55)

(16.24) (20.48)

ΔNet Assets/ME 0.156*** 0.179***, #

0.139*** 0.164***, #

0.173*** 0.202***

(20.03) (20.29)

(14.74) (15.23)

(14.09) (14.02)

ΔR&D/ME 1.365*** 0.943***, #

1.860*** 1.359***

0.880*** 0.815***

(7.94) (5.51)

(7.84) (5.37)

(3.68) (3.60)

ΔInterest/ME -1.079*** -1.513***, ###

-0.896*** -1.463***, ###

-1.427*** -1.657***

(-10.44) (-12.87)

(-7.40) (-10.54)

(-7.83) (-8.02)

ΔDividends/ME 3.232*** 3.416***

4.353*** 4.037***

1.389*** 2.760***, ##

(11.53) (12.08)

(13.47) (10.68)

(2.74) (6.37)

Lagged Cash/ME 0.329*** 0.291***, #

0.330*** 0.204***, ###

0.303*** 0.346***

(19.24) (17.03)

(16.01) (10.15)

(10.96) (13.24)

Leverage -0.503*** -0.504***

-0.448*** -0.465***

-0.587*** -0.548***, #

(-43.37) (-43.73)

(-30.07) (-29.86)

(-34.83) (-32.80)

Net Finance/ME 0.030* 0.052***

0.030 0.045**

0.026 0.042

(1.94) (3.12)

(1.63) (2.24)

(0.99) (1.50)

Lagged Cash/ME *ΔCash/ME -0.681*** -0.819***

-0.443*** -0.604***

-1.009*** -1.058***

(-9.24) (-10.48)

(-5.00) (-6.20)

(-8.79) (-9.01)

Leverage*ΔCash/ME -1.251*** -1.097***

-0.940*** -0.650***

-1.395*** -1.333***

(-12.80) (-10.56)

(-7.43) (-4.83)

(-9.55) (-8.17)

Constant 0.080*** 0.061***, ###

0.084*** 0.091***

0.079*** 0.041***, ###

(16.08) (13.19)

(12.71) (13.56)

(10.89) (6.50)

Observations 46908 48707 23425 21810 23483 26897

Adj. R-squared 0.199 0.201 0.209 0.197 0.201 0.212

F-test (p-value) 6.28 (0.00) 5.86 (0.00) 4.27 (0.00)

119

Table 3.A.1: Replication of Faulkender and Wang (2006) – Summary

Statistics

This table replicates Table I of Faulkender and Wang (2006), providing summary statistics for those variables used

in the model on value of cash over the period 1972 to 2001. The sample contains 82,100 firm-year observations.

The details on variable definitions are in Appendix 3.A.

Variable Mean 1st Quartile Median 3

rd Quartile SD

, ,

B

i t i tr R -0.0121 -0.3573 -0.0914 0.2031 0.5799

ΔCash/ME 0.0018 -0.0398 -0.0008 0.0348 0.1538

Lagged Cash/ME 0.1780 0.0350 0.0953 0.2178 0.2384

ΔEarnings/ME 0.0098 -0.0425 0.0062 0.0487 0.2474

ΔNet Assets/ME -0.0076 -0.1063 0.0248 0.1590 0.5943

ΔR&D/ME 0.0003 0.0000 0.0000 0.0005 0.0202

ΔInterest/ME 0.0004 -0.0045 0.0000 0.0077 0.0376

ΔDividends/ME -0.0002 0.0000 0.0000 0.0001 0.0087

Leverage 0.2937 0.0694 0.2420 0.4706 0.2508

Net Finance/ME 0.0505 -0.0319 0.0013 0.0898 0.2743

120

Table 3.A.2: Replication of Faulkender and Wang (2006) – Regression

Results

This table replicates the main part of Table II of Faulkender and Wang (2006), which covers the period from 1972

to 2001. The regression model is

, , , , , ,

, , 0 1 2 3 4 5 6

, 1 , 1 , 1 , 1 , 1 , 1

, 1 , , 1 , ,

7 8 , 9 10 11 , ,

, 1 , 1 , 1 , 1 , 1

* *

i t i t i t i t i t i tB

i t i t

i t i t i t i t i t i t

i t i t i t i t i t

i t i t i t

i t i t i t i t i t

C E NA RD I Dr R

M M M M M M

C NF C C CL L

M M M M M

Data are from the CRSP-Compustat merged database (Fundamental Annual) for the period of 1972–2007. I exclude

companies from the financial (SIC 6000–6999) and utility (SIC 4900–4999) industries. I also exclude all

observations with missing data. The details on variable definitions are in Appendix 3.A. Robust t-statistics are

reported in parentheses. For each regression, adjusted R2 is reported. The standard errors are adjusted for clustering

on firms. They are computed assuming observations are independent across firms, but not across time. *, **, and

