ARTICLE IN PRESS

Contents lists available at ScienceDirect

Journal of Financial Economics

Journal of Financial Economics 92 (2009) 40–65

0304-40

doi:10.1

$ For

Gang H

Susan S

particip

Confere

Shangh

tricht, t

particip

versity.� Cor

E-m

journal homepage: www.elsevier.com/locate/jfec

Product market advertising and new equity issues$

Thomas Chemmanur a,�, An Yan b

a Carroll School of Management, Boston College, Chestnut Hill, MA 02467, USAb Graduate School of Business Administration, Fordham University, New York, NY 10023, USA

a r t i c l e i n f o

Article history:

Received 10 January 2006

Received in revised form

5 December 2007

Accepted 28 February 2008Available online 3 December 2008

JEL classification:

G32

Keywords:

Advertising

Initial public offerings

Seasoned equity offerings

IPO underpricing

5X/$ - see front matter & 2008 Elsevier B.V.

016/j.jfineco.2008.02.009

helpful comments or discussions, we than

u, Blake LeBaron, Holger Muller, Debarshi Nan

hu, Narayanan Subramanian, and James W

ants at the 2004 Financial Intermediation

nce at Capri, the 2004 China Internatio

ai, the 2004 European Finance Association

he 2006 Frontiers of Finance Conference at Bo

ants at Boston College, Brandeis University

We alone are responsible for any errors or om

responding author.

ail address: chemmanu@bc.edu (T. Chemman

a b s t r a c t

We analyze the interaction between a firm’s product market advertising and its

corporate financing decisions. We consider a firm that faces asymmetric information in

both the product and financial markets and that needs to raise external financing to

fund its growth opportunity (new project). Any product market advertising undertaken

by the firm is visible to the financial market as well. In equilibrium, the firm uses a

combination of product market advertising, equity underpricing, and underfinancing

(raising a smaller amount of external capital than the full information optimum) to

convey its true product quality and the intrinsic value of its projects to consumers and

investors. The following two predictions arise from our theoretical analysis for the

relation between product market advertising and equity underpricing around new

equity issues. First, firms choose a higher level of product market advertising when they

are planning to issue new equity, compared with situations in which they have no

immediate plans to do so. Second, product market advertising and equity underpricing

are substitutes for a firm issuing new equity. We empirically test the above two

predictions and find supporting evidence in the context of firms making initial public

offerings and seasoned equity offerings.

& 2008 Elsevier B.V. All rights reserved.

1. Introduction

The role of the underpricing of initial public offerings(IPOs) in signaling firm insiders’ private information to theequity market has been extensively analyzed (see, e.g.,Allen and Faulhaber, 1989; or Welch, 1989). However,recently, some authors have questioned whether under-pricing is the most efficient way to signal firm value, and

All rights reserved.

k Sonia Falconieri,

dy, Imants Paeglis,

eston, as well as

Research Society

nal Conference at

Meetings at Maas-

naire, and seminar

, and Tilburg Uni-

issions.

ur).

they have raised the possibility that it could be moreefficient for firms to use other signals around new equityissues. For example, Ritter and Welch (2002) comment intheir review of the IPO literature: ‘‘On theoretical grounds,however, it is unclear why underpricing is a more efficientsignal than, say, advertising.’’

Advertising is a particularly interesting signalingalternative to underpricing, because some anecdotalevidence exists that, in practice, some managers couldattempt to convey their firm’s intrinsic value to thefinancial market by making use of product marketadvertising (particularly in the context of an upcomingIPO). Consider, for example, this quote (Wall StreetJournal, 1999): ‘‘As they plaster ads everywhere consu-mers might turn, companies are hoping to catch investors’eyes too. Businesses are often as interested in selling stockas in selling products; a high voltage advertising spreecould serve as a critical prelude to an initial publicoffering.’’ Another article (Boston Globe, 2000) deals with

ARTICLE IN PRESS

T. Chemmanur, A. Yan / Journal of Financial Economics 92 (2009) 40–65 41

television advertising during the Super Bowl: ‘‘Hopingto impress Wall Street as well as fans, advertiserswith names such as Pets.com, Lifeminders.com, andOurbeginning.com will pony up at least $2.2 million foreach 30 second slot on next Sunday’s football game. SkipPile, of Pile and Co., a Boston firm that helps companieshire ad agencies, offers the following appraisal: ‘A SuperBowl ad can legitimize a brand among multiple constitu-encies: the company’s employees, venture capitalists, theinvestment community, and lastly, the target audience offootball fans.’’’ Whether these and similar anecdotesreflect the special situation, during a special time period,of only a few companies (e.g., Internet firms going publicduring the bubble period), or whether they reflect thegeneral situation of firms making equity issues (either anIPO or a seasoned equity offering (SEO)), is an empiricalquestion that has been unanswered so far in the literature.

The objective of this paper is to explore how the extentof product market advertising undertaken by a firm couldaffect (and be affected by) the prospect of an upcomingnew equity issue. The economic environment that weanalyze is a setting where insiders of a firm, withinformation about its intrinsic value superior to outsiders,raise external financing to fund a positive net presentvalue (NPV) project by making a new equity issue.We address several related questions in the above setting.First, how will a firm choose the extent of its productmarket advertising in a setting where this advertising isvisible to the financial market as well as the productmarket? Second, will the equilibrium level of advertisingchosen by a firm be different in situations in which thefirm plans to make a public offering of new equity (eitheran IPO or an SEO) compared with a situation in which ithas no immediate plans to make such an offering of equityor other financial assets?1 Third, of three alternativesignals easily available to firm insiders, namely, under-pricing, advertising, and underfinancing (raising a smalleramount of equity than the optimal amount in a fullinformation setting), under what conditions will eachsignal be employed (either individually or in combinationwith the other signals)? In particular, are advertising andunderpricing substitutes for a firm in the context of newequity issues? While our primary objective in this paper isto answer these questions empirically, we first brieflydevelop a theoretical analysis that generates testablehypotheses that we subsequently test. We empiricallyshow how firms alter the extent of their advertising in thecontext of an upcoming new equity issue and how theextent of advertising relates to the extent of underpricingin IPOs as well as SEOs.

We consider a firm that has an existing product, anongoing project, and a growth opportunity (new project).Firm insiders have private information not only about the

1 For concreteness, we present much of our theoretical analysis in

the context of a private firm raising external capital by making an IPO of

equity. However, our analysis goes through with minor modifications for

the case of a publicly traded firm making a seasoned issue of equity or

other information-sensitive financial assets. Our theory therefore has

implications for these situations as well, and we test the implications of

our model not only for IPOs, but also for SEOs.

quality of the firm’s product, but also about the true valueof its projects. In other words, the firm faces asymmetricinformation in both the product and financial markets.While the firm has some internal capital available, thiscapital is not adequate to cover both the investmentrequired in its ongoing project and to fund its growthopportunity. The firm therefore needs to raise externalfinancing for investment by making a new equity issue. Weassume that any product market advertising undertakenby the firm is visible to financial market investors as well.

In this setting, product market advertising can bethought of as playing two different roles. The first roleplayed by advertising is that of signaling quality tothe product market, thereby allowing consumers to pricethe firm’s products correctly in equilibrium. This productmarket role of advertising in our analysis is similar to therole played by advertising in the industrial organizationliterature, which has long argued that product marketadvertising can help convey information about productquality to consumers (see, e.g., Nelson, 1974; Kihlstromand Riordan, 1984; or Milgrom and Roberts, 1986). In oursetting, however, product market advertising plays asecond role; that of signaling the true value of a firm’sprojects to potential stock market investors, thus allowingthem to price the firm’s equity correctly in equilibrium.Because a firm’s product quality and the value of itsprojects might not be perfectly correlated, an outsider(consumer or potential investor) who knows only the truequality of a firm’s existing product cannot infer the truevalue of its projects. Conversely, an outsider who knowsonly the true value of its projects cannot infer existingproduct quality. However, the firm needs not to useproduct market advertising alone as a signal, either to theproduct market or to the financial market. In a settingwhere the firm interacts with the equity market as well asthe product market, it can also signal by underpricingequity in its new equity issue or by underfinancing.

In equilibrium, the firm uses the least-cost combina-tion of the three signals to convey its product quality andproject value to outsiders. The equilibrium choice ofsignaling mix by the firm depends on the extent ofasymmetric information facing the firm and the internalcapital available to it. First consider when the extent ofasymmetric information facing the firm is relatively small.In this case, firms with superior quality products andhigher intrinsic value projects (which we refer to as highertype firms) use underfinancing alone as a signal, because,by itself, underfinancing is a less costly signal for thehigher type firm to use than either advertising or under-pricing. While underfinancing requires the higher typefirm to scale back its investment in its growth opportu-nity, the cost of this underinvestment is partially offset bythe reduced dilution in insiders’ equity holdings (whichresults from its raising a smaller amount of externalfinancing).2 Therefore, if the extent of underfinancing

2 By dilution, we refer to the fact that, when a firm sells equity at a

lower price, insiders have to give up a greater share of the firm’s equity to

new investors in return for external financing. If the firm raises only a

smaller amount of external financing, this dilution in insiders’ equity

holdings is smaller.

ARTICLE IN PRESS

T. Chemmanur, A. Yan / Journal of Financial Economics 92 (2009) 40–6542

required to deter mimicking by lower type firms (thosewith either inferior products, lower intrinsic valueprojects, or both) is small enough that the highertype firm has to sacrifice investment only in theless productive ranges of its growth opportunity, then itcan be shown that the firm uses only underfinancing as asignal.

Now consider when the extent of asymmetric informa-tion facing the firm is more severe, so that, if the firm wereto use only underfinancing as a signal, the reduction ininvestment required would be significant enough that thefirm would have to sacrifice investment in the higherproductivity ranges of its growth opportunity. In this case,the cost of signaling by underfinancing alone becomesprohibitive, and the higher type firm can lower itsaggregate signaling cost by adding either advertising,underpricing, or both, to its signaling mix. The firm’schoice between advertising and underpricing as the signalto add to underfinancing depends upon the internalcapital available to it prior to the equity issue. Toadvertise, the higher type firm needs to cut back oninvestment in its ongoing project, thereby reducing firmvalue and diluting insiders’ equity holdings. If the internalcapital available to the firm is large enough that onlyinvestment in the lower productivity ranges of theongoing project has to be sacrificed to fund the requiredamount of advertising, then it can be shown that the firmadds only advertising to the signaling mix. If, however, theinternal capital available to the firm is smaller, so that thefirm has to sacrifice investment in the higher productivityranges of its ongoing project to undertake the requiredamount of advertising, then minimizing the signaling costinvolves the firm adding both advertising and IPO under-pricing to the signaling mix, thus using all three signals toconvey its true type to outsiders.3

Our paper is the first in the literature to analyze howproduct market advertising can serve as a signal to thefinancial market in the context of new equity issues. It isalso the first to study the interaction between product

3 The relative cost of underfinancing, advertising, and underpricing

as signals depends upon how expensive it is for the higher type firm to

undertake each activity, relative to the cost to lower type firms

undertaking the same activity to a similar extent (their cost of

mimicking). When the extent of asymmetric information facing the firm

is relatively small, underfinancing is used alone as a signal, because the

cost of a given extent of underfinancing is lower for a higher type firm

compared with the same cost for a lower type firm that seeks to mimic it.

In contrast, if a higher type firm funds advertising by cutting back only

on investment in the lower productivity ranges of its ongoing project,

then the cost of advertising for the higher type firm is similar in

magnitude to the cost for a lower type firm that seeks to mimic the

higher type firm. Thus, advertising alone is a costlier signal compared

with underfinancing in our setting. Finally, for underpricing alone to

serve as a signal, the higher type firm has to price equity in its IPO below

the intrinsic value of the equity of the lower type firm (because,

otherwise, the lower type would not incur any cost from mimicking).

Given this, the cost to a higher type firm of diluting insiders’ equity

holdings arising from underpricing is always greater than the cost to a

lower type firm that seeks to mimic the higher type firm. This means

that underpricing alone is a costlier signal than both underfinancing and

advertising.

market advertising and underpricing in the context ofeither IPOs or SEOs and the conditions under whichvarious combinations of advertising and underpricing areemployed by firms to signal their private information.Thus, we provide insight into the ongoing debate in theIPO literature regarding the efficiency of using under-pricing as a signal. Our analysis indicates that it is efficientfor firms that are financially constrained to include bothunderpricing and advertising in their signaling mix. Ourpaper makes an additional contribution by demonstratinghow underpricing can allow the firm to signal qualitymore effectively to the product market by reducing theamount of advertising required.

Two testable predictions emerge from our theoreticalanalysis for product market advertising in the context ofnew equity issues, as well for equity underpricing (IPOs orSEOs). The first prediction is that firms choose a higherlevel of product market advertising when they areplanning to issue new equity compared with situationsin which they have no immediate plans to sell new equity.The second prediction is that, in the context of the newequity issue, product market advertising and equityunderpricing are substitutes. The greater the extent ofproduct market advertising, the lower the extent ofunderpricing.

We test the two predictions using advertising (andother product market) data and the financial market datafrom firms making new equity issues. In particular, westudy two different samples of firms making new equityissues: a sample of firms going public and a sample offirms making SEOs. We test the first prediction bystudying the change in a firm’s advertising level inresponse to its plan to issue new equity in the financialmarket. In this test, we first compare a firm’s advertisinglevel in its IPO or SEO year with its advertising level innon-IPO and non-SEO years (years when the firm had noimmediate plans to make an equity offering). We find thata firm’s advertising expenditures in its IPO or SEO year aregreater than those in its non-IPO and non-SEO years, evenafter controlling for product market considerations. In thefive-year span around the equity issue year (from twoyears before the equity issue to two years after), the peakadvertising level is reached in the year of the new equityissue. We also construct a matching sample of non-equityissuing firms, both for our IPO and for our SEO sample,using the matching algorithm of Loughran and Ritter(1997). We find that a firm’s advertising expenditures inits IPO year or SEO year are higher than the advertisingexpenditures of its matching firm. In contrast to the casein IPO and SEO years, the advertising expenditures of thefirms in our IPO and SEO samples in non-IPO and non-SEOyears are generally similar to the advertising expendituresof their matching firms. Thus, our findings are consistentwith the first prediction. They confirm that firms increasetheir product market advertising in the years when theyplan to issue new equity. They also suggest that suchincreases are related to firms’ decisions to issue newequity, instead of being driven by product marketconsiderations.

We test the second prediction by studying thesubstitution effect between advertising and underpricing

ARTICLE IN PRESS

5 The broader theoretical literature on the going public decision (e.g.,

Chemmanur and Fulghieri, 1999) and the broader theoretical literature

on IPO underpricing (e.g., Chemmanur, 1993) are also indirectly related

to this paper. Our paper is also indirectly related to the literature on

product and financial market interactions outside the context of equity

T. Chemmanur, A. Yan / Journal of Financial Economics 92 (2009) 40–65 43

for firms making IPOs and SEOs. In addition to an OLSregression analysis, we present results from seeminglyunrelated regressions (SURE) to address any potentialsimultaneity issues involved in a firm’s choice of adver-tising expenditures and the pricing of its new equity.Consistent with the second prediction, we find thatproduct market advertising and underpricing are sub-stitutes. The extent of underpricing in a firm’s IPO or SEOis smaller as the extent of its product market advertising isgreater. To further address the endogeneity and simulta-neity issues, we also try other econometric techniquessuch as two-stage least squares regressions (2SLS) andthree-stage least squares regressions (3SLS). Our resultsfrom both the 2SLS and 3SLS techniques are similar tothose presented here. They are available in the workingpaper version of this article.

Our finding that advertising and underpricing aresubstitutes is particularly interesting in light of behavioralnotions that product market advertising around a firm’sIPO might generate a media buzz among retail investorsregarding the IPO, inducing them to buy the IPO firm’sequity at a high price on the first trading day (thusgenerating excess demand for it and further driving upits price). However, such behavioral arguments wouldpredict that a higher level of advertising around a firm’sIPO will be associated with larger initial returns(underpricing) for its equity, which is directly contra-dicted by our empirical finding that advertisingand underpricing are substitutes.4 In sum, while ourtheoretical analysis predicts that the product marketadvertising of a firm affects its equity issues in importantways even in an equity market dominated by rationalinvestors, our empirical results indicate that these effectsare consistent with the predictions of our theory. In otherwords, firm managers do take into account the effect oftheir product market advertising on the financial marketwhen making their advertising and corporate financingdecisions.

Our research is related to the theoretical literature onthe use of underpricing to signal insiders’ privateinformation at the time of an IPO. The existing literature(e.g., Allen and Faulhaber, 1989; Welch, 1989) hasdemonstrated that underpricing can signal insiders’private information about firm value to the financialmarket. In this literature, IPO underpricing works as asignal only because insiders price equity in the IPO inanticipation of a second round of financing subsequent tothe IPO and the possibility of true firm value beingrevealed exogenously between the two rounds of finan-cing. In contrast to the above literature, here IPO under-pricing serves as a signal in the context of a one-shotequity offering.

Our paper is also related to the theoretical literature onthe interaction between the product and financial marketsin the context of IPOs. See, e.g., Bhattacharya and Ritter(1983), who point out that one cost of going public arises

4 We thank an anonymous referee for pointing out these behavioral

arguments and the fact that our empirical findings contradict the

implications of these arguments.

from the need to release information in a firm’s IPOprospectus, and Maksimovic and Pichler (2001), whostudy how the possibility of such an information releaseto product market competitors could affect the timing of afirm’s going public decision. Stoughton, Wong, andZechner (2001) further argue that the decision of a firmto go public could serve to signal high quality to theproduct market.5 None of the above papers, however,addresses the role of product market advertising in IPOs(or SEOs). They also do not study how product marketadvertising interacts with equity underpricing.

Finally, our research is related to the large empiricalliterature on IPO underpricing (see Ritter and Welch,2002, for a review).6 Two empirical papers related to oursare Grullon, Kanatas, and Weston (2004) and Demers andLewellen (2003). The former paper shows that firms witha greater level of product market advertising have asignificantly larger number of both individual and institu-tional investors holding their equity, lower bid-askspreads (indicating a smaller amount of adverse selectionin the market for these stocks), smaller price impacts, andgreater market depth. In contrast to Grullon, Kanatas, andWeston (2004), who focus on the behavior of stock marketinvestors in response to increased product marketadvertising by a firm, our empirical analysis focuses onhow firms choose their advertising levels in the context ofa new equity issue. However, our results suggest thatproduct market advertising plays a role in conveyingfirm insiders’ private information to equity marketinvestors and are thereby consistent with the findings inGrullon, Kanatas, and Weston (2004) on the reduction inadverse selection associated with a greater level ofproduct market advertising. Demers and Lewellen(2003) show that, for a sample of Internet firms, IPOunderpricing is associated with a post-IPO increase intheir website traffic. Unlike their paper, where the focus ison how Internet firms’ IPOs affect their product marketdemand, our empirical analysis focuses on how theprospect of a firm’s IPO affects its choice of advertisinglevel and how its choice of product market advertisingaffects underpricing in its IPO.

The rest of this paper is organized as follows. Toprovide a theoretical framework for our empirical analysisand develop testable hypotheses, Section 2 and Section 3briefly develop a theoretical analysis of a firm’s choice ofsignaling mix in a setting of asymmetric information inthe product and financial markets. Section 4 discusses thesample selection and the econometric methodology usedin our empirical analysis. Section 5 presents our empirical

issues. See, e.g., Gertner, Gibbons, and Scharfstein (1988), who analyze an

informed firm’s choice of financial structure when the financing contract

is observed not only by the capital market but also by a competing firm.6 Substantial empirical literature demonstrates the underpricing of

SEOs, which is also indirectly related to this paper. See, e.g., Corwin

(2003).

ARTICLE IN PRESS

1 2 3

Firm determines the amount of advertising to be undertakenFirm invests in the existing project, using its internal capital

Firm goes public in the stock market, determining the amount of capital raised and the pricing of equity in the IPO Firm implements its investment in the new project using the external capital raised from the IPO and the internal capital left over from time 0 First round of sales of the firm’s product

Second round of sales of the firm’s product

All cash flows are realized including the cashflows from the two rounds of sales in the productmarket and that from the investments in the twoprojectsAll the asymmetric information in the productand financial markets is resolved

t = 0

Interim signal of product quality is received

Fig. 1. Sequence of events.

T. Chemmanur, A. Yan / Journal of Financial Economics 92 (2009) 40–6544

results for IPO firms, and Section 6 presents our empiricalresults for SEO firms. Section 7 discusses potentialalternative explanations that could explain some of ourempirical findings. Section 8 concludes.

2. Theoretical framework

To provide a theoretical context for our empirical testsand develop testable hypotheses, we now briefly describethe essential features of the economic environmentstudied in the paper. Consider an entrepreneur owning aprivate firm, which has available to it a certain amount ofinternal capital. For concreteness, we conduct our discus-sion in the context of a firm going public, though ourresults apply to firms making SEOs as well. Our model hasfour dates (time 0, 1, 2, and 3). At time 0, the firm has anexisting positive NPV project that it plans to fund using(part of) this internal capital.7 For simplicity, we assumethat issuing equity is the only source of external financing,so that firms cannot fund their projects by issuing debt orother financial assets. The firm has an existing product,which it plans to sell in the product market in two roundsof sales, at time 1 and time 2.

A new positive NPV project (growth opportunity)becomes available to the firm at time 1. To fund thisnew project, the firm has the opportunity to go public attime 1. If it chooses to go public, outsiders come to knowthis decision right away. The firm chooses the price of theequity to be issued in the IPO, as well as the amount of theexternal capital to be raised in the IPO. If the firm goes

7 Our assumption of an existing project merely serves to capture the

notion that the firm has other ongoing business activities for which it

also requires funding. In other words, the assumption of an existing

project allows us to endogenize the fact that the firm has an opportunity

cost of internal funding, with this opportunity cost increasing as the

amount of internal financing available is smaller. All our results go

through even in the absence of an existing project if we make the above

assumption about the opportunity cost of the firm’s internal financing

exogenously.

public at time 1, it invests the external capital raised, inaddition to any amount left over from its internal capital,in the new project. If, however, the firm does not go publicat time 1, it invests only the leftover capital in the newproject at time 1.

At time 0, the firm chooses the amount of investmentto be made in the existing project and the amount ofproduct market advertising to undertake. Any advertisingthe firm undertakes in the product market is observableby potential consumers in the product market and bypotential investors in the financial market.

All cash flows to the firm are realized at time 3including the cash flows from the two rounds of sales ofthe firm’s product and that from the firms’ investments inits two projects. We assume that all agents (entrepre-neurs, consumers, and investors) are risk-neutral and thatthe risk-free rate of return is zero. The sequence of eventsis depicted in Fig. 1.

2.1. Information structure in the product and financial

markets

Both the product and financial markets are character-ized by asymmetric information. In particular, we assumethat there are products of two quality levels in the productmarket: superior and poor, with the quality of superiorproducts being higher than that of poor products. We alsoassume that there are two kinds of projects: good and bad.For any given level of investment, good projects yieldhigher cash flows than bad projects (i.e., the NPV of goodprojects is greater than that of bad projects at any giveninvestment level). See Fig. 2. While firm insiders know thetrue quality of the firm’s product as well as the NPV of itsprojects, outsiders (be they consumers in the productmarket or investors in the financial market) observe onlythe prior probability distributions over product andproject types.

To capture the above asymmetric information in thesimplest possible manner, we assume that there are threetypes of firms: H (high), M (medium) and L (low).

ARTICLE IN PRESS

NPV of marginal investment, B (I)-1

NPV of marginal investment, G (I)-1

Amount ofinvestment

NPV

i1

NPV of marginal investment, 0

NPV of marginal investment, 0

The type G project

The type B project

Fig. 2. Investment technologies for good and bad projects.

T. Chemmanur, A. Yan / Journal of Financial Economics 92 (2009) 40–65 45

We assume that a type H firm always has superior qualityproducts as well as good projects; a type M firm has poorquality products but good projects; a type L firm has poorquality products and bad projects.8 While firm insidersobserve the type of their own firm, outsiders observe onlythe prior probability distribution across firm types attime 0. At time 3, all cash flows are realized so that theasymmetric information between firm insiders and out-siders is resolved completely.

2.2. The firm’s product and the product market

We assume that quality is not a choice variable for thefirm. Each firm is endowed with technology that allows itto manufacture products only of a given quality. In theabsence of asymmetric information, consumers valuesuperior quality products higher than poor qualityproducts. Each kind of product is sold in the productmarket for two rounds, at time 1 and time 2. Betweentime 1 and time 2, after the first round of sales of thefirm’s product but before the second round, a signal ofproduct quality becomes publicly available. We can thinkof this signal as a (probabilistic) breakdown of the firm’sproduct. We assume that the probability of breakdown fora poor quality product is higher than that for a superiorquality product. We also assume that the demand for thefirm’s products is perfectly elastic and that the productioncapacity of the firm is not variable. The production costsfor superior and poor quality products are assumed to bethe same and normalized to zero for analytical simplicity.

8 Thus, we assume that the product quality and project value are

correlated, but not perfectly so. Clearly, assuming that both product

quality and project value are perfectly correlated is somewhat less

interesting, because, in that case, if product quality is revealed, project

value is completely known as well. Our assumption of these types of

firms allows us to introduce an imperfect but positive correlation

between product quality and project value in a parsimonious manner.

(This would not be possible in a two-type model.) Our results are not

driven by the correlation between product quality and project value. At

the expense of some additional modeling complexity, it can be shown

(using a four-type model) that our results hold even when product

quality and project value are completely uncorrelated.

2.3. The firm’s projects and the financial market

We assume that the payoff of good projects is G(I) andthe payoff of bad projects is B(I), where I is firminvestment in a project. Both G(I) and B(I) are commonknowledge. We also make the following assumptions onthe investment technology for good and bad projects.First, we assume that the investment in both types ofprojects are of positive NPV and has a decreasing marginalreturn to scale, G0(I)41, B0(I)41,G00(I)o0, and B00(I)o0 forIA[0, i1). Second, G0(i1) ¼ B0(i1) ¼ 1. Thus, we can think ofi1 as the full investment level (in the sense that, for eitherkind of project, if the investment made is below i1, thenNPV is wasted). Third, the NPV of marginal investment ishigher in good projects than in bad projects, G0(I)4B0(I) forIA[0, i1). Thus, G(I)/B(I) is increasing in IA[0, i1). Fourth, weassume that the relative marginal profits G0(I)/B0(I) isdecreasing in investment IA[0, i1). Finally, we assume that,when investment is small, the difference in marginalprofit between good and bad projects is larger than thedifference in product quality between superior andinferior products. We depict the investment technologyfor firms with both good and bad projects in Fig. 2.

The firm has two projects. Both projects of any givenfirm are of the same type.9 We assume that the internalcapital owned by each firm before going public is belowthe full investment level i1, so that any firm can finance atmost only one project (either the existing or the newproject) to its full investment level by using internalcapital. Thus, if the firm wishes to fund both its existingand its new project to the full investment level, it has toundertake an IPO at time 1. In this case, the firm funds itsnew project partially or fully using the external capitalraised from its IPO, depending on whether any funds fromits internal capital are left over after investing in theexisting project. Thus, the amount of the internal capital

9 This assumption is made only for convenience. Our results are

driven only by the asymmetric information about the new project, and

are therefore qualitatively unchanged in the absence of the asymmetric

information about the firm’s existing project.

ARTICLE IN PRESS

T. Chemmanur, A. Yan / Journal of Financial Economics 92 (2009) 40–6546

available to the firm is a measure of the degree of thefinancial constraint faced by the firm prior to its IPO.

2.4. The firm’s objective

The firm’s objective is to maximize the expected long-term (time 3) value of equity held by the entrepreneur(or, equivalently, the present value of expected total cashflows to the entrepreneur). The firm’s revenue arises fromtwo sources: the two rounds of sales of its product, andthe revenue stream from investment in its existing project(at time 0) and in its new project (at time 1). Thus, theentrepreneur strategically chooses the firm’s expenditureon advertising, the amount of external capital raised bygoing public, the amount of investment in its projects, andthe price of equity in the firm’s IPO to maximize thisobjective.

3. Equilibrium choice of signaling mix and testablehypotheses

Equilibrium strategies and beliefs in our model aredefined as separating equilibria constituting a Paretodominant or efficient Perfect Bayesian Equilibrium(PBE).10 Due to space limitations, we discuss here onlyintuitively the equilibrium choices of signals by varioustypes of firm in different situations, focusing primarily onthe type H firm. The detailed characterization and formalderivations of the equilibria in our model are provided inthe working paper version of this article (available fromthe authors’ websites).

3.1. Trade-offs driving the equilibrium

The objective of the type H firm is to maximize theexpected long-term (time 3) value of the equity held bythe entrepreneur. This is accomplished by maximizing thetype H’s value while minimizing the dilution in theentrepreneur’s equity holdings in the firm. This, in turn,is achieved by ensuring that the firm is able to raise theoptimal amount of external capital, while separating itselffrom lower type (type M and type L) firms. The type H canuse three alternative signals to accomplish this: (1) raisingless capital externally than the amount that would beraised in a full information setting (underfinancing), withthe resulting underinvestment in its new project;11 (2)expending resources on advertising; and (3) underpricingthe firm’s equity in its IPO.

10 In other words, the equilibrium concept we use is a Perfect

Bayesian Equilibrium that minimizes the dissipative costs of separation

incurred. See Milgrom and Roberts (1986) or Engers (1987) for a detailed

discussion of why the notion of a Pareto dominant or efficient PBE is the

appropriate equilibrium concept here. See Fudenberg and Tirole (1991)

for a formal definition of a PBE.11 We assume that the amount raised by the firm in its IPO is

publicly observable, as is the case in practice. Given the observability of

the amount raised, outsiders can also infer the investment level chosen

by the firm. Therefore, throughout this paper, we use the terms

‘‘underfinancing’’ and ‘‘underinvestment’’ interchangeably.

To understand the trade-offs driving the type H firm’sequilibrium choice of signals, it is useful to study the costof each signal separately. Consider first the case in whichthe type H firm attempts to use underfinancing in the IPOmarket alone as a signal. It has to cut back on investmentin its new project, thus losing part of its value. If the typeM or type L firm chooses to mimic the type H, it also has toincur a similar loss in value as the type H firm. However,this cost in firm value due to underfinancing is partiallyoffset by the reduced dilution in the entrepreneur’s equitythat results from the firm raising a smaller amount ofexternal financing. The benefit from this reduction indilution is greater for the type H firm (because its intrinsicvalue is greater) than for the type M or the type L firm,thus allowing the type H to use underfinancing as a signal.

Consider now the case in which the type H firmattempts to use advertising alone as a signal. To advertise,the type H firm needs to cut back on investment in itsexisting project, thereby reducing firm value. This loss invalue also results in an increase in the dilution of theentrepreneur’s equity holdings in the firm. If the type M ortype L firm attempts to mimic the type H, it also has toincur a loss in firm value of similar magnitude to thetype H. However, given that the intrinsic value of the typeH is greater than that of the type M or type L, this dilutionin insiders’ equity holdings always imposes a greatercost on the type H firm compared with that on the type Mor type L firm. Thus, advertising alone is a costlier signalcompared with underfinancing, so that it enters theequilibrium signaling mix only in combination withunderfinancing.

Finally, consider the case in which the type H attemptsto signal using IPO underpricing alone. Given that theunderpricing of a firm is not directly observable byoutsiders (only the IPO share price is observable), thetype H can use underpricing as a signal only by pricing itsequity below the intrinsic value of the type M firm (if itwishes to prevent the type M from mimicking) or belowthe intrinsic value of the type L firm (if it wishes toprevent both the type M and the type L from mimicking).Further, given the type H firm’s higher intrinsic value, thecost arising from the dilution in insiders’ equity holdings(resulting from underpricing) can be shown to be greaterfor the type H firm than the corresponding cost to thetype M and the type L firms if they attempt to mimic thetype H. This means that underpricing alone is a costliersignal to the type H than either underfinancing oradvertising, so that it enters the equilibrium signalingmix only in combination with the other two signals.

The trade-offs facing the type M in choosing theoptimal combination of signals are similar to the trade-offs discussed above in the context of the type H firm.However, while the type H firm is concerned about thetype L mimicking it in terms of both product quality andproject value, the type M is concerned about the type Lmimicking it only in terms of project value. Finally, thetype L’s problem is straightforward, because giventhe equilibrium choices made by the type H and M firms,the type L is always worse off mimicking the other firmtypes compared with its payoff if it separates by followingits full information equilibrium strategy.

ARTICLE IN PRESS

12 While the substitution effect predicted by our model is an

equilibrium phenomenon, the prediction is that for any given firm type,

advertising and underfinancing are substitutes. However, if we compare

across firm types, a high type firm could have higher levels of advertising

as well as underfinancing compared with a medium type firm. In other

words, for us to empirically detect the above substitution effect

predicted by our model, either one of two conditions has to be satisfied.