*** indicate statistical significance at the 10%, 5%, and 1% levels, respectively.

Independent Variables [1] [2]

ΔCash/ME 0.773*** 1.488***

(36.91) (38.56)

ΔEarnings/ME 0.403*** 0.400***

(33.23) (33.49)

ΔNet Assets/ME 0.152*** 0.162***

(24.50) (26.28)

ΔR&D/ME 1.395*** 1.331***

(10.60) (10.25)

ΔInterest/ME -1.296*** -1.245***

(-15.98) (-15.49)

ΔDividends/ME 3.359*** 3.280***

(15.82) (15.53)

Lagged Cash/ME 0.372*** 0.298***

(27.66) (20.80)

Leverage -0.491*** -0.496***

(-54.59) (-55.64)

Net Finance/ME 0.079*** 0.055***

(6.66) (4.73)

Lagged Cash/ME *ΔCash/ME -0.712***

(-12.45)

Leverage*ΔCash/ME -1.353***

(-18.60)

Constant 0.059*** 0.063***

(16.43) (17.92)

Observations 82,100 82,100

Adj. R-squared 0.18 0.20

121

Figure 3.1: Rolling Windows

This plot reports the coefficient estimates of the interaction term between HTDummy and the change in cash in

equation (1) over a series of three-year rolling windows. HTdummy takes a value of 1 if a firm belongs to the high-

tech sector defined according to the U.S. Department of Commerce, which includes seven industries defined by 3-

digit SIC codes 283, 357, 366, 367, 382, 384, and 737.

122

Appendix 3.A: Variable Definitions

Variable Definition

Market value of equity (Mt) the number of shares (CSHPRI) multiplied by the stock‟s closing price at the end

of fiscal year t (PRCC_F)

Cash holdings (Ct) cash plus marketable securities (CHE)

Net assets (NAt) total assets (AT) minus cash holdings (CHE)

Earnings (Et) earnings before extraordinary items plus interest, deferred tax credits, and

investment tax credits (IB+XINT+TXDI+ITCI)

Total dividends (Dt) common dividends paid (DVC)

Leverage (Lt) the market debt ratio, calculated as the sum of long term debt and short term debt

(DLTT+DLC) over the sum of total debt and the market value of equity

Net financing (NFt) total equity issuance (SSTK) - repurchases (PRSTKC) + debt issuance (DLTIS) -

debt redemption (DLTR)

R&D expenditures (RDt) R&D expenditure (XRD), which equals zero if missing

Interest expense (It) interest expense during the fiscal year t (XINT)

Excess stock return during

fiscal year t (, ,

B

i t i tr R ) The stock return minus the return on a benchmark portfolio over the fiscal year. ,i tr

is stock i‟s return over fiscal year t. ,

B

i tR is stock i's benchmark return over fiscal

year t. The benchmark portfolio is Fama and French (1993) size and book-to-market

matched portfolio. Data are available from Ken French‟s website.

Cash flow over assets

(CF/TA)

operating income before depreciation (OIBDP), less interest (XINT) and taxes

(TXT), and then scaled by total assets (AT)

Capital expenditures

(Capex/TA)

the ratio of Capital expenditures (CAPX) to total assets (AT)

Industry cash flow volatility

(Industry Sigma)

For each firm-year, I compute the standard deviation of cash flow over assets for the

previous 10 years if there are at least 3 observations. Industry sigma is calculated as

the mean of cash flow standard deviations of firms in the same industry, defined by

2-digit SIC code

Net debt issuance (NetDiss) long-term debt issuance (DLTIS) minus long-term debt reduction (DLTR), scaled

by total assets (AT)

Net equity issuance

(NetEiss)

the sale of common and preferred stock (SSTK) minus the purchase of common and

preferred stock (PRSTKC), scaled by total assets (AT)

Cash-to-assets ratio

(Cash/TA)

cash plus marketable securities (CHE), scaled by total assets (AT)

All names in parentheses refer to the Compustat (XPF version, Fundamental Annual) item names. All the

key values are converted to 2008 dollars.