We should be able to control for firm type adequately in our empirical

T. Chemmanur, A. Yan / Journal of Financial Economics 92 (2009) 40–65 47

3.2. Equilibrium choice of signals

Given the relative cost of the three possible signals, theequilibrium (least-cost) combination of these signals forthe type H firm is determined as follows. The equilibriumchoice of signaling mix by the type H firm depends on theextent of asymmetric information facing the firm and theinternal capital available to it. Consider first the case inwhich the extent of asymmetric information facing thetype H firm is relatively small. In this case, the type H firmuses underfinancing alone as a signal. The extent ofunderfinancing required to deter mimicking by the type Mand the type L firms results in the type H firm cutting backon investment in its new project only in the lowerproductivity ranges of its investment opportunity set.The type M firm underfinances as well in this case, thoughto a lesser extent compared with the level of under-financing incurred by the type H. This is becauseinformation asymmetry exists only about the type M’sproject value while it exists about both the type H’sproduct quality and its project value. In contrast, the type Lfirm does not underfinance. It raises the full investmentamount i1 from its IPO.

Consider now the case in which the extent ofasymmetric information facing the type H firm is moresevere. Signaling by underfinancing alone would be costly,because the extent of underfinancing required to determimicking by lower firm types is so large as to require thefirm to cut back on investment in the higher productivity(as well as the lower productivity) ranges of investment inits new project. To minimize its aggregate cost ofsignaling, the type H firm can add either advertising aloneor both advertising and underpricing to the equilibriumsignaling mix. The choice between advertising and under-pricing as an additional signal depends on the amount ofinternal capital available to the firm. When the internalcapital available is still sufficient for the type H firm tofund the required amount of advertising by cutting backonly on the lower productivity ranges of investment in itsexisting project, the opportunity cost of adding onlyadvertising as a signal in addition to underfinancing islower than that of adding both advertising and under-pricing to the signaling mix. In this case, the type H firmadds only advertising to the signaling mix, so that thetype H uses a combination of product market advertisingand underfinancing to signal its type. In this equilibrium,the type M firm could also advertise, as well as under-finance, to signal its firm type (depending on the extent ofasymmetric information). However, in the case thatthe type M advertises, it advertises less than the type H.The type L does not advertise at all. All three firm typesprice their equity in the IPO at their respective intrinsicvalues.

If, however, the internal capital available to the type Hfirm is smaller, the type H firm adds both underpricingand advertising, in addition to underfinancing, to thesignaling mix. In this case, if the type H firm chose to useunderfinancing alone to signal its firm type, it would haveto cut back on investment in the higher productivityranges of investment in its new project. Further, if it choseto fund advertising to the extent required to deter

mimicking by the lower firm types, it would have to cutback on the higher productivity ranges of investment in itsexisting project. As a result, the least-cost combination ofsignals in this equilibrium involves the type H firm usingIPO underpricing, as well as advertising and underfinan-cing, as signals. Similarly, the type M firm could under-price the equity in its IPO as well in this equilibrium,depending on the extent of asymmetric information itfaces and the amount of internal capital available to it.In the case that the type M underprices, the level of itsunderpricing is smaller than that of the type H firm, whilethe level of its advertising is the same as that of the type Hfirm. The type L firm does not underprice the equity in itsIPO. Neither does it advertise in equilibrium.

3.3. Empirical implications and testable hypotheses

Our theory has two important testable implications.The first implication is on product market advertisingaround new equity issues. The above theory predicts that,on average, a firm advertises more extensively in yearswhen it plans to make a public equity issue (IPO or SEO),relative to years when it does not plan to make such a newequity issue. When a higher intrinsic-value firm does notplan to issue new equity, the firm advertises only to theextent required to distinguish itself from lower intrinsic-value firms in the product market alone. When it plans toissue new equity, the higher intrinsic-value firm adver-tises more, because it needs to distinguish itself not onlyin the product market, but also in the financial market.This implication is the first hypothesis (H1) we test below.

The second implication is on the relation betweenunderpricing and product market advertising. We predictthat, when a firm faces significant asymmetric informa-tion and plans to issue new equity (whether in an IPO or inan SEO), it uses product market advertising and equityunderpricing as substitutes to signal to the financialmarket. Consider the equilibrium when a higher intrin-sic-value firm uses both product market advertising andunderpricing (together with underfinancing) to signal itsfirm type. In this equilibrium, when a higher intrinsic firmbecomes more financially constrained, signaling by ad-vertising becomes more expensive relative to signaling byunderpricing, because the investment cut required by thefirm to finance the additional advertising is moreexpensive to the firm than any losses incurred due tounderpricing. In this case (when the firm is morefinancially constrained), the firm could use less advertis-ing and more underpricing to signal its firm type, toreduce its aggregate signaling cost. This substitution effectbetween advertising and underpricing is the secondhypothesis (H2) we test below.12 For each of these two

ARTICLE IN PRESS

T. Chemmanur, A. Yan / Journal of Financial Economics 92 (2009) 40–6548

hypotheses, we first test the hypothesis on a sample of IPOfirms and then on a sample of SEO firms.

4. Sample selection

In this section, we describe our sample selection andvariable construction for both the IPO sample and the SEOsample.

4.1. IPO sample and variables

We obtain our initial sample of IPOs from the ThomsonFinancial’s Securities Data Corporation (SDC) new issuesdatabase. It covers the period from 1990 to 2000. Weextract financial statement information for IPO firms inour sample from the Standard & Poor’s Compustat filesand stock prices from the University of Chicago’s Centerfor Research in Securities Prices (CRSP). We then excludefrom our initial IPO sample spin-offs, American Deposi-tory Receipts (ADRs), unit offerings, leveraged buyouts(LBOs), limited partnerships, financial firms (SIC codes6000–6999), and utilities (SIC codes 4900–4999). We alsoexclude from our sample those firms with missing data onadvertising expenditures in the IPO year, when advertisingexpenditures are the cost of advertising, media, andpromotional expenses from Compustat item ]45. Thus,our final sample of IPOs consists of 884 firms. Panel A ofTable 1 reports the annual breakdown of our IPO sample.Panel A also reports the annual mean and medianamounts of the equity raised from IPOs, the first dayreturn, and advertising expenditures in the IPO year. Thispanel indicates that the time series pattern of the averagefirst-day return in our sample is similar to that in Ritterand Welch (2002), although we have a smaller sample. Forexample, the average first-day return is 13.2% between1990 and 1994 in our sample, 19.0% between 1995 and1998, and 61.2% between 1999 and 2000. The correspond-ing first day returns for these three periods are 11.2%,18.1%, and 65.0%, respectively, in Ritter and Welch (2002).However, the average amount of equity raised from IPOsin our sample is larger than that in Ritter and Welch(2002). In the empirical tests in Section 5, we could beconstrained to use only part of the sample due toincomplete information on lagged values of advertisingexpenditures or due to incomplete information in eitherthe CRSP or the Compustat database regarding otherfinancial variables.13

(footnote continued)

tests. Alternatively, the across-type variation in our data should be

dominated by the variations within types (all types of firms should be

well represented in our sample). We thank an anonymous referee for

pointing out this requirement of our empirical tests.13 Our sample reduces to 719 firms when we exclude those firms

with missing values on advertising expenditures one year prior to the

IPO year, to 459 firms when we exclude firms with missing values one

year subsequent to the IPO year, to 376 firms when we exclude firms

with missing values two years subsequent to the IPO year, and to 189

firms when we exclude those firms with missing values two years prior

to the IPO year. In the last case, our sample is further reduced to 177

firms if we exclude those firms with missing values on other variables.

We use all of the four samples in our test of hypothesis H1 and use the

In our study on the IPO sample, we define year t as IPOyear, year t�1 as one year prior to the IPO year, etc. Wefollow the industrial organization (IO) literature andconstruct advertising intensity in the IPO year t as ADVt/SLSt, the advertising expenditures in year t (ADVt) scaledby the sales revenue in the same year (SLSt). We alsoconstruct another advertising variable ADVt/SLSt�1, theadvertising expenditures in year t scaled by the salesrevenue in year t�1. The difference between ADVt/SLSt andADVt/SLSt�1 is that ADVt/SLSt (partially) excludes theadvertising in the IPO year t related to the product marketpurposes (e.g., to maintain or increase sales revenue SLSt)while ADVt/SLSt�1 captures the level of advertising in theIPO year t related to both the product market and thefinancial market purposes. To study the relation betweenadvertising and IPO underpricing, we also calculate thefirst-day return on the firm’s equity (IPO underpricing),RET, as the percentage change from the IPO offer price tothe first day closing price. RET, ADVt/SLSt, and ADVt/SLSt�1

are further winsorized at the 1% level in both tails of thedistribution to avoid the outlier effect.

The other variables used in our study of IPO firms areas follows. First, we follow the IPO literature and constructthe following financial market variables that could affectIPO first-day return RET. We measure underwriter rank,RANK, following Loughran and Ritter (2004) (see alsoCarter, Dark, and Singh, 1998), with higher ranks repre-senting higher quality. Chemmanur and Fulghieri (1994)demonstrate that a higher-ranked underwriter has anincentive to more accurately certify the quality of theissuing firm so as to protect its reputation, leadinginvestors to demand a smaller IPO discount. Similarly, afirm in an IPO underwritten by a syndicate or a firm listedin NYSE or a firm backed by venture capitalists could havea smaller IPO discount. We construct syndication dummy,SYND, equal to one if an IPO is underwritten by a syndicateand zero otherwise; exchange dummy, EXCHANGE, equalto one if a firm is listed in the NYSE or AMEX and zerootherwise; and VENTURE; equal to one if the firm makingthe IPO is backed by venture capitalists and zerootherwise. Further, Michaely and Shaw (1994) argue thatlarger issues can be harder to sell, thereby inducingunderwriters to underprice them by a larger amount. Wecalculate AMOUNT as the amount of money raised fromthe IPO scaled by the book value of assets in year t�1.The IPO literature also suggests that underpricing could berelated to the change of offer price from the filing price.

(footnote continued)

sample with 719 firms and the sample with 189 firms in our test of

hypothesis H2. The distributions of the various variables in the first three

samples are similar to those in the whole sample with 884 firms (e.g., in

terms of the average advertising expenditures and the average first day

return). However, for the sample with 189 firms, both advertising

expenditures and the first day return, either on average or across time,

are higher than those for the whole sample. The greater advertising

expenditures and greater first day return in the latter sample potentially

could affect our tests of H2 by rejecting the substitution effect between

advertising and underpricing. In other words, the latter sample is biased

against finding evidence in support of hypothesis H2. Thus, the fact that

we find evidence consistent with the substitution hypothesis even using

the sample with 189 firms reinforces our confidence in the substitution

effect between advertising and underpricing.

ARTICLE IN PRESS

Table 1Distribution of initial public offerings (IPOs) and seasoned equity offerings (SEOs) across years. Panels A and B present the number of firms, the mean and

median amounts of IPO and SEO proceeds each year, first day return, and advertising expenditures for a sample of firms going public and a sample of firms

issuing seasoned equity, respectively, between 1990 and 2000.

Panel A. A sample of firms going public

IPO proceeds (millions of dollars) First day return Advertising expenditures (millions of dollars )

Year Number of firm Mean Median Mean Median Mean Median

1990 41 19.83 18.10 0.09 0.04 1.88 0.16

1991 99 28.47 25.00 0.15 0.11 1.82 0.32

1992 115 38.89 24.80 0.11 0.04 1.98 0.22

1993 154 35.65 26.50 0.15 0.09 1.57 0.42

1994 47 30.05 22.00 0.12 0.09 1.74 0.65

1995 46 33.23 22.65 0.17 0.10 6.20 0.98

1996 85 34.01 27.00 0.20 0.13 2.34 0.60

1997 74 55.43 31.95 0.17 0.11 9.29 1.15

1998 51 45.30 32.40 0.22 0.08 5.51 1.00

1999 115 84.83 51.20 0.66 0.35 6.86 1.41

2000 57 109.39 60.00 0.51 0.29 4.78 1.13

Total 884 47.32 29.90 0.24 0.10 3.76 0.63

Panel B. A sample of firms issuing seasoned equity

SEO proceeds (millions of dollars) First day return Advertising expenditures (millions of dollars)

Year Number of firms Mean Median Mean Median Mean Median

1990 31 53.55 29.70 0.014 0.005 6.97 0.00

1991 93 105.52 34.10 0.023 0.010 44.07 0.43

1992 98 87.15 37.95 0.022 0.010 30.13 0.10

1993 121 87.21 44.80 0.025 0.016 17.50 0.15

1994 29 69.30 42.60 0.024 0.009 19.90 4.52

1995 38 115.54 52.95 0.015 0.000 112.56 3.19

1996 40 82.65 49.20 0.030 0.021 11.83 4.05

1997 36 99.42 65.35 0.027 0.004 73.65 4.03

1998 41 132.98 84.50 0.020 0.010 69.28 10.50

1999 74 214.41 123.65 0.020 0.004 37.59 9.56

2000 62 308.95 169.65 0.033 0.027 34.89 1.75

Total 663 127.19 56.00 0.024 0.010 37.93 2.11

T. Chemmanur, A. Yan / Journal of Financial Economics 92 (2009) 40–65 49

We construct REVISION as IPO offer price minus the mid-point of the filing range scaled by IPO offer price. Recently,Loughran and Ritter (2004) find that conditions in thestock market affect initial IPO returns, though Lowry andSchwert (2002) find this effect to be insignificant.We measure stock market conditions by MRUN, calculatedas the compounded return of an equally weighted marketindex 60 days prior to the IPO, NIPO, the number of all IPOsin the firm’s IPO year, and BUBBLE, a dummy variable equalto one if the IPO year is between 1999 and 2000 and zero ifit is between 1990 and 1998. We have also tried othermarket variables such as the average first day return of allIPOs in the firm’s IPO year, the number of seasoned equityissues in the IPO year, and contemporary price-to-earningsratio of S&P 500 in the IPO year. Our results would remainunchanged with the inclusion of these variables.

Second, we follow the industrial organization literatureand construct the following product market variables tocapture the factors that could affect a firm’s advertising

decision. We calculate LOGSLS as the log of sales revenuein year t; DLOGSLSt as the change in the log of salesrevenue from year t�1 to year t; 1/SLSt as the inverse ofsales revenue in year t; FSALE as the fraction of a firm’ssales revenue to the total sales revenue of its industry,where industries are grouped by two-digit SIC codes; andadvertising intensity in pre-IPO years. Advertising inten-sity in pre-IPO years measures the established advertisingintensity of a firm in the product market (given no equityissue in these years). FSALE measures the extent of productmarket competition in the industry.

Finally, we also construct the following variables tocontrol for various firm characteristics. We construct atech dummy TECH equal to one if a firm is classified as atech firm and zero otherwise. The definition of a tech firmis based on the firm’s SIC code and follows that inLoughran and Ritter (2004). Tech firms are defined asthose in SIC codes 3571, 3572, 3575, 3577, 3578, 3661,3663, 3669, 3674, 3812, 3823, 3825–3827, 3829, 3841,

ARTICLE IN PRESS

T. Chemmanur, A. Yan / Journal of Financial Economics 92 (2009) 40–6550

3845, 4812, 4813, 4899, 7370–7375, 7378, and 7379. Firmsize, SIZEt�1, is calculated as the log of the book value ofassets in year t�1. Firm age, AGE, is calculated as thedifference between the IPO year and the founding year.EBITt�1 is calculated as the earnings before interest, taxes,depreciation and amortization in year t�1 scaled by thebook value of assets in the same year. LDRt�1 is calculatedas the ratio of long-term debt to the book value of assets.Column 1 of Table 2 reports the sample statistics of allvariables used in our study on the IPO sample.

4.2. SEO sample and variables

We obtain our initial sample of SEOs from the SDCdatabase as well. It covers the period from 1990 to 2000.We exclude from our SEO sample IPOs, financial firms(SIC codes 6000–6999), utilities (SIC codes 4900–4999),closed-end funds, Real Estate Investment Trusts (REITs),

Table 2Summary statistics of the IPO sample and the SEO sample. This table presents

change from the offer price to the first day closing price. ADVt/SLSt and ADVt/SLSt

year t or year t�1, respectively. LOGSLSt is the log of sales revenue. DLOGSLS is

money raised from IPO or SEO scaled by the book value of assets in year t�1. SY

the book value of assets. EBITt�1 is earnings before interest, taxes, depreciation a

high technology firms. RANK is the proxy for underwriter ranks. EXCHANGE is a

dummy equal to one if the issuer is backed by venture capitalists. AGE is the diff

long-term debt to the book value of assets. MRUN is the compounded return of

offer date. BUBBLE is a dummy for years 1999 and 2000. NIPO is the log of the

industry sales revenue in the past year. MBt�1 is the ratio of the market value t

price minus the mid-point of filing range scaled by the offer price.

(1) IPO sample

Number of observations Mean Med

RET 884 0.244 0.1

ADVt/SLSt 884 0.059 0.0

ADVt/SLSt�1 730 0.379 0.0

Variables related to IPO or SEO in financial market

VENTURE 869 0.459 0

SYND 884 0.868 1

RANK 866 7.033 8

EXCHANGE 884 0.092 0

AMOUNT 730 3.720 1.5

BUBBLE 884 0.195 0

NIPO 884 6.216 6.2

MRUN 884 0.101 0.0

REVISION 873 0.370 0.0

Variables related to advertising in product market

ADVt�1/SLSt�1 719 0.049 0.0

DLOGSLS 730 0.521 0.3

LOGSLSt 884 3.454 3.5

FSALE 730 0.011 0.0

Other variables

TECH 884 0.381 0

AGE 861 13.235 7

SIZEt�1 730 2.875 2.8

MBt�1

LDRt�1 730 0.262 0.0

EBITt�1 728 �0.056 0.1

units, and non-US shares. Closed-end funds, REITs, units,and non-US shares are identified using SDC classificationsand by CRSP share codes. We also check each firm’s salesfrom common and preferred stock from Compustat (item]108) and keep only those firms with sales of more than$2 million in the SEO year. We further exclude from oursample those firms involving in stock dividends, stocksplits, and acquisitions and reorganizations within tentrading days around the SEO offer date, which areidentified by CRSP distribution codes. Finally, we requirethe firms in our final sample to have data available onadvertising expenditures in the SEO year. Thus, our finalsample consists of 663 firms. Panel B of Table 1 reports theannual breakdown of the SEO sample. Panel B also reportsthe annual mean and median amounts of the equity raisedfrom SEOs and advertising expenditures in the SEO year.

Lease, Masulis, and Page (1991) note that SDC-statedoffer dates for SEOs are often inappropriate for analyzing

sample statistics of all variables used in the paper. RET is the percentage

�1 are advertising expenditures in the IPO or SEO year t scaled by sales in

the change in LOGSLS from year t�1 to year t. AMOUNT is the amount of

ND is a dummy variable for syndicated underwriting. SIZEt�1 is the log of

nd amortization scaled by the book value of assets. TECH is a dummy for

dummy equal to one if a firm is listed in NYSE or AMEX. VENTURE is a

erence between the IPO year and the founding year. LDRt�1 is the ratio of

an equally weighted market index during 60 days prior to the IPO or SEO

numbers of IPO deals. FSALE is the fraction of a firm’s sales revenue to

o the book value of assets. REVISION is calculated as the IPO or SEO offer

(2) SEO sample

ian Number of observations Mean Median

00 663 0.024 0.010

19 663 0.031 0.015

29 602 0.053 0.022

663 0.053 0

663 0.848 1

649 8.022 8.425

663 0.474 0

49 602 0.718 0.364

663 0.205 0

32 663 6.176 6.232

98 651 0.116 0.105

00 663 �0.074 �0.044

19 602 0.030 0.015

55 602 0.418 0.287

89 663 5.195 5.204

02 602 0.050 0.012

663 0.253 0

37 602 4.965 4.804

542 3.151 1.966

97 601 0.234 0.177

24 602 0.075 0.123

ARTICLE IN PRESS

T. Chemmanur, A. Yan / Journal of Financial Economics 92 (2009) 40–65 51

price effects because some SEOs take place after the closeof trading. We follow Safieddine and Wilhelm (1996) andapply the volume-based algorithm to correct the offerdate provided by SDC. In particular, if trading volume onthe day following the SDC-stated offer date is more thantwice the trading volume on the SDC-stated offer date, wedesignate the day following the SDC-stated offer date asthe offer date. This correction results in an offer-datechange for 52% of our sample.

In our study on the SEO sample, we define year t as SEOyear, year t�1 as one year prior to the SEO year, etc. Theadvertising variables ADVt/SLSt�1 and ADVt/SLSt are calcu-lated in the same way as in Section 4.1. We define SEOunderpricing RET as the percentage change from the SEOoffer price to the first-day closing price (see, e.g., Corwin,2003). Similar to the study on IPO firms, we control in thestudy on SEO firms for the financial market variablesrelated to a firm’s decision to issue new seasoned equity,consisting of VENTURE, SYND, RANK, EXCHANGE, BUBBLE,NIPO, MRUN, REVISION and AMOUNT, where AMOUNT isthe amount of equity raised from seasoned offeringsscaled by the book value of assets in year t�1. We havealso tried to control for other financial market variablessuch as the average first day return of all IPOs in the SEOyear, the number of seasoned equity issues in the SEOyear, contemporary price-to-earnings ratio of S&P 500, thecompounded stock return during 60 days prior to the SEOoffer date, the coefficient of variation of the daily stockreturns during 60 days prior to the SEO offer date, and thelog of the market value one day before the SEO offer day.Our results would remain unchanged by including theseadditional variables.

We further control for the product market variablesrelated to a firm’s advertising policy, consisting ofDLOGSLSt, LOGSLSt, 1/SLSt, FSALE, and lagged advertisingintensity. Finally, we control for the other firm character-istics consisting of TECH, LDRt�1, EBITt�1, SIZEt�1, andMBt�1, where MBt�1 is calculated as the ratio of the marketvalue of assets in year t�1 to the book value of assets inthe same year.14 Column 2 of Table 2 reports the samplestatistics of the variables for our SEO sample.

4.3. Matching firms for IPO and SEO samples

We also construct matching firms that have no plan toissue new equity. We match each IPO firm and each SEOfirm with a Compustat-listed non-IPO and non-SEO firmon the basis of industry, asset size, and sales revenue,following the matching algorithm in Loughran and Ritter(1997). The matching procedure consists of several steps.First, we construct a preliminary matching sample, usingnon-IPO and non-SEO Compustat firms with sales fromcommon and preferred stock totaling less than $2 million.

14 If we compare the variables that we use for the IPO and SEO

analyses, variable AGE is used for IPO firms but not for SEO firms due to

the unavailability of data. We use variable MBt-1 for SEO firms but not for

IPO firms because pre-IPO market data for IPO firms are not available.

Our results do not change qualitatively if we drop the additional control

variables and keep the same set of control variables for both our IPO and

SEO analyses.

Thus, the advertising expenditures of our matching firmsare unlikely to be affected by new equity issues. Second,we use this preliminary matching sample to match eachIPO firm and each SEO firm to those non-issuing firmswith the same two-digit SIC code and whose asset size liesbetween 25% and 250% of the size of the correspondingIPO or SEO firm in year t. If the matching yields zero non-issuing firms, we relax the industry matching (based onSIC code) and the asset matching requirements. Finally,the selected non-issuing firms are ranked by their salesrevenue in year t. The firm with the closest sales revenueto the IPO firm’s sales revenue in year t is selected as thematching firm for the IPO firm. Similarly, the firm with theclosest sales revenue to the SEO firm’s sales revenue inyear t is selected as the matching firm for the SEO firm.

We present in Panel A, Table 3, the sample statistics ofselected variables for IPO firms, their matching firms, andthe difference between IPO firms and matching firms. Wepresent in Panel B, Table 3, similar statistics for SEO firmsand their matching firms. Table 3 shows that, comparedwith matching firms, both IPO firms and SEO firms havehigher sales growth (larger DLOGSLS), higher salesrevenue (larger LOGSLS), lower profit (smaller EBIT), andless market share (smaller FSALE). SEO firms also havemore firm growth options (larger MB) than their matchingfirms.

5. Empirical results for IPO firms

In this section, we test our hypotheses in a sample ofIPO firms.

5.1. Change in advertising expenditures around IPOs:

univariate tests

Hypothesis H1 predicts that a firm advertises morewhen it plans to go public. However, a firm’s advertisingexpenditures in its IPO year could consist of both theadvertising expenditures incurred due to the firm’santicipation of going public and that incurred due toproduct market considerations. Thus, to study howadvertising expenditures are related only to the firm’sgoing public decision, we need to exclude the effect ofproduct market considerations on the firm’s advertisingexpenditures. We do so by studying a firm’s advertisingintensity. Compared with advertising expenditures, ad-vertising intensity is scaled by a firm’s sales revenue andis thereby less affected by a firm’s product marketconsiderations (for example, considerations of boostingsales revenue).

In the following, we test hypothesis H1 in two ways.First, we study a firm’s advertising intensity around its IPOyear. We expect that an IPO firm’s advertising intensitypeaks in its IPO year. Second, we compare the change inadvertising intensity around the IPO year between an IPOfirm and its matching firm that does not make a newequity issue. We expect that the peak in advertisingexpenditures in the IPO year occurs only to the IPO firmbut not to its matching firm.

ARTICLE IN PRESS

Table 3Comparison of firms issuing new equity and matching firms. This table presents sample statistics of selected variables for firms issuing new equity, their

matching firms, and the pair-wise difference between issuing firms and their matching firms. Panel A is for the IPO sample and panel B is for the SEO

sample. A matching firm is constructed for each IPO firm and each SEO firm following the method in Loughran and Ritter (1997), by industry, size, and

sales revenue at the new equity issue year t. EBITt�1 is earnings before interest, taxes, depreciations and amortization scaled by the book value of assets.

MBt�1 is the ratio of the market value to the book value of assets. LOGSLSt is the log of sales revenue in year t. DLOGSLS is the change in LOGSLS from year

t�1 to year t. FSALE is the fraction of a firm’s sales revenue to industry sales revenue in the past year. Industries are identified using two-digit SIC codes.

Tests on means and medians are based on t-tests and Wilcoxon tests, respectively. *, **, and *** indicate significant difference from zero at the 10%, 5%, and

1% levels, respectively, using two-tailed tests.

Panel A. Selected firm characteristics of IPO firms and their matching firms

(1) IPO sample (2) Matching firms for IPO sample (3) IPO firms–matching firms

Number of observations Mean Median Number of observations Mean Median Number of observations Mean Median

MBt�1 818 1.715 1.272

EBITt�1 728 �0.056 0.124 834 0.069 0.106 720 �0.120*** �0.007***

DLOGSLS 730 0.521 0.355 837 0.052 0.064 724 0.470*** 0.313***

LOGSLSt 884 3.454 3.589 884 3.404 3.569 884 0.0492*** 0.002***

FSALEt�1 837 0.014 0.002 730 0.011 0.002 724 �0.002*** �0.000***

Panel B. Selected firm characteristics of SEO firms and their matching firms

(1) SEO sample (2) Matching firms for SEO sample (3) SEO firms–matching firms

Number of observations Mean Median Number of observations Mean Median Number of observations Mean Median

MBt�1 542 3.151 1.966 663 1.532 1.248 663 1.208*** 0.717***

EBITt�1 602 0.075 0.123 624 0.113 0.120 579 �0.033*** �0.001

DLOGSLS 602 0.418 0.287 624 0.053 0.055 579 0.353*** 0.248***

LOGSLSt 663 5.195 5.204 663 4.907 4.947 663 0.288*** 0.100***

FSALEt�1 602 0.050 0.012 624 0.044 0.011 579 0.005*** 0.000

T. Chemmanur, A. Yan / Journal of Financial Economics 92 (2009) 40–6552

5.1.1. IPO years versus non-IPO years

We first test hypothesis H1 by comparing a firm’sadvertising intensity in its IPO year with that in non-IPOyears. We study four non-IPO years: the year two yearsprior to IPO (year t�2), the year one year prior to IPO (yeart�1), the year one year subsequent to IPO (year t+1), andthe year two years subsequent to IPO (year t+2).

To relate the difference in advertising between the IPOyear and non-IPO years to the decision of going public, werequire non-IPO years to be unaffected by the decision toissue new equity. Clearly, year t�2 satisfies this require-ment because the advertising two years prior to the IPOyear is unlikely to reach potential investors in the IPOmarket. However, the size of the sample in year t�2 issmall due to the fact that many IPO firms in our sample donot have financial information (including advertisingexpenses) in year t�2. Unlike year t�2, there could existsome lingering effects of advertising in year t�1 onpotential investors in the IPO year t. However, as long asthe advertising in year t�1 and year t affects potentialinvestors in the IPO year t differently, our hypothesis H1still predicts an increase in advertising expenditures fromyear t�1 to year t. Compared with the years prior to IPO,the problem of data availability is considerably mitigatedfor the years subsequent to IPO, years t+1 and t+2.15

15 We thank an anonymous referee for pointing out the possibility of

using the years subsequent to IPO as non-IPO benchmark years.

However, a firm could choose to issue new seasonedequity in years t+1 and t+2. Thus, for t+1 and t+2 to beselected into our study, we further require that the firmdoes not have any seasoned equity issues in these years,i.e., the firm-year is not covered in our SEO sample and thefirm’s sales of common and preferred stock in these yearsare less than $2 million according to Compustat. By doingso, we ensure a firm’s advertising policy in years t+1 andt+2 to be unaffected by its decision to issue new equity.

We report in Column 1 of Panel A, Table 4, the samplestatistics of the level of advertising intensity in the yearsaround the IPO year and the pair-wise differences inadvertising intensity between the IPO year and each non-IPO year. Our results show that a firm’s advertisingintensity is greater in its IPO year than that in the non-IPO years. The difference in advertising intensity issignificant for all the non-IPO years in the t-test and it issignificant for most non-IPO years in the Wilcoxon non-parametric test. For example, the mean change inadvertising intensity from year t�2 to year t, (ADVt/SLSt)�(ADVt�2/SLSt�2), is 0.012 and the mean of (ADVt/SLSt)�(ADVt�1/SLSt�1) is 0.008. These statistics suggest that afirm on average increases its advertising expenditures by1.2% of its sales revenue from year t�2 to the IPO year t

and by 0.8% of its sales revenue from year t�1 to the IPOyear t. Further, the mean of (ADVt/SLSt)�(ADVt+1/SLSt+1) is0.007 and the mean of (ADVt/SLSt)�(ADVt+2/SLSt+2) is0.022, which suggest that a firm on average decreases itsadvertising expenditures by 0.7% of its sales revenue from

ARTICLE IN PRESS

Table 4Summary statistics of the change in advertising expenditures between the initial public offering (IPO) year and non-IPO years. Panel A reports the levels of

advertising intensity (ADV/SLS) in the five years around the IPO year and the difference in ADV/SLS between non-IPO years and the IPO year. ADV/SLS is

advertising expenditures scaled by sales revenue in the same year. A matching firm is constructed for each IPO firm following the method in Loughran and

Ritter (1997), by industry, size, and sales revenue in the IPO year t. Panel B reports the change in ADV/SLS for IPO firms surviving both years t�1 and t�2

and their matching firms. Panel C reports the change in ADV/SLS for IPO firms surviving both years t+1 and t+2 and their matching firms. In Panel D,

surviving and entering IPO firms are those reporting and not reporting ADV/SLS in the pre-change year, respectively. In Panel E, surviving and exiting firms

are those IPO firms reporting and not reporting ADV/SLS in the subsequent year, respectively. In Panel F, we create subsamples of matching firms based on

whether a matching firm’s EBITt�1, calculated as earnings before interest, taxes, depreciation and amortization scaled by the book value of assets, is higher

or lower than the sample median (high and low EBITt�1) and on whether a matching firm’s MBt�1, the ratio of the market value to the book value of assets,

is higher or lower than the sample median (high and low MBt�1). Tests on means and medians are based on t-tests and Wilcoxon tests, respectively. *, **,

and *** indicate significant difference from zero at the 10%, 5%, and 1% levels, respectively.