123

Appendix 3.B: Replicating Faulkender and Wang (2006)

This paper uses the modified version of the empirical methodology designed by Faulkender and

Wang (2006) to estimate the shareholder‟s valuation of a marginal dollar of cash on a firm‟s

balance sheet. The sample in Faulkender and Wang (2006) covers all the publicly listed U.S.

firms during the period of 1972-2001. Since my sample has extended the investigating period to

2007, it is import to show that I can replicate their results during their time period.

[Put Table 3.A.1 here]

[Put Table 3.A.2 here]

Table A.1 and A.2 are the replications of their summary statistics and main regression results.

During the period of 1972-2001, there are 82,100 observations used in the regression. The

sample size, summary statistics, and coefficient estimates are quantitatively close to their results.

The marginal value of cash is decreasing in the level of a firm‟s cash position and leverage

position. During this period, the marginal value of extra one dollar of cash to shareholders in the

mean firm is $0.96 (=$1.488 + ($0.712*0.1780) + (-$ 1.353*0.2937)).

124

References

Acharya, A., Almeida, H., Campello, M., 2007. Is cash negative debt? A hedging perspective on

corporate financial policies. Journal of Financial Intermediation 16, 515-554.

Aivazian, V., Booth, L., Cleary, S., 2006. Dividend smoothing and debt ratings. Journal of Financial and

Quantitative Analysis 41, 439-453.

Akdogu, E., MacKay, P., 2008. Investment and competition. Journal of Financial and Quantitative

Analysis 43, 299-330.

Ali, A., Klasa, A., Yeung, E., 2009. The limitations of industry concentration measures constructed with

Compustat data: implications for finance research. Review of Financial Studies 22, 3839-3871.

Altman, E.I., Haldeman, R.G., Narayanan, P., 1977. Zeta analysis: a new model to identify bankruptcy

risk of corporations. Journal of Banking and Finance 1, 29-54.

Baker, M. P., Ruback, R.S., Wurgler, J., 2004. Behavioral corporate finance: a survey. in Espen Eckbo,

ed.: The Handbook of Corporate Finance: Empirical Corporate Finance (Elsevier, Amsterdam).

Baker, M., Wurgler, J., 2006. Investor sentiment and the cross-section of stock returns. Journal of

Finance, 61, 1645-1680.

Baker, M., Wurgler, J., 2007. Investor sentiment in the stock market. Journal of Economic Perspectives,

21, 129-151.

Barber, B.M., Lyon,J.D., 1996. Detecting abnormal operating performance: the empirical power and

specification of tests-statistics. Journal of Financial Economics 41, 359–399.

Bates, T.W., Kahle, K.M., Stulz, R.M., 2009. Why do U.S. firms hold so much more cash than they used

to? Journal of Finance 64, 1985- 2021.

Baumol, W.J, 1982. Contestable markets: an uprising in the theory of industry structure. American

Economic Review 72, 1-15.

Bekaert, G., Harvey, C., Lundblad, C., Siegel, S., 2008. What segments equity markets? Unpublished

working paper, Columbia University.

Bernard, A.B., Jensen, J.B., Schott, P.K., 2006. Survival of the best fit: exposure to low-wage countries

and the (uneven) growth of us manufacturing plants. Journal of International Economics 68, 219-

237.

Bhojraj, S., Lee, C.M.C., Oler, D., 2003. What‟s my line? A comparison of industry classification

schemes for capital market research. Journal of Accounting Research 41, 745–74.

Booth, L., 1980. Stochastic demand, output and the cost of capital: a clarification. Journal of Finance 35,

795-798.

Booth, L., 1981. Market structure uncertainty and the cost of equity capital. Journal of Banking and

Finance 5, 467-482.

Booth, L., Xu, Z., 2007. Who smooth dividends? Unpublished working paper, University of Toronto.

Bradley, M., Jarrell, G., Kim, E.H., 1984. On the existence of an optimal capital structure: theory and

evidence. Journal of Finance 39, 857-878.

125

Brav, A., Graham, J.R., Harvey, C.R., Michaely, R., 2005. Payout policy in the 21st century. Journal of

Financial Economics 77, 483-527.

Brav, A., Graham, J.R., Harvey, C.R., Michaely, R., 2008. Managerial response to the May 2003

dividend tax cut. Financial Management 37, 611-624.

Brown, G., Kapadia, N., 2007. Firm-specific risk and equity market development. Journal of Financial

Economics 84, 358–388.