(1) IPO firms (2) Matching firms (3) IPO firms–matching firms

Number of

observations

Mean Median Number of observations Mean Median Number of

observations

Mean Median

Panel A. Pair-wise difference in advertising intensity: IPO firms versus matching firms

ADVt�2/SLSt�2 189 0.055 0.021 693 0.043 0.023 189 0.014** �0.002

ADVt�1/SLSt�1 719 0.049 0.019 837 0.040 0.021 719 0.009** �0.003

ADVt/SLSt 884 0.059 0.019 884 0.039 0.021 884 0.020*** �0.003

ADVt+1/SLSt+1 459 0.049 0.019 749 0.039 0.023 459 0.011*** �0.003

ADVt+2/SLSt+2 376 0.036 0.018 636 0.038 0.023 376 �0.001 �0.003

ADVt/SLSt–ADVt�2/SLSt�2 189 0.012** 0.000 693 �0.004*** �0.001*** 189 0.012** 0.001

ADVt/SLSt–ADVt�1/SLSt�1 719 0.008*** 0.000** 837 �0.001 0.000 719 0.009*** 0.001***

ADVt/SLSt–ADVt+1/SLSt+1 459 0.007*** 0.000** 749 0.001 0.000 459 0.005* 0.000

ADVt/SLSt–ADVt+2/SLSt+2 376 0.022*** 0.001*** 636 0.002* 0.001*** 376 0.020*** �0.001

Panel B. Change in advertising intensity, IPO firms and matching firms surviving two years prior to the IPO year

ADVt/SLSt–ADVt�2/SLSt�2 189 0.012** 0.000 189 �0.001 �0.001 189 0.012** 0.001

ADVt/SLSt–ADVt�1/SLSt�1 189 0.011*** 0.000* 189 0.000 0.000 189 0.011*** 0.001

Panel C: Change in advertising intensity: IPO firms and matching firms surviving two years subsequent to the IPO year

ADVt/SLSt–ADVt+1/SLSt+1 275 0.007** 0.000* 275 0.000 0.000*** 275 0.007*** 0.000

ADVt/SLSt–ADVt+2/SLSt+2 275 0.021*** 0.001*** 275 0.001 0.001 275 0.019** 0.000

Panel D: Change in advertising intensity: IPO firms entering the sample versus IPO firms surviving the past year

(1) Entering IPO firms (2) surviving IPO firms (3) Entering-surviving firms

Number Mean Median Number Mean Median Number Mean Median

ADVt/SLSt–ADVt�1/SLSt�1 530 0.006** 0.000 189 0.011*** 0.000* 719 �0.004 0.000

ADVt/SLSt–ADVt+1/SLSt+1 78 0.004 0.000 381 0.007*** 0.000*** 459 �0.003 0.000

ADVt/SLSt–ADVt+2/SLSt+2 64 0.019* 0.000 312 0.023*** 0.001*** 376 -0.003 0.000

Panel E: Change in advertising intensity: IPO firms exiting the sample versus IPO firms surviving the subsequent year

(1) Exiting IPO firms (2) surviving IPO firms (3) Exiting-surviving firms

Number Mean Median Number Mean Median Number Mean Median

ADVt/SLSt–ADVt�2/SLSt�2 88 0.006 0.000 101 0.017** 0.000 189 �0.011 0.000

ADVt/SLSt–ADVt�1/SLSt�1 338 0.009*** 0.000 381 0.006** 0.000** 719 0.003 0.000

ADVt/SLSt–ADVt+1/SLSt+1 184 0.006 0.000 275 0.007** 0.000* 459 �0.001 0.000

Panel F: Change in advertising intensity of matching firms: ADVt/SLSt-ADVt�1/SLSt�1

Number Mean Median Number Mean Median

Firms with high MBt�1 409 �0.001 0.000 Firms with high EBITt�1 417 0.002** 0.001***

Firms with low MBt�1 409 �0.001 �0.000 Firms with low EBITt�1 417 �0.004*** �0.001***

Difference (high

MBt�1�low MBt�1)

818 0.000 0.001 Difference (high

EBITt�1�low EBITt�1)

834 0.006*** 0.002***

T. Chemmanur, A. Yan / Journal of Financial Economics 92 (2009) 40–65 53

ARTICLE IN PRESS

T. Chemmanur, A. Yan / Journal of Financial Economics 92 (2009) 40–6554

the IPO year t to year t+1 and by 2.2% of sales revenuefrom the IPO year t to year t+2. These results suggest that afirm’s advertising expenditures peak in the IPO yearduring a five-year window surrounding the firm’s IPO.Remember that advertising intensity is scaled by salesrevenue and the non-IPO years are not affectedby the decision to issue new equity. Thus, our results areconsistent with hypothesis H1. They suggest that thepeaking of advertising in the IPO year is most likely due tothe firm’s going public decision, not any product marketconsiderations.

5.1.2. IPO firms versus matching firms

Second, we test hypothesis H1 by comparing theadvertising intensity around the IPO year between IPOfirms and their matching firms. We provide the level andthe change in advertising intensity for the matching firmsaround the IPO year in Column 2 of Panel A, Table 4. Wefind that the advertising intensity of matching firmsdecreases over the years from year t�2 to year t+2.For example, the mean of (ADVt/SLSt)�(ADVt�2/SLSt�2) is�0.004 and the mean of (ADVt/SLSt)�(ADVt+2/SLSt+2) is0.002. Both differences are significant in the t-test andWilcoxon test. Such a pattern of declining advertisingintensity over the years could be due to the decreasingreturns to scale of advertising in the product market.When a firm’s sales revenue grows above a certain level,the marginal advertising expenditures required to supportan additional increase in sales becomes smaller. We alsoprovide in Column 3 of Panel A, Table 4, the difference inthe change in advertising intensity between IPO firms andtheir matching firms. We find that from year t�2 to theIPO year t, IPO firms experience a larger increase in theiradvertising expenditures than their matching firms. Fromyear t to year t+2, IPO firms experience a larger decrease intheir advertising expenditures than their matching firms.These results suggest that the peak in advertising in theIPO year t occurs only to IPO firms and it does not occur tothe matching firms. Remember that our matching firmsare matched by industry, size, and sales revenue and thatthey are constructed in such a way that they have no planto issue new equity in the sample years (see Section 4.3).Thus, our results here support hypothesis H1 as well. Theysuggest that an IPO firm’s advertising growth in its IPOyear is most likely to be driven by the firm’s going publicdecision instead of by any industry-wide or any productmarket considerations.

5.1.3. Robustness checks

In the above test, our sample sizes in the IPO year andthe non-IPO years are different. Thus, our results couldbe driven by the changes in the composition of firms inthe different samples. In the first robustness check, weaddress this concern by keeping the same set of firmsexisting in the IPO year and the pre-IPO years (i.e., yearst�2, t�1, and t). Using these firms, we study the change inadvertising intensity from the pre-IPO years to the IPOyear. The results are presented in Panel B, Table 4. We alsostudy the change in advertising intensity from the IPOyear to the post-IPO years based on the same set of firmsexisting in these years (i.e., years t+2, t+1, and t). We

present the results from this study in Panel C, Table 4. Theresults from both Panels B and C are similar to thosepresented in Panel A: advertising expenditures peak in theIPO year for IPO firms but not for matching firms. Thus,our results discussed in Section 5.1.2 are unlikely to bedriven by the composition of firms.

Our second robustness check is on the possibility ofsurvivorship bias. One concern for our results is that theycould be driven by the new firms entering the sample orthe firms exiting the sample. We address this concern intwo steps. We first study whether or not firms enteringthe sample are more likely to increase their advertisingexpenditures. We disaggregate the IPO sample intosubsamples of entering firms and surviving firms, whereentering firms are those firms with no data available onadvertising intensity in the previous year and survivingfirms are those with data available in the previous year.For example, in the test on (ADVt/SLSt)�(ADVt�1/SLSt�1),subsamples of entering firms and surviving firms areconstructed based on data availability in year t�2. In boththe test on (ADVt/SLSt)�(ADVt+1/SLSt+1) and the test on(ADVt/SLSt)�(ADVt+2/SLSt+2), subsamples are constructedbased on data availability in year t�1. We then comparethe change of advertising intensity between enteringfirms and surviving firms. The results from this studyare presented in Panel D of Table 4. We find that theadvertising growth in the IPO year is not differentbetween entering firms and surviving firms.

We also address the concern of survivorship bias bystudying whether or not firms experiencing an increase inadvertising are more likely to disappear from the sample.In this study, we disaggregate the sample into exiting IPOfirms and surviving IPO firms based on whether firms inour IPO sample have data available on advertisingintensity in the subsequent year. For example, in boththe test on (ADVt/SLSt)�(ADVt�1/SLSt�1), and the test on(ADVt/SLSt)�(ADVt�2/SLSt�2), we select exiting and surviv-ing firms based on data availability in year t+1. In the teston (ADVt/SLSt)�(ADVt+1/SLSt+1), our selection is based ondata availability in year t+2. The results from this studyare presented in Panel E of Table 4. They show that exitingfirms demonstrate a similar pattern on the change inadvertising intensity compared with surviving firms. Thus,our evidence on both entering firms and exiting firmssuggest that our study on advertising intensity around IPOyears is unlikely to be affected by survivorship bias.

Finally, we check whether our results on the differencein advertising intensity between IPO firms and matchingfirms are driven by product market considerations such asfirm growth and firm profitability. Our sample statistics inTable 3 show that IPO firms have lower profitability (EBIT)and higher sales growth (DLOGSLS) than matching firms.IPO firms also could have higher firm growth (MTOB).In our previous studies, we control for the sales con-sideration by constructing advertising intensity. However,the difference in advertising intensity between IPO firmsand matching firms could still be driven by theirdifference in firm growth or profitability. We address thisconcern by studying how firm growth and firm profit arerelated to advertising in matching firms. Such a relationderived from matching firms provides a benchmark for

ARTICLE IN PRESS

T. Chemmanur, A. Yan / Journal of Financial Economics 92 (2009) 40–65 55

IPO firms on how firm growth and firm profitability wouldaffect the advertising of IPO firms through product marketconsiderations. If IPO firms’ advertising is related to firmgrowth and profitability in a way that cannot be explainedby the relations predicted by their matching firms, thenwe can reject the possibility that the peak of advertising inIPO firms in their IPO years is driven by firm growth orfirm profitability.

We disaggregate the matching sample into subsampleswith high firm growth and low firm growth, where highand low growth firms are defined based on whether afirm’s MTOBt�1 is above or below the median MTOBt�1 ofall matching firms. We also disaggregate the matchingsample into the high profitability and low profitabilitysubsamples based on whether a firm’s EBITt�1 is above orbelow the median EBITt�1 of all matching firms. We thenstudy the difference in the change in advertising intensity(ADVt/SLSt)�(ADVt�1/SLSt�1) between the high growthfirms and the low growth firms and the differencebetween the high profitability firms and the low profit-ability firms. The results from this study are provided inPanel F of Table 4. We find that matching firms with lowprofitability tend to cut their advertising expenditures andmatching firms with high profitability tend to increasetheir advertising expenditures. We also find that firmgrowth (MTOB) in general does not affect a matchingfirm’s advertising expenditures. Recall our findings thatIPO firms increase their advertising expenditures in theirIPO years even though they have lower profitability andhigher firm growth. Thus, the pattern of advertising, EBIT,and MTOB in IPO firms is not consistent with what ispredicted by their matching firms and therefore unlikelyto be explained by product market considerations alone.

5.2. Change in advertising expenditures around IPOS:

multivariate analysis

In Section 5.1, we use advertising intensity to controlfor the sales consideration. However, advertising intensitycould still be affected by other product market factors. Weaddress this issue in our multivariate analysis.

5.2.1. Advertising intensity in IPO firms

We first study the difference in IPO firms’ advertisingintensity between their IPO years and non-IPO years. Werun the following regression:

Advertising Intensity ¼ a0 þ a1ISSUEþ a2PRODþ �. (1)

The sample of regression Eq. (1) consists of both IPOfirms in the IPO year and IPO firms in non-IPO years. Weselect non-IPO years from years t�2, t�1, t+1, and t+2 inthe same way as in Section 5.1.1. Advertising intensity, thedependent variable, is measured either in the IPO year orin non-IPO years. ISSUE is a dummy variable for a newequity issue, which equals one if the correspondingdependent variable is measured in the IPO year and zeroif it is measured in non-IPO years. PROD is a vector ofproduct market variables that could affect a firm’s productmarket advertising and e is an error term.

We control for four product market variables, followingthe IO literature. We control for a firm’s sales growth from

the prior year to the current year, DLOGSLS; the inverse ofa firm’s sales revenue in the current year, 1/SLS; thecompetition that a firm faces in its industry, FSALE; and afirm’s lagged advertising intensity. All the control vari-ables are measured corresponding to the year when thedependent variable is measured. For example, in aregression on the difference in advertising intensitybetween the IPO year t and the non-IPO year t�1, thedependent variable in Eq. (1) is either ADVt/SLSt or ADVt�1/SLSt�1. For those observations in which the correspondingdependent variable is ADVt/SLSt (measured in the IPO year t),the control variable DLOGSLS is measured as the salesgrowth from year t�1 to year t; 1/SLS is measured in year t;and both FSALE and lagged advertising intensityare measured in year t�1. In the same regression, forthose observations where the corresponding dependentvariable is ADVt�1/SLSt�1 (measured in the non-IPO yeart�1), the control variable DLOGSLS is measured as thesales growth from year t�2 to year t�1; 1/SLS is measuredin year t�1; and both FSALE and lagged advertisingintensity are measured in year t�2. Finally, we alsocontrol for cross-sectional differences across firms byrunning a random effect model.

In regression Eq. (1), a1 measures the difference inadvertising intensity between the IPO year and non-IPOyears for an average IPO firm. Given our control forproduct market variables PROD, such a differencelikely arises from financial market considerations ratherthan product market considerations. In this sense,a1 can be viewed as measuring the average level ofadvertising related to the IPO firm’s going public decision.We expect a1 to be positive to be consistent withhypothesis H1.

We present the results from Eq. (1) in Table 5, withyear t�2, t�1, t+1, or t+2 as the non-IPO year. In the firstfour columns, we use the sample with unbalanced panels,i.e., we do not require IPO firms to have data available inboth the IPO year and the non-IPO years. In Column 1,where we use an unbalanced sample of firms in the IPOyear t and the non-IPO year t�2, we do not control forDLOGSLS, FSALE, and lagged advertising intensity due tothe lack of information in year t�3 for IPO firms. Ingeneral, our results are similar in the first four columns.The coefficient of ISSUE, a1, is positive and significant. Forexample, the results in Columns 2 and 3 suggest that,controlling for product market factors, a firm increases itsadvertising expenditures by 0.8% of its sales revenue fromyear t�1 to the IPO year t and then decreases advertisingexpenditures by 0.9% of its sales revenue from the IPOyear t to year t+1.

To ensure the robustness of our results, we also reportin Columns 5–8 the results based on the sample withbalanced panels, which consists of only those firms withdata available in both the IPO year and the non-IPO year.Our results in Columns 5–8 are similar to those reportedin Columns 1–4: a1 is positive and significant in all theregressions. Thus, our results in these regressions supporthypothesis H1. They suggest again that an IPO firm’sadvertising expenditures peak in the IPO year, even afterwe control for the change in product market factorsaround the IPO year.

ARTICLE IN PRESS

Table 5Change in advertising expenditures from non-IPO years to IPO Year: multivariate results. This table reports the results from the regressions on advertising

intensity of IPO firms in the IPO year and non-IPO years. Advertising intensity is advertising expenditures scaled by sales revenue in the same year. The

non-IPO year in regressions (1) and (5) is the year two years prior to the IPO year, year t�2. It is the year one year prior to the IPO year, year t�1, in

regressions (2) and (6); one year subsequent to the IPO year, year t+1, in regressions (3) and (7); and two years subsequent to the IPO year, year t+2, in

regressions (4) and (8). The independent variables consist of ISSUE, a dummy variable equal to one if a firm is in its IPO year and zero if a firm is in non-IPO

years; 1/SLS, where SLS is sales revenue in the current year; DLOGSLS, the change in the log of sales revenue from the past year to the current year; lagged

advertising intensity; and FSALE, fraction of a firm’s sales revenue to industry sales revenue in the past year. Industries are identified using two-digit SIC

codes. Regressions (1)–(4) use samples with unbalanced panels. Regressions (5)–(8) use samples with balanced panels. We run a random effect model for

all the regressions. Standard errors are reported in parentheses. *, **, and *** indicate significant difference from zero at the 10%, 5%, and 1% levels,

respectively

(1) (2) (3) (4) (5) (6) (7) (8)

Year t�2 Year t�1 Year t+1 Year t+2 Year t�2 Year t�1 Year t+1 Year t+2

Constant 0.045*** 0.004 0.010*** �0.005 0.050*** 0.007 0.008** 0.001

[0.006] [0.004] [0.003] [0.004] [0.007] [0.006] [0.003] [0.004]

ISSUE 0.010* 0.008** 0.009*** 0.012*** 0.014** 0.010** 0.011*** 0.014***

[0.005] [0.003] [0.003] [0.004] [0.006] [0.004] [0.003] [0.005]

1/SLS 0.015*** 0.010*** 0.013*** 0.013*** 0.032*** 0.016** 0.006** 0.006**

[0.002] [0.002] [0.002] [0.002] [0.008] [0.007] [0.002] [0.003]

DLOGSLS 0.007** 0.005* 0.014*** 0.009 0 0.011**

[0.003] [0.003] [0.003] [0.007] [0.004] [0.005]

FSALE �0.142 �0.164* �0.046 �0.139 �0.101 0.001

[0.097] [0.090] [0.082] [0.134] [0.092] [0.098]

Lagged advertising intensity 0.829*** 0.691*** 0.814*** 0.819*** 0.752*** 0.657***

[0.026] [0.023] [0.023] [0.039] [0.023] [0.033]

Balanced panel? No No No No Yes Yes Yes Yes

Number of observations 1,073 908 1,178 970 378 378 762 422

Number of firms 884 719 797 759 189 189 381 211

R-squared 0.04 0.68 0.67 0.67 0.05 0.73 0.69 0.54

T. Chemmanur, A. Yan / Journal of Financial Economics 92 (2009) 40–6556

5.2.2. Advertising intensity in matching firms

We also run regression Eq. (1) in a sample of matchingfirms in the IPO year t and non-IPO benchmark years. Inthis regression, a1 measures the difference in advertisingintensity between the IPO year and non-IPO years for anaverage matching firm. Because matching firms have noplan to issue any new equity in both the IPO year and anynon-IPO year, we expect a1 to be insignificant in thisregression.

Similarly, we run regressions for matching firms withyear t�2, t�1, t+1, or t+2 as the non-IPO benchmark year.We present the results from these new regressions inTable 6. The results reported in the first four columns arebased on the samples with unbalanced panels and theresults reported in the last four columns are based on thesamples with balanced panels. In the regressions withyear t�2 as the non-IPO year (reported in Columns 1 and 5),the coefficient of ISSUE, a1, is negative and significant.In all other regressions, a1 is insignificant. These resultssuggest that the peak of advertising in the IPO year t doesnot occur to matching firms.

5.3. Advertising and IPO underpricing as substitutes

In this section, we test hypothesis H2. Hypothesis H2predicts that advertising and underpricing serve assubstitutes when a firm uses them strategically to signalits firm type around its new equity issue. We test thishypothesis by estimating two regression equations withIPO underpricing, RET, and the amount of advertising inthe IPO year, ADVt/SLSt�1, as dependent variables. The

advertising variable we choose here is ADVt/SLSt�1, thelevel of advertising scaled by the lagged sales revenue,instead of advertising intensity ADVt/SLSt, the level ofadvertising scaled by the contemporary sales revenue.Every dollar of advertising expenditures can serve as asignal to both the product and financial markets. Adver-tising for product market purposes can also serve as asignal to the financial market due to the interactionbetween the financial market and the product market.Thus, if we scale advertising expenditures by the con-temporary sales revenue, we would exclude the level ofcontemporary product market advertising to some degree.In this case, our estimation of the substitution effectbetween advertising and underpricing would be biased.We avoid this problem by using ADVt/SLSt�1.

In the first regression equation (return equation here-after), we regress IPO underpricing, RET, on the amount ofadvertising in the IPO year, ADVt/SLSt�1, controlling for thefinancial market variables that affect the level of under-pricing according to the IPO literature. In particular, thereturn equation is

RET ¼ b0 þ b1

ADVt

SLSt�1þ b2FIN þ �. (2)

Here, FIN is a vector of financial market variablesconsisting of VENTURE, SYND, RANK, EXCHANGE, AMOUNT,BUBBLE, NIPO, MRUN, and REVISION. In this regression, wechoose not to control for the product market variables asin regression Eq. (1), because including these variablesdoes not add any explanatory power to the estimationwhile introducing the potential for multicollinearity with

ARTICLE IN PRESS

Table 6Change in advertising expenditures for the matching firms of the IPO sample: multivariate results. This table reports the results from regressions on

advertising intensity of the matching firms in the IPO year t and non-IPO years. Advertising intensity is advertising expenditures scaled by sales revenue in

the same year. A matching firm is constructed for each IPO firm following the method in Loughran and Ritter (1997), by industry, size, and sales revenue in

the IPO year t. The sample in regressions (1) and (5) consists of matching firms in year t and year t�2. The sample in regressions (2) and (6) consists of year

t-1 and year t. The sample in regressions (3) and (7) consists of year t+1 and year t; and the sample in regressions (4) and (8) consists of year t+2 and year t.

The independent variables consist of ISSUE, a dummy variable equal to one if the corresponding dependent variable is measured in the IPO year and zero if

measured in non-IPO years; 1/SLS, where SLS is sales revenue in the current year; DLOGSLS, the change in the log of sales revenue from the past year to the

current year; lagged advertising intensity; and FSALE, fraction of a firm’s sales revenue to industry sales revenue in the past year. Industries are identified

using two-digit SIC codes. Regressions (1)–(4) use samples with unbalanced panels. Regressions (5)–(8) use samples with balanced panels. We run a

random effect model for all the regressions. Standard errors are reported in parentheses. *, **, and *** indicate significant difference from zero at the 10%,

5%, and 1% levels, respectively.

(1) (2) (3) (4) (5) (6) (7) (8)

Year t�2 Year t�1 Year t+1 Year t+2 Year t�2 Year t�1 Year t+1 Year t+2

Constant 0.042*** 0.007*** 0.008*** 0.006*** 0.042*** 0.007*** 0.008*** 0.005***

[0.002] [0.001] [0.001] [0.001] [0.002] [0.001] [0.001] [0.001]

ISSUE �0.004*** 0.001 0.000 0.000 �0.004*** 0.002 0.001 0.000

[0.001] [0.001] [0.001] [0.001] [0.001] [0.001] [0.001] [0.001]

1/SLS 0.001*** �0.001*** �0.003*** 0.000 0.001** �0.001*** �0.004*** 0.000

[0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000]

DLOGSLS �0.002 �0.001 0.002 �0.002 �0.002 �0.003*

[0.001] [0.001] [0.001] [0.002] [0.001] [0.002]

FSALE �0.005 �0.03 �0.005 �0.007 �0.034* �0.006

[0.016] [0.019] [0.017] [0.015] [0.020] [0.017]

Lagged advertising intensity 0.775*** 0.789*** 0.818*** 0.789*** 0.800*** 0.869***

[0.013] [0.014] [0.013] [0.013] [0.015] [0.013]

Balanced panel? No No No No Yes Yes Yes Yes

Number of observations 1,577 1,520 1,585 1,400 1,386 1,366 1,418 1,066

Number of firms 884 837 876 867 693 683 709 533

R-squared 0.01 0.77 0.73 0.76 0.01 0.78 0.73 0.80

T. Chemmanur, A. Yan / Journal of Financial Economics 92 (2009) 40–65 57

ADVt/SLSt�1. Further, certain firm characteristics such asTECH, AGE, SIZEt�1, LDRt�1, and EBITt�1 could affect both afirm’s advertising policy and its decision in IPOs. Includingthese variables in regression Eq. (2) would, on the onehand, improve the quality of control and, on the otherhand, introduce the multicollinearity problem betweenthese variables and ADVt/SLSt�1. Thus, we present resultsfrom both the regressions controlling for these firmcharacteristics and the regressions without these controlsto demonstrate the robustness of our results. Underhypothesis H2, we expect the coefficient of ADVt/SLSt�1,b1, to be negative.

In the second regression equation (advertising equa-tion hereafter), we regress advertising expenditures (ADVt/SLSt�1) on IPO underpricing (RET), controlling for theproduct market factors that could affect the firm’sadvertising decision. The following is the advertisingequation we estimate in our test of H2:

ADVt

SLSt�1¼ b00 þ b01RET þ b02PRODþ �0. (3)

The set of product market variables PROD is similar tothat in regression Eq. (1), consisting of DLOGSLS, 1/SLSt�1,FSALEt�1, and ADVt�1/SLSt�1. Here, we choose not tocontrol for the set of financial variables FIN as inregression Eq. (2), because their explanatory power hasalready been incorporated into RET, and adding themwould only cause multicollinearity with RET. Similar to

the case in regression Eq. (2), we also choose to control forfirm characteristics such as TECH, etc. in some specifica-tions in addition to the control for PROD, to demonstratethe robustness of our results. Under hypothesis H2, weexpect b01 to be negative.

In the estimation of the above two equations, we startwith the ordinary least squares (OLS) approach. In both thereturn and the advertising regressions, we conduct signifi-cance tests using heteroskedasticity-consistent standarderrors following the Huber–White procedure. We presentthe results in Columns 1 and 2 of Table 7. In general, theresults in both specifications are similar. The overall fit ofboth the return and the advertising regressions is quite good,in view of the adjusted R2 of at least 30% for the returnregression (on RET) and 46% for the advertising regression(on ADVt/SLSt�1). Also as predicted, the coefficient of ADVt/SLSt�1 in the return regression (b1) is negative andsignificant at either the 5% or the 1% level. The coefficientof RET in the advertising regression (b01) is negative as welland significant at the 1% level. For example, according to thereturn regression in Column 2, an increase in ADVt/SLSt�1 byone standard deviation is associated with a decrease in RET

by 0.105 times standard deviation (2.9%). However, accord-ing to the advertising regression in Column 2, an increase inRET by one standard deviation is associated with a decreasein ADVt/SLSt�1 by 0.111 times standard deviation (16% of thelagged sales revenue). Thus, our results here are consistentwith hypothesis H2, suggesting that a firm uses advertising

ARTICLE IN PRESS

Table 7Substitution effect between IPO underpricing and product market advertising. This table presents the results from ordinary least square (OLS) and

seemingly unrelated (SURE) regressions. The limited sample consists of firms with data available in both years t and t�2. The dependent variable is RET,

the first day equity return, or ADVt/SLSt�1, the amount of advertising expenditures in year t divided by sales revenue in year t�1. The independent variables

include ADVt�1/SLSt�1; DLOGSLS, change in the log of sales revenue from year t�1 to year t; AMOUNT, IPO proceeds scaled by the book value of assets in

year t�1; SYND, a dummy for syndicated underwriting; SIZEt�1, the log of the book value of assets; EBITt�1, earnings before interest, taxes, depreciation and

amortization scaled by the book value of assets; TECH, a dummy for high-technology firms; RANK, the proxy for underwriter ranks; VENTURE, a dummy for

issuers backed by venture capitalists; EXCHANGE, a dummy for firms listed in NYSE or AMEX; LDRt�1, the ratio of long-term debt to the book value of

assets; FSALE, fraction of a firm’s sales revenue to industry sales revenue in year t�1; BUBBLE, a dummy for years 1999 and 2000; MRUN, the return of an

equally weighted market index 60 days prior to the IPO date; NIPO, the log of the number of IPO deals; AGE, the difference between the IPO year and the

founding year; and REVISION, the change from the mid-point of filing range to IPO offer price scaled by IPO offer price. Standard errors are provided in

parentheses. *, **, and *** indicate significant difference from zero at the 10%, 5%, and 1%, respectively. Coefficients of EXCHANGE, RANK, VENTURE, AGE,

FSALE, and MRUN are insignificant and not reported.

(1) OLS with the full sample (2) OLS with the full sample (3) OLS with the limited sample (4) SURE with the full sample

Variables RET ADVt/SLSt�1 RET ADVt/SLSt�1 RET ADVt/SLSt�1 RET ADVt/SLSt�1

Constant �0.077 �0.382*** 0.061 �0.313*** �0.505 �0.521*** 0.079 �0.209*

[0.169] [0.061] [0.176] [0.114] [0.573] [0.137] [0.174] [0.113]

ADVt/SLSt�1 �0.015** �0.021*** �0.054** �0.027***

[0.006] [0.007] [0.024] [0.007]

RET �0.489*** �0.562*** �0.438*** �0.780***

[0.145] [0.152] [0.134] [0.152]

SYND �0.072** �0.061** �0.055 �0.052*

[0.030] [0.031] [0.067] [0.031]

AMOUNT 0.003** 0.003 0.011* 0.002

[0.001] [0.002] [0.006] [0.002]

REVISION �0.199*** �0.198*** �0.223*** 0.199***

[0.019] [0.019] [0.043] [0.027]

BUBBLE 0.239*** 0.195*** 0.132*** 0.028

[0.024] [0.027] [0.050] [0.028]

NIPO 0.051* 0.029 0.126 �0.198***

[0.027] [0.028] [0.095] [0.019]

ADVt�1/SLSt�1 7.380*** 7.470*** 5.565*** 7.146***

[0.497] [0.553] [0.521] [0.543]

DLOGSLS 0.814*** 0.782*** 0.334*** 0.709***

[0.069] [0.074] [0.109] [0.073]

1/SLSt�1 0.023*** 0.022*** 2.118*** 0.023***

[0.006] [0.006] [0.280] [0.006]

TECH 0.062*** �0.044 0.073 0.044 0.061*** �0.033

[0.021] [0.090] [0.052] [0.099] [0.021] [0.089]

SIZEt�1 0.005 0.001 0.009 0.060* 0.003 �0.004

[0.011] [0.033] [0.022] [0.033] [0.011] [0.033]

EBITt�1 �0.034 �0.137 �0.092 0.062 �0.041 �0.214*

[0.026] [0.111] [0.062] [0.115] [0.025] [0.110]

LDRt�1 �0.113*** �0.293* �0.159* �0.075 �0.118*** �0.315*

[0.038] [0.165] [0.081] [0.155] [0.037] [0.162]

Number of obs. (R2) 706 719 682 698 177 187 675 675

(0.301) (0.467) (0.339) (0.476) (0.421) (0.657) (0.337) (0.462)

16 It is worth emphasizing that our objective in presenting results

from this simultaneous equation estimation technique is not to identify

any dynamic causal relation between advertising and underpricing. Our

hypothesis H2 predicts only that the two should be negatively correlated

in equilibrium.

T. Chemmanur, A. Yan / Journal of Financial Economics 92 (2009) 40–6558

and IPO underpricing as substitutes to signal to the productand financial markets.

We then conduct two robustness checks on the aboveresults. First, we run OLS regressions based on a sample offirms with the data on advertising expenditures availablein the pre-IPO years t�1 and t�2. In this robustness check,we intend to verify that our results on the substitutabilitybetween advertising and underpricing are not driven bythe change in the composition of firms over years. Theresults from this robustness check are presented inColumn 3 of Table 7. Again, we find that both b1 and b01are negative and significant, thereby supporting hypoth-esis H2.

In the second robustness check, we use a seeminglyunrelated estimation (SURE) approach to account for thepotential simultaneity problem. In our theoretical frame-

work, a firm chooses both the level of underpricing andthe amount of advertising simultaneously to signal itsfirm type. Thus, both ADVt/SLSt�1 and RET could beendogenous variables and the OLS estimator could sufferfrom this simultaneity problem. We use the SUREregression technique to address this issue.16 In the SUREestimation, we rely on the covariance matrix estimatedfrom the OLS results and estimate the return Eq. (2) andthe advertising Eq. (3) simultaneously. The results arepresented in Column 4 of Table 7. Our results in the SURE

ARTICLE IN PRESS

T. Chemmanur, A. Yan / Journal of Financial Economics 92 (2009) 40–65 59

estimation are stronger than those in the OLS estimation.Again, b1 and b01 are negative and significant at the 1%level, thereby confirming the robustness of our earlier OLSresults.

In addition to the SURE regression technique, there areother alternatives to estimate simultaneous equations,such as the 2SLS approach and the 3SLS approach. It iswell known that, asymptotically, 2SLS dominates OLS andfull-information estimators such as 3SLS and SURE aremore efficient than limited-information estimators suchas 2SLS and OLS. However, the small-sample properties ofthese estimators are not clear. In a small sample, 2SLS issensitive to outliers and the choice of instrumentalvariables, thus the net benefit of using 2SLS is ambiguous.Also, the benefit of using system methods (SURE and 3SLS)instead of single-equation methods is modest in a smallsample, because the finite-sample variation of distur-bances or specification errors could be transmittedbetween the two equations in a joint estimation and thusreduce the efficiency of estimation. Given that each of theabove estimation techniques has its own individuallimitations in a small sample, we choose to provide onlythe results based on the SURE regression technique. Theresults based on the 2SLS and the 3SLS techniques aresimilar to the results presented here. These results areavailable in the working paper version of this article,available on the authors’ websites and at ssrn.com.

In sum, our results in all regressions support hypoth-esis H2. They suggest that advertising and IPO under-pricing can serve as substitutes for a firm to signal its firmtype to outsiders. These results are robust to variousestimation techniques such as the SURE estimationtechnique, and various compositions of samples.17

6. Empirical results for SEO firms

We study a sample of SEO firms, following the samesteps as in our earlier study on IPO firms. We start withthe test on hypothesis H1 by comparing a firm’s advertis-ing policy when the firm plans to issue seasoned equitywith a benchmark advertising policy when the firm doesnot have any plan to issue new equity.