Brown, J.R., Fazzari, S.M., Petersen, B.C., 2009. Financing innovation and growth: cash flow, external

equity, and the 1990s R&D boom. Journal of Finance 64, 151-185.

Brown, J.R., Petersen, B.C., 2009. Why has the investment-cash flow sensitivity declined so sharply?

Rising R&D and equity market developments. Journal of Banking and Finance, 33, 971-984.

Brown, J. R., Petersen, B.C., 2010. Cash holdings and R&D smoothing. Journal of Corporate Finance

forthcoming

Byoun, S., Moore, W., Xu, Z., 2008. Why do some firms become debt-free? Unpublished working paper,

Baylor University.

Carpenter, R.E., Petersen, B.C., 2002. Capital market imperfections, high-tech investment, and new

equity financing. Economic Journal 112, 54-72.

Chan, L., Lakonishok, J., Sougiannis T., 2001. The stock market valuation of research and development

expenditures. Journal of Finance 56, 2431-2456.

Chan, L.K.C., Lakonishok, J., Swaminathan, B., 2007. Industry classifications and return comovement.

Financial Analysts Journal 63, 56-70.

Chay, J.B., Suh, J., 2009. Payout policy and cash-flow uncertainty. Journal of Financial Economics 93,

88-107.

Damodaran, A., 2006. Damodaran on Valuation: Security Analysis for Investment and Corporate

Finance, 2nd Edition, Wiley.

Daniel, K., Grinblatt, M., Titman, S., Wermers, R., 1997. Measuring mutual fund performance with

characteristic-based benchmarks. Journal of Finance 52, 1035-1058.

DeAngelo, H., DeAngelo, L., Skinner, D.J., 2004. Are dividends disappearing? dividend concentration

and the consolidation of earnings. Journal of Financial Economics 72, 425–456.

DeAngelo, H., DeAngelo, L., Stulz, R.M., 2006. Dividend policy and the earned /contributed capital mix:

a test of the life-cycle theory. Journal of Financial Economics 81, 227-254.

DeAngelo, H., DeAngelo, L., 2007. Capital structure, payout policy, and financial flexibility.

Unpublished working paper, University of Southern California.

De Long, J. B., Shleifer, A., Summers, L.H., Waldmann, R.J., 1990. Noise trader risk in financial

markets. Journal of Political Economy 98, 703-738.

Denis, D.J., Sibilkov, V., 2009. Financial constraints, investment, and the value of cash holdings. Review

of Financial Studies, forthcoming.

Dittmar, A., 2008. Corporate cash policy and how to manage it with stock repurchases. Journal of

Applied Corporate Finance 20, 22-34.

126

Dittmar, A., Mahrt-Smith, J., 2007. Corporate governance and the value of cash holdings. Journal of

Financial Economics 83, 599-634.

Dittmar, A., Mahrt-Smith, J., Servaes, H., 2003. International corporate governance and corporate cash

holdings. Journal of Financial and Quantitative Analysis 38, 111–133.

Duchin, R., 2008. Cash holdings and corporate diversification. Unpublished working paper, University

of Michigan.

Duchin, R., Ozbas, O., Sensoy, B.A., 2009. Costly external finance, corporate investment, and the

subprime mortgage credit crisis. Unpublished working paper, University of Michigan.

Fama, E.F., French, K.R., 1993. Common risk factors in the returns on stocks and bonds. Journal of

Financial Economics 33, 3–56.

Fama, E.F., French, K.R., 1998. Taxes, financing decisions, and firm value. Journalof Finance 53, 819–

843.

Fama, E.F., French, K.R., 2001. Disappearingdividends: changing firm characteristics or lower

propensity to pay? Journal of Financial Economics 60, 3–43.

Fama, E.F., French, K.R., 2002. Testing trade-off and pecking order predictions about dividends and

debt. Review of Financial Studies 15, 1-33.

Fama, E.F., French, K.R., 2004. New lists: fundamentals and survival rates. Journal of Financial

Economics 73, 229–269.

Fama, E.F., MacBeth, J.D., 1973. Risk, return, and equilibrium: Empirical tests. Journal of Political

Economy 81, 607–636.

Faulkender, M., Wang, R., 2006. Corporate financial policy and the value of cash. Journal of Finance 61,

1957–1990.

Feenstra, R.C., 1996. U.S. imports 1972–1994: data and concordances. NBER Working Paper 5515.