6.1. Change in advertising expenditures around SEOs:

univariate tests

We first compare a firm’s advertising expenditures inits SEO year t with the firm’s advertising expenditures inits non-SEO years, consisting of years t�2, t�1, t+1, andt+2. We select only those non-SEO years in which no newequity is issued, i.e., the firm-year is not covered in ourSEO and IPO samples and the firm’s sales of common and

17 Our results from the OLS and SURE regressions also show that a

firm underprices more in its IPO when it is a high-tech firm, when its

issue is not syndicated, and when its IPO occurs in the bubble period in

years 1999 and 2000. These results are consistent with the earlier

findings in the IPO literature (e.g., Ritter Loughran, 2004). We also find

that a firm advertises more when a firm’s sales growth is higher (with a

larger LOGSLS) or when it is more financially constrained (with a smaller

LDRt�1).

preferred stock is less than $2 million according toCompustat. By doing so, we ensure that a firm’s advertis-ing policy in the selected non-SEO years is unaffected byits decision to issue new equity.

Similar to our study on the advertising in the IPO year,we use advertising intensity to measure a SEO firm’sadvertising policy. Column 1 of Panel A, Table 8, reportsthe sample statistics of advertising intensity in the yearsaround the SEO year based on the whole sample and thepair-wise differences in advertising intensity between theSEO year and the non-SEO years. Our results show that afirm’s advertising intensity in its SEO year ADVt/SLSt isgreater than those in years t�2, t�1, and t+2, i.e., ADVt�2/SLSt�2, ADVt�1/SLSt�1, and ADVt+2/SLSt+2. Both (ADVt/SLSt)�(ADVt�1/SLSt�1) and (ADVt/SLSt)�(ADVt�2/SLSt�2) aresignificant in the t-test and the Wilcoxon test. (ADVt/SLSt)�(ADVt+2/SLSt+2) is positive and significant in thet-test as well. These results are consistent with hypothesisH1. However, (ADVt/SLSt)�(ADVt+1/SLSt+1) is negative andsignificant. Thus, seasoned equity issues in the SEO year t

could have a lingering effect on a firm’s advertising policyin year t+1 and this effect levels off in year t+2.

Second, we test hypothesis H1 by comparing theadvertising intensity around the SEO year between SEOfirms and their matching firms. Column 2 of Panel A,Table 8, presents the level and the change in advertisingintensity for the matching firms around year t. Column 3of Panel A, Table 8, presents the differences between SEOfirms and their matching firms. We find that from eitheryear t�2 or year t�1 to the SEO year t, SEO firmsexperience a larger increase in their advertising intensitythan their matching firms. Further, from the SEO year t toyear t+2, SEO firms experience a larger decrease in theiradvertising intensity than their matching firms. Theseresults are consistent with hypothesis H1. However, fromthe SEO year t to year t+1, SEO firms experience a largerincrease in advertising intensity than matching firms do,probably because the increase in SEO firms’ advertising inthe SEO year has some lingering effect on the yearsubsequent to the SEO year.

Following the steps in our earlier study of IPO firms, wealso conduct three robustness checks on our univariateresults of SEO firms. First, we study the change inadvertising intensity for those SEO firms surviving bothyears t�1 and t�2 and for those surviving both years t+1and t+2. In this robustness check, we intend to findwhether our results on advertising around the SEO yearare driven by the different compositions of firms indifferent years. The results of this robustness check arepresented in Panels B and C of Table 8. They are similar tobut weaker than the results presented in Panel A. In thesecond robustness check, we disaggregate SEO firms intosubsamples of entering firms and surviving firms based ondata availability in the previous year. We also disaggregateSEO firms into subsamples of exiting firms and survivingfirms based on data availability in the subsequent year.We then compare surviving firms with both entering firmsand exiting firms to check whether our earlier results onSEO firms are driven by survivorship bias. The results fromthe second robustness check are presented in Panels D andE of Table 8. In general, we find that the change in

ARTICLE IN PRESS

Table 8Summary statistics of the change in advertising expenditures between the SEO year and non-SEO years: univariate results. Panel A reports the level and

the change of advertising intensity (ADV/SLS) in the five years around the SEO year. ADV/SLS is advertising expenditures scaled by sales revenue in the

same year. A matching firm is constructed for each SEO firm following the method in Loughran and Ritter (1997), by industry, size, and sales revenue in

the SEO year t. Panel B reports the change in ADV/SLS for SEO firms surviving both years t�1 and t�2 and their matching firms. Panel C reports the change

in ADV/SLS for SEO firms surviving year t+1 and year t+2 and their matching firms. Panel D is based on SEO firms surviving the past year (surviving firms)

and SEO firms not reporting advertising in the past year (entering firms). Panel E is based on SEO firms surviving the subsequent year (surviving firms) and

SEO firms not reporting advertising in the subsequent year (exiting firms). In Panel F, we create subsamples by comparing a matching firm’s EBITt�1,

calculated as earning before interest, taxes, depreciation and amortization scaled by the book value of assets, with the sample median (high and low

EBITt�1) and by comparing a matching firm’s MBt�1, the ratio of the market value to the book value of assets, with the sample median (high and low

MBt�1). Tests on means and medians are based on t-tests and Wilcoxon tests, respectively. *, **, and *** indicate significant difference from zero at the 10%,

5%, and 1%, respectively.

(1) SEO firms (2) Matching firms (3) SEO firms–matching firms

Number of

observations

Mean Median Number of

observations

Mean Median Number of

observations

Mean Median

Panel A. Pair-wise difference in advertising intensity: SEO firms versus matching firms

ADVt�2/SLSt�2 311 0.025 0.014 540 0.026 0.014 311 0.001 0.000

ADVt�1/SLSt�1 286 0.023 0.013 587 0.026 0.014 286 �0.001 0.000

ADVt/SLSt 663 0.031 0.015 663 0.025 0.013 663 0.005*** 0.000**

ADVt+1/SLSt+1 256 0.031 0.014 535 0.025 0.015 256 0.005 0.000

ADVt+2/SLSt+2 191 0.024 0.015 417 0.028 0.017 191 �0.003 0.000

ADVt/SLSt–ADVt�2/SLSt�2 311 0.003* 0.000 540 �0.001** 0.000*** 311 0.004** 0.000**

ADVt/SLSt–ADVt�1/SLSt�1 286 0.001* 0.000 587 �0.001 0.000*** 286 0.002** 0.000**

ADVt/SLSt–ADVt+1/SLSt+1 256 �0.003** 0.000*** 535 0.001*** 0.000*** 256 �0.004*** �0.001***

ADVt/SLSt–ADVt+2/SLSt+2 191 0.006*** 0.000** 417 0.001 0.000 191 0.005** 0.000

Panel B. Change in advertising intensity: SEO firms and matching firms surviving two years prior to the SEO year

ADVt/SLSt–ADVt�2/SLSt–2 177 0.002 0.000 177 �0.002* 0.000*** 177 0.003* 0.000**

ADVt/SLSt–ADVt�1/SLSt�1 177 0.001 0.000 177 �0.001 0.000*** 177 0.002* 0.000

Panel C. Change in advertising intensity: SEO firms and matching firms surviving two years subsequent to the SEO year

ADVt/SLSt–ADVt+1/SLSt+1 137 �0.004*** 0.000** 137 0.002* 0.000** 137 �0.006*** �0.001***

ADVt/SLSt–ADVt+2/SLSt+2 137 0.004** 0.000 137 0.002 0.000** 137 0.002 0.000

Panel D. Change in advertising intensity: SEO firms entering the sample versus SEO firms surviving the past year

(1) Entering SEO firms (2) Surviving SEO firms (3) Entering–surviving firms

Number Mean Median Number Mean Median Number Mean Median

ADVt/SLSt–ADVt�1/SLSt�1 109 0.002* 0.000 177 0.001 0.000 286 0.002 0.000

ADVt/SLSt–ADVt+1/SLSt+1 108 �0.003 �0.000*** 148 �0.002* 0.000 256 �0.001 0.000

ADVt/SLSt–ADVt+2/SLSt+2 91 0.010*** 0.001* 100 0.003** 0.000 191 0.007* 0.001

Panel E. Change in advertising intensity: SEO firms exiting the sample versus SEO firms surviving the subsequent year

(1) Exiting SEO firms (2) Surviving SEO

firms

(3) Exiting–surviving firms

Number Mean Median Number Mean Median Number Mean Median

ADVt/SLSt–ADVt�2/SLSt�2 166 0.001 0.000 146 0.004* 0.000 311 �0.003 0.000

ADVt/SLSt–ADVt�1/SLSt�1 138 0.001 0.000 148 0.002** 0.000 286 �0.001 0.000

ADVt/SLSt–ADVt+1/SLSt+1 119 �0.002 0.000* 137 �0.004*** 0.000** 256 0.002 0.000

Panel F. Change in advertising intensity of matching firms: ADVt/SLSt–ADVt�1/SLSt�1

Number Mean Median Number Mean Median

Firms with high MBt�1 277 �0.000 �0.000* Firms with high EBITt�1 286 �0.000 0.000

Firms with low MBt�1 296 �0.001** �0.000*** Firms with low EBITt�1 301 �0.001 �0.000***

Difference (high

MBt�1�low MBt�1)

573 �0.001 0.000 Difference (high EBITt�1�low EBITt�1) 587 0.001 0.001***

T. Chemmanur, A. Yan / Journal of Financial Economics 92 (2009) 40–6560

ARTICLE IN PRESS

T. Chemmanur, A. Yan / Journal of Financial Economics 92 (2009) 40–65 61

advertising intensity of surviving firms is not significantlydifferent compared with entering firms and exiting firms,thereby rejecting the possibility of survivorship bias.Finally, we also study the difference in advertising growthbetween matching firms with high firm growth (highMTOBt�1) and low firm growth (low MTOBt�1) andbetween matching firms with high firm profitability (highEBITt�1) and low firm profitability (low EBITt�1). Wepresent the results from this robustness check in Panel Fof Table 8. The results are similar to those based on thematching firms of the IPO sample. In general, theadvertising growth of the matching firms is positivelyrelated to firm profitability but unrelated to growthoptions. Under the same rationale discussed in Section5.1.3, these evidence rejects the possibility that SEO firms’advertising growth in their SEO years is driven by firmgrowth or firm profitability.

6.2. Change in advertising expenditures around SEOS:

multivariate analysis

In this section, we test hypothesis H1 by runningregression Eq. (1). Similar to our regressions based on theIPO sample, we first pool the SEO year with non-SEO yearsfor each SEO firm, so that the dependent variable(advertising intensity) is measured for either the SEOyear or non-SEO years. The dummy variable ISSUE equalsone if the dependent variable is measured in the SEO yearand zero if measured in non-SEO years. The coefficient of

Table 9Change in advertising expenditures from non-SEO years to SEO year: multivariat

intensity of SEO firms in the SEO year and non-SEO years. Advertising intensity

non-SEO year in regressions (1) and (5) is the year two years prior to the SEO

regressions (2) and (6); one year subsequent to the SEO year, year t+1, in regres

regressions (4) and (8). The independent variables consist of ISSUE, a dummy var

the SEO year and zero if measured in a non-SEO year; 1/SLS, where SLS is sales re

from the past year to the current year; lagged advertising intensity; and FSALE, f

Industries are identified using two-digit SIC codes. Regressions (1)–(4) use s

balanced panels. We run a random effect model for all the regressions. Stand

difference from zero at the 10%, 5%, and 1% levels, respectively.

(1) (2) (3)

Year t�2 Year t�1 Year t+1

Constant 0.002 0.001 0.004**

[0.002] [0.001] [0.002]

ISSUE 0.004*** 0.002** 0.001

[0.001] [0.001] [0.001]

1/SLS 0.008 0.058*** 0.025

[0.016] [0.021] [0.022]

DLOGSLS �0.002 0.000 �0.011***

[0.002] [0.002] [0.002]

FSALE 0.001 0.002 �0.003

[0.010] [0.010] [0.012]

Lagged advertising intensity 0.820*** 0.881*** 0.998***

[0.024] [0.024] [0.025]

Balanced panel? No No No

Number of observations 488 529 542

Number of firms 350 286 394

R-squared 0.81 0.86 0.84

ISSUE, a1, measures the average difference in advertisingintensity between the SEO year and non-SEO years.Hypothesis H1 predicts a1 to be positive. In this regres-sion, we use both the sample with unbalanced panels andthe sample with balanced panels, and we control for thecross-sectional differences across firms by running arandom effect model. The results from this regressionare presented in Table 9 with year t�2, t�1, t+1, or t+2 asthe non-SEO year, respectively. As expected, the coeffi-cient of ISSUE, a1, is positive and significant in theregressions with year t�2, t�1, or t+2 as the non-SEOyear. For example, according to the regressions withunbalanced panels (as shown in Columns 1 and 2), aSEO firm increases its advertising expenditures by 0.4% ofits sales revenue from year t�2 to the SEO year t and by0.2% of its sales revenue from year t�1 to year t. Column 4further shows that a SEO firm decreases its advertisingexpenditures by 0.5% of its sales revenue from the SEOyear t to year t+2. However, while a1 in the regression withyear t+1 as the non-SEO year is positive (as expected), it isinsignificant.

We also run regression Eq. (1) for a sample of matchingfirms. Because matching firms have no plan to issue anynew equity in either the SEO year or the non-SEO years, weexpect a1 to be insignificant in the new regression. Theresults based on the matching firms of the SEO sample arepresented in Table 10. As expected, in all the regressionswith year t�2, t�1, t+1, or t+2 as the non-SEO year, a1 isinsignificant. Thus, the advertising intensity of the matchingfirms in the SEO year is similar to that in the non-SEO years.

e results. This table reports the results from the regressions on advertising

is advertising expenditures scaled by sales revenue in the same year. The

year, year t�2. It is the year one year prior to the SEO year, year t�1, in

sions (3) and (7); and two years subsequent to the SEO year, year t+2, in

iable equal to one if the corresponding dependent variable is measured in

venue in the current year; DLOGSLS, the change in the log of sales revenue

raction of a firm’s sales revenue to industry sales revenue in the past year.

amples with unbalanced panels. Regressions (5)–(8) use samples with

ard errors are reported in parentheses. *, **, and *** indicate significant

(4) (5) (6) (7) (8)

Year t+2 Year t�2 Year t�1 Year t+1 Year t+2

�0.002 0 0.002 0.003** 0

[0.002] [0.001] [0.001] [0.001] [0.002]

0.005*** 0.004*** 0.002* 0.002 0.005***

[0.001] [0.001] [0.001] [0.001] [0.002]

0.001 �0.038** 0.025 0.000 0.019

[0.027] [0.017] [0.020] [0.019] [0.030]

0.004 �0.002 �0.002 �0.018*** �0.003

[0.002] [0.002] [0.002] [0.003] [0.003]

0.015 0.002 0.001 �0.005 �0.002

[0.010] [0.007] [0.009] [0.009] [0.008]

0.846*** 0.899*** 0.884*** 1.083*** 0.902***

[0.025] [0.020] [0.022] [0.023] [0.029]

No Yes Yes Yes Yes

473 276 486 296 198

374 138 243 148 99

0.77 0.88 0.87 0.92 0.84

ARTICLE IN PRESS

Table 10Change in advertising expenditures of matching firms for the SEO sample: multivariate results. This table reports the results from regressions on

advertising intensity of the matching firms in the SEO year t and non-SEO years. Advertising intensity is advertising expenditures scaled by sales revenue

in the same year. A matching firm is constructed for each SEO firm following the method in Loughran and Ritter (1997), by industry, size, and sales

revenue in the SEO year t. The sample in regressions (1) and (5) consists of matching firms in year t�2 and year t. The sample in regressions (2) and (6)

consists of year t�1 and year t. The sample in regressions (3) and (7) consists of year t+1 and year t; and the sample in regressions (4) and (8) consists of

year t+2 and year t. The independent variables consist of ISSUE, a dummy variable equal to one if the corresponding dependent variable is measured in a

SEO year and zero if measured in a non-SEO year; 1/SLS, where SLS is sales revenue in the past year; DLOGSLS, the change in the log of sales revenue from

the past year to the current year; lagged advertising intensity; and FSALE, fraction of a firm’s sales revenue to industry sales revenue in the past year.

Industries are identified using two-digit SIC codes. Regressions (1)–(4) use samples with unbalanced panels. Regressions (5)–(8) use samples with

balanced panels. We run a random effect model for all the regressions. Standard errors are reported in parentheses. *, **, and *** indicate significant

difference from zero at the 10%, 5%, and 1% levels, respectively.

(1) (2) (3) (4) (5) (6) (7) (8)

Year t�2 Year t�1 Year t+1 Year t+2 Year t�2 Year t�1 Year t+1 Year t+2

Constant 0.000 0.001** 0.001* 0.001** 0.000 0.001 0.000 0.001

[0.001] [0.001] [0.001] [0.001] [0.001] [0.001] [0.002] [0.002]

ISSUE 0.000 0.001 0.000 �0.001 0.000 0.001 0.002 0.000

[0.001] [0.001] [0.001] [0.001] [0.001] [0.001] [0.002] [0.002]

1/SLS 0.016*** 0.010* 0.013* 0.024*** 0.011 0.005 0.082*** 0.060***

[0.005] [0.006] [0.007] [0.007] [0.010] [0.008] [0.022] [0.019]

DLOGSLS �0.003** �0.005*** �0.003 �0.008*** �0.004 �0.007*** 0.002 �0.002

[0.001] [0.001] [0.002] [0.002] [0.003] [0.002] [0.004] [0.005]

FSALE 0.003 0.004 0.000 �0.001 0.000 0.000 �0.001 �0.001

[0.003] [0.003] [0.004] [0.004] [0.006] [0.005] [0.009] [0.010]

Lagged advertising intensity 0.960*** 0.893*** 0.906*** 0.943*** 0.975*** 0.875*** 0.902*** 0.869***

[0.010] [0.010] [0.011] [0.011] [0.021] [0.016] [0.027] [0.032]

Balanced panel? No No No No Yes Yes Yes Yes

Number of observations 1,012 1,120 998 998 296 418 260 252

Number of films 858 907 864 870 148 209 130 126

T. Chemmanur, A. Yan / Journal of Financial Economics 92 (2009) 40–6562

In general, both our multivariate results and theunivariate results for SEO firms support hypothesis H1.However, they are weaker compared with our earlierresults for IPO firms. One possible reason for the weakerresults in the context of SEOs compared with IPOs is thatfirms making SEOs, being already public, suffer from onlya smaller extent of information asymmetry comparedwith private firms going public, so that we would expectSEO firms’ advertising policy to be less targeted at theirfinancial market investors and more targeted at theirproduct market customers (compared with a privatefirm’s advertising policy when it plans to go public).

18 According to the results from the return regression, a firm

underprices more in its SEO when the reputation of its underwriter is

lower. According to the results from the advertising regression, a firm

advertises more during its SEO year when the firm’s size is larger, when

the firm’s sales growth is higher, and when the firm has less liabilities.

6.3. Advertising and SEO underpricing as substitutes

We study the substitution effect between advertisingand SEO underpricing, based on the return Eq. (2) and theadvertising Eq. (3). Similar to the steps in the study on IPOfirms, here we first run regressions with only productmarket variable PROD as the independent variable in theadvertising regression and the financial market variableFIN as the independent variable in the return regression.We then run regressions including the firm characteristicsthat could be related to both product market considera-tions and financial market considerations. To supporthypothesis H2, we expect both the coefficient of ADVt/SLSt�1 in the return regression (b1) and the coefficient ofRET in the advertising regression (b01) to be negative.

We present the regression results in Table 11. Columns1 and 2 of Table 11 present the results based on OLSregressions. As expected, both b1 and b01 are negative in allspecifications. However, most coefficients are insignificantexcept b1 in the return regression in Column 1, which issignificant at the 5% level. We also use the SUREregression technique to control for the potential problemof simultaneity. We present the results from the SUREregressions in Columns 3 and 4 of Table 11. We find thatboth b1 and b01 are negative and significant in theregressions in Column 3. However, in Column 4, after wecontrol for firm characteristics such as SIZEt�1, b1 and b01become insignificant though remaining negative.18

Overall, our results on the substitution effect betweenproduct market advertising and underpricing are weakerin the context of SEOs relative to those in the context ofIPOs. This is to be expected for two reasons. First, firmsmaking SEOs, being already public, suffer from only asmaller extent of information asymmetry compared withprivate firms going public, so that they face only a smallerneed to signal to the financial markets either throughadvertising or underpricing. Second, firms making SEOs,being already public, are less financially constrained than

ARTICLE IN PRESS

Table 11Test of substitution effect between SEO underpricing and product market advertising. This table presents the results from ordinary least squares (OLS) and

seemingly unrelated (SURE) regressions. The dependent variable is RET, the first day equity return, or ADVt/SLSt�1, the amount of advertising expenditures

in the SEO year t divided by sales in year t�1. The independent variables include DLOGSLS, change in the log of sales revenue from year t�1 to year t;

AMOUNT, the amount of money raised from SEO scaled by the book value of assets in year t�1; SYND, a dummy for syndicated underwriting; SIZEt�1, the

log of the book value of assets; EBITt�1, earnings before interest, taxes, deprecation and amortization scaled by the book value of assets; TECH, a dummy for

high technology firms; RANK, the proxy for underwriter ranks; EXCHANGE, a dummy for firms listed in NYSE or AMEX; MBt�1, the ratio of the market value

to the book value of assets; LDRt�1, the ratio of long-term debt to the book value of assets; VENTURE, a dummy for issuers backed by venture capitalists;

REVISION, the change from the mid point of filing range to SEO price scaled by IPO offer price; MRUN, return of an equally weighted market index during

60 days prior to the SEO issuing date; BUBBLE, a dummy for years 1999 and 2000; NIPO, the log of the number of IPO deals; and FSALE, fraction of a firm’s

sales revenue to industry sales revenue in the past year. Standard errors are provided in parentheses. *, **, and *** indicate significant difference from zero

at the 10%, 5%, and 1% levels, respectively. Coefficients of EXCHANGE, TECH, SYND, MRUN, BUBBLE, MTOB, REVISION, and EXCHANGE are insignificant and not

reported.

(1) OLS (2) OLS (3) SURE (4) SURE

Variable RET ADVt/SLSt�1 RET ADVt/SLSt�1 RET ADVt/SLSt�1 RET ADVt/SLSt�1

Constant 0.008 �0.010*** �0.014 �0.027*** 0.008 �0.009*** �0.014 �0.026***

[0.038] [0.003] [0.040] [0.010] [0.037] [0.003] [0.039] [0.010]

ADVt/SLSt�1 �0.056** �0.033 �0.073*** �0.041

[0.028] [0.030] [0.028] [0.030]

RET �0.022 �0.002 �0.063* �0.016

[0.038] [0.039] [0.038] [0.039]

RANK �0.004*** �0.003* �0.004*** �0.003*

[0.001] [0.002] [0.001] [0.002]

AMOUNT 0.005 0.004 0.005* 0.004

[0.003] [0.005] [0.003] [0.005]

VENTURE 0.005 �0.007 0.005 �0.007

[0.009] [0.010] [0.009] [0.010]

NIPO 0.008 0.009 0.008 0.01

[0.006] [0.006] [0.006] [0.006]

ADVt�1/SLSt�1 1.422*** 1.452*** 1.428*** 1.443***

[0.046] [0.048] [0.047] [0.049]

DLOGSLS 0.056*** 0.048*** 0.054*** 0.045***

[0.004] [0.005] [0.005] [0.005]

1/SLSt�1 0.001 �0.022 0.003 �0.019

[0.024] [0.025] [0.024] [0.025]

FSALE �0.067** �0.042 �0.048* �0.046

[0.028] [0.045] [0.028] [0.045]

SIZEt�1 �0.001 0.004*** �0.001 0.004***

[0.002] [0.001] [0.002] [0.001]

EBITt�1 �0.003 0.012 �0.003 0.008

[0.014] [0.015] [0.014] [0.016]

LDRt�1 0.01 �0.018** 0.009 �0.018**

[0.010] [0.009] [0.010] [0.009]

Number of obs. (R2) 584 602 530 541 584 584 530 530

(0.036) (0.700) (0.031) (0.708) (0.035) (0.688) (0.031) (0.697)

T. Chemmanur, A. Yan / Journal of Financial Economics 92 (2009) 40–65 63

private firms going public, so that they are less likely tosubstitute underpricing for advertising (consistent withthe prediction of our theoretical analysis).

7. Alternative explanations

In this section, we briefly discuss how our empiricalresults could be explained by considerations other thanthe signaling argument we develop in Sections 2 and 3.19

One such alternative explanation is that, instead of anincrease in advertising being caused by the desire of thefirm to signal to the financial market and make an IPO, thecausality could be the other way around. The firm couldundergo an increase in sales growth that causes it to

19 We thank an anonymous referee for suggesting these alternative

explanations.

increase its advertising. The above increase in salesgrowth could also prompt the firm to go public, asdemonstrated in the empirical literature on the goingpublic decisions (see, e.g., Pagano, Panetta, and Zingales,1998). In other words, it could be an increase in salesgrowth that simultaneously causes the firm to go publicand increase product market advertising. One problemwith this explanation is that it is inconsistent with ourempirical finding that advertising peaks in the IPO year. Ifsales growth were causing the firm to increase advertis-ing, one would expect the firm to continue to increaseadvertising at least for one or two years after the IPO year.It is also worth noting that our results continue to holdeven after controlling for sales growth in our multivariateanalysis of hypothesis H1.

A second potential (though less plausible) explanationis that our results are driven by certain industriescharacterized by high levels of advertising (for example,

ARTICLE IN PRESS

T. Chemmanur, A. Yan / Journal of Financial Economics 92 (2009) 40–6564

industries with certain products having a first-moveradvantage in the product market) and that are, for somereason, more likely to go public as well. We attempt torule out such explanations by following the industrialorganization literature and controlling for various indus-try variables (e.g., FSALE, the fraction of a firm’s salesrevenue to industry sales revenue) in our multivariate testof hypothesis H1. In our test of hypothesis H2, we also ranregressions controlling for heteroskedasticity across dif-ferent industries, in which we define industries based ontwo-digit SIC codes. The results from these regressions aresimilar to those presented here.

Finally, a third explanation could arise from behavioralnotions that product market advertising around a firm’sIPO might generate a media buzz among retail investorsregarding the IPO, inducing them to buy the IPO firm’sequity at a high price on the first trading day (thusgenerating excess demand for it and further driving up itsprice). However, such behavioral arguments would predictthat a higher level of advertising around a firm’s IPO isassociated with larger initial returns for its equity, whichis directly contradicted by our empirical finding thathigher levels of advertising are associated with lowerlevels of initial returns.

In summary, while each of the three potential alter-native explanations is consistent with some of ourempirical findings, they seem to be unable to explain allour empirical findings simultaneously. In contrast, theimplications of the signaling argument that we develop inSections 2 and 3 are consistent with all three of our mainempirical findings; namely, the increase in advertisinglevel in both the IPO year and the SEO year compared withprior years; the peaking of advertising in the IPO year (inparticular, the decline in advertising from the IPO year tosubsequent years); and the fact that advertising andunderpricing are substitutes for firms going public.However, as is true for many phenomena in corporatefinance, one or more of the alternative explanations couldcoexist along with the signaling explanation we focus onin this paper.

8. Conclusion

Practitioners note that firms tend to increase theirproduct market advertising prior to an IPO or a seasonedequity issue. In this paper, we present a theory of theinteraction between a firm’s product market advertisingand its corporate financing decisions in this context. Weconsider a firm that faces asymmetric information in boththe product and the financial markets (about the qualityof its products and the intrinsic value of its projects) andthat needs to raise external financing to fund its growthopportunity (new project). Any product market advertis-ing undertaken by the firm is visible in the financialmarket as well. In equilibrium, the firm uses a combina-tion of product market advertising, equity underpricing,and underfinancing (raising a smaller amount of externalcapital than the full information optimum) to convey itstrue product quality and the intrinsic value of its projectsto consumers and investors.

Our theoretical analysis generates two sets of testablepredictions for firms making new equity issues. First,firms choose a higher level of product market advertisingwhen they are planning to issue new equity, comparedwith situations in which they have no such plan to makean equity issue. Second, product market advertising andequity underpricing are substitutes for a firm going public.We test the above two predictions of our theoreticalanalysis on two samples of firms making IPOs and SEOs,respectively. The empirical evidence supports these pre-dictions. First, firms increase their product marketadvertising in their IPO years or SEO years relative tonon-IPO and non-SEO years when they have no plan toissue new equity. In other words, in the five year spanaround the equity issue year (from two years before theequity issue to two years after), the peak advertising levelis reached in the equity issue year. We also find that IPOfirms and SEO firms experience a greater increase in theiradvertising expenditures in the IPO year or the SEO year,compared with their matching firms. Second, the extent ofunderpricing is smaller as the level of product marketadvertising is greater, both for IPO and SEO firms. In otherwords, firms make use of product market advertising andequity underpricing as substitutes in signaling to thefinancial market around a new equity issue.

References

Allen, F., Faulhaber, G., 1989. Signaling by underpricing in the IPO market.Journal of Financial Economics 23, 303–323.

Bhattacharya, S., Ritter, J.R., 1983. Innovation and communication:signaling with partial disclosure. Review of Economic Studies 50,331–346.

Boston Globe, 2000. Internet firms bet on Super Bowl ads to reachinvestors, consumers, January 23.

Carter, R., Dark, F., Singh, A., 1998. Underwriter reputation, initial returns,and the long-run underperformance of IPO stocks. Journal of Finance53, 285–311.

Chemmanur, T., 1993. The pricing of initial public offerings: A dynamicmodel with information production. Journal of Finance 48, 285–304.

Chemmanur, T., Fulghieri, P., 1994. Investment bank reputation, informa-tion production, and financial intermediation. Journal of Finance 49,57–79.

Chemmanur, T., Fulghieri, P., 1999. A theory of the going-public decision.Review of Financial Studies 12, 249–279.

Corwin, S.A., 2003. The determinants of underpricing for seasoned equityoffers. Journal of Finance 58, 2249–2279.

Demers, E., Lewellen, K., 2003. The marketing role of IPOs: evidence frominternet stocks. Journal of Financial Economics 68, 413–437.

Engers, M., 1987. Signaling with many signals. Econometrica 55,663–674.

Fudenberg, D., Tirole, J., 1991. Perfect bayesian equilibrium andsequential equilibrium. Journal of Economic Theory 53, 236–260.

Gertner, R., Gibbons, R., Scharfstein, D., 1988. Simultaneous signaling tothe capital and product markets. Rand Journal of Economics 19,173–190.

Grullon, G., Kanatas, G., Weston, J.P., 2004. Advertising, breadth ofownership, and liquidity. Review of Financial Studies 17, 439–461.

Kihlstrom, R.E., Riordan, M.H., 1984. Advertising as a signal. Journal ofPolitical Economy 92, 427–451.

Lease, R.C., Masulis, R.W., Page, J.R., 1991. An investigation of marketmicrostructure impacts on event study returns. Journal of Finance46, 1523–1536.

Loughran, T., Ritter, J.R., 1997. The operating performance of firmsconducting seasoned equity offerings. Journal of Finance 52,1823–1850.

Loughran, T., Ritter, J.R., 2004. Why has IPO underpricing changed overtime? Financial Management 33, 5–37.

Lowry, M., Schwert, G.W., 2002. IPO market cycles: bubbles or sequentiallearning. Journal of Finance 62, 1171–1200.

ARTICLE IN PRESS

T. Chemmanur, A. Yan / Journal of Financial Economics 92 (2009) 40–65 65

Maksimovic, V., Pichler, P., 2001. Technological innovation and initialpublic offerings. Review of Financial Studies 14, 459–494.

Michaely, R., Shaw, W., 1994. The pricing of initial public offerings: testsof adverse-selection and signaling theories. Review of FinancialStudies 7, 279–319.

Milgrom, P., Roberts, J., 1986. Price and advertising signals of productquality. Journal of Political Economy 94, 796–821.

Nelson, P., 1974. Advertising as information. Journal of Political Economy81, 729–754.

Pagano, M., Panetta, F., Zingales, L., 1998. Why do companies go public?An empirical analysis. Journal of Finance 53, 27–64.

Ritter, J.R., Welch, I., 2002. A review of IPO activity, pricing, andallocations. Journal of Finance 57, 1795–1828.

Safieddine, A., Wilhelm, W.J., 1996. An empirical investigation of short-selling activity prior to seasoned equity offerings. Journal of Finance51, 729–749.

Stoughton, N.M., Wong, K.P., Zechner, J., 2001. IPOs and product quality.Journal of Business 74, 375–408.

Wall Street Journal, 1999. In Web firms’ ad blitz, an eye on Wall Street.August 19.

Welch, I., 1989. Seasoned offerings, imitation costs, and the underpricingof initial public offerings. The Journal of Finance 44, 421–449.