Feenstra, R.C., 1997. U.S. exports 1972–1994: with state exports and other U.S. data. NBER Working

Paper 5990.

Feenstra, R.C., Romalis, J., Schott, P.K., 2002. U.S. imports, exports, and tariff data, 1989–2001. NBER

Working Paper 9387.

Fenn, G.W., Liang, N., 2001. Corporate payout policy and managerial stock incentives. Journal of

Financial Economics 60, 45-72.

Financial Accounting Standards Board (FASB), 1974. Accounting for research and development costs,

Statement of Financial Accounting Standards, No. 2. Stamford, CT.

Foley, C. F., Hartzell, J., Titman, S., Twite, G.J., 2007. Why do firms hold so much cash? A tax-based

explanation. Journal of Financial Economics 86, 579-607.

Frank, M.Z., Goyal, V., 2009. Capital structure decisions: which factors are reliably important? Financial

Management 38, 1-39.

Froot, K.A., Scharfstein, D., Stein, J., 1993. Risk management: coordinating corporate investments and

financing policies. Journal of Finance 5, 1629–1658.

Gaspar, J., Massa, M., 2006. Idiosyncratic volatility and product market competition. Journal of Business

79, 3125-3152.

127

Graham, B., Dodd, D.L., 1951. Security Analysis. McGraw-Hill Companies.

Graham, J., Harvey, C., 2001. The theory and practice of corporate finance: evidence from the field.

Journal of Financial Economics 60, 187-243.

Greenspan, A., 2002. Economic volatility. At a symposium sponsored by the Federal Reserve Bank of

Kansas City, Wyoming. http://www.federalreserve.gov/boarddocs/speeches/2002/20020830.

Grömping, U., 2006. Relative importance for linear regression in R: the package relaimpo. Journal of

Statistical Software 17, 1-27.

Grömping, U., 2007. Estimators of relative importance in linear regression based on variance

decomposition. The American Statistician 61, 139-147.

Grullon, G., Michaely, R., Swaminathan, B., 2002. Are dividend changes a sign of firm maturity?

Journal of Business 75, 387–424.

Grullon, G., Michaely, R., 2008. Corporate payout policy and product market competition. Unpublished

working paper, Rice University and Cornell University.

Hall, B.H., 2002. The financing of research and development. Oxford Review of Economic Policy 18,

35-51.

Hall, B.H., Lerner, J., 2009. The financing of R&D and innovation. NBER working paper, forthcoming

2010 in Hall, B. H. and N. Rosenberg (eds.), Handbook of the Economics of Innovation, Elsevier-

North Holland.

Han, S., Qiu, J., 2007. Corporate precautionary cash holdings. Journal of Corporate Finance 13, 43–57.

Harford, J., 1999. Corporate cash reserves and acquisitions. Journal of Finance 54, 1969-1997.

Harford, J., Mansi, S., Maxwell, W., 2008. Corporate governance and firm cash holdings. Journal of

Financial Economics 87, 535-555.

Helpman, E., Krugman, P., 1985. Market Structure and Foreign Trade. MIT Press, Cambridge, MA.

Helpman, E., Krugman, P., 1989. Trade Policy and Market Structure. MIT Press, Cambridge, MA.

Herfindahl, O.C., 1950. Concentration in the U.S. steel industry. Unpublished doctoral dissertation,

Columbia University.

Hertzel, M.G., Li, Z., 2009. Behavioral and rational explanations of stock price performance around

SEOs: Evidence from a decomposition of market-to-book ratios. Forthcoming, Journal of

Financial and Quantitative Analysis.

Hoberg, G., Prabhala, N., 2009. Disappearing dividends: the importance of idiosyncratic risk and the

irrelevance of catering. Review of Financial Studies 22, 79-116.

Hou, K., Robinson, D., 2006. Industry concentration and average stock returns. Journal of Finance 61,

1927-1956.

Irvine, P.J., Pontiff, J., 2009. Idiosyncratic return volatility, cash flows and product market competition.

Review of Financial Studies 22, 1149-1177.

Kecskes, A., 2008. Why are firms that raise more financing worth more? Unpublished working paper,

University of Toronto.

Keynes, J.M., 1936. The General Theory of Employment, Interest, and Money. Harcourt, London.

128

Kim, C.S., Mauer, D.C., Sherman, A.E., 1998. The determinants of corporate liquidity: theory and

evidence. Journal of Financial and Quantitative Analysis 33, 305-334.

Lemmon, M.L., Roberts, M.R., Zender, J.F., 2008. Back to the beginning: persistence and the cross-

section of corporate capital structure. Journal of Finance 63, 1575-1608.

Lerner, A. P., 1934. The concept of monopoly and the measurement of monopoly power. Review of

Economic Studies 1, 157–75.

Lev, B., Sougiannis T., 1996. The capitalization, amortization, and value relevance of R&D. Journal of

Accounting and Economics 21, 107-138.

Li, K., Zhao, X., 2008. Asymmetric information and dividend policy. Financial Management 37, 673-

694.

Lintner, J., 1956. Distribution of incomes of corporations among dividends, retained earnings, and taxes.

The American Economic Review 46, 97-113.

Loughran, T., Ritter, J., 2004. Why has IPO underpricing changed over time? Financial Management 33,

5–37.

MacKay, P., Phillips, G., 2005. How does industry affect firm financial structure? Review of Financial

Studies 18, 1433-1466.

McLean, R. D., 2010. Share issuance and cash savings. Unpublished working paper, University of

Alberta.

Modigliani, F., Miller, M., 1958. The cost of capital, corporation finance and the theory of investment.

American Economic Review 48, 261–297.

Myers, S.C., 1984. The capital structure puzzle. Journal of Finance 39, 575-592.

Myers, S.C., Majluf, N.S., 1984. Corporate financing and investment decisions when firms have

information that investors do not have. Journal of Financial Economics 13, 187-221.

Newey, W., West, K., 1987. A simple positive semi-definite, heteroscedasticity and autocorrelation

consistent covariance matrix. Econometrica 55, 703–708.

Opler, T., Pinkowitz, L., Stulz, R.M., Williamson, R., 1999. The determinants and implications of cash

holdings. Journal of Financial Economics 52, 3-46.

Passov, R., 2003. How much cash does your company need? Harvard Business Review, 119-128.

Peress, J., 2010. Product market competition, insider trading and stock market efficiency. Journal of

Finance 65, 1-43.

Petersen, M.A., 2009. Estimating standard errors in finance panel data sets: comparing approaches.

Review of Financial Studies 22, 435-480.

Pinkowitz, L., Stulz, R.M., Williamson, R., 2006. Do firms in countries with poor protection of investor

rights hold more cash? Journal of Finance 61, 2725-2751.

Pinkowitz, L., Williamson, R., 2007. What is a dollar worth? The market value of cash holdings. Journal

of Applied Corporate Finance 19, 74-81.

Ritter, J., Welch, I., 2002. A review of IPO activity, pricing, and allocations. Journal of Finance 57,

1795–1828.

129

Schmalensee, R., 1989. Inter-industry studies of structure and performance. In Richard Schmalensee and

Robert Willig, (eds.), Handbook of Industrial Organization, Amsterdam: North-Holland.

Schroth, E., Szalay, D., 2009. Cash Breeds Success: The Role of Financing Constraints in Patent Races.

Review of Finance forthcoming.

Shleifer, A., Vishny, R., 1997. The limits of arbitrage. Journal of Finance 52, 35–55.

Skinner, D.J., 2008. The evolving relation between earnings, dividends, and stock repurchases. Journal

of Financial Economics 87, 582-609.

Strebulaev, I.A., Yang, B., 2007. The mystery of zero-leverage firms. Unpublished working paper,

Stanford University.

Subrahmanyam, M. G., Thomadakis, S. B., 1980. Systematic risk and the theory of the firm. Quarterly

Journal of Economics 94, 437-451.

Sullivan, T. G., 1978. The cost of capital and the market power of firms. Review of Economics and

Statistics 60, 209–217.

Tong, Z., 2009. Firm diversification and the value of corporate cash holdings. Journal of Corporate

Finance, forthcoming

Tybout, J. R., 2003. Plant- and firm-level evidence on „new‟ trade theories. In: E. K. Choi and J.,

Harrigan (Eds.), Handbook of International Trade, Basil-Blackwell, Oxford.

Welch, I., 1989. Seasoned offerings, imitation costs, and the underpricing of initial public offerings.

Journal of Finance 44, 421–449.

Welch, I., 1996. Equity offerings following the IPO theory and evidence. Journal of Corporate Finance 2,

227–259.

Xu, J., 2008. The product market competition and capital structure: evidence from import penetration,

Working Paper, Purdue University.

Zhou, J., 2009. Increase in cash holdings: pervasive or sector-specific? Unpublished working paper,

University of Toronto.