Supply Matters for Asset Prices: Evidence from IPOs in Emerging Markets

Transcription

1 No Supply Matters for Asset Prces: Evdence from IPOs n Emergng Markets Matías Braun and Borja Larran Abstract: We show that the ntroducton of a new asset affects the prces of prevously exstng assets n a market. Usng data from 254 IPOs n emergng markets, we fnd that stocks n ndustres that covary hghly wth the ndustry of the IPO experence a larger declne n prces relatve to other stocks durng the month of the IPO. The effects are stronger when the IPO s ssued n a market that s less ntegrated nternatonally, and when the IPO s bg. The evdence supports the dea that the composton of asset supply affects the cross secton of stock prces. JEL Codes: G11, G14, G15 Matías Braun s an assstant professor of economcs and fnance at the UCLA Anderson School of Management. Borja Larran s an economst at the Federal Reserve Bank of Boston. Ther emal addresses are matas.braun@anderson.ucla.edu and borja.larran@bos.frb.org, respectvely. Ths paper, whch may be revsed, s avalable on the web ste of the Federal Reserve Bank of Boston at The vews expressed n ths paper are solely those of the authors and do not reflect offcal postons of the Federal Reserve Bank of Boston or the Board of Governors of the Federal Reserve System. The followng ndvduals provded nsghtful comments and suggestons: Pedro Santa Clara, Jeremy Sten, Rossen Valkanov, and Paul Wllen. Elena Myronova provded excellent research assstance. We also thank semnar partcpants at the Federal Reserve Bank of Boston for comments. Ths verson: December 2005 (frst draft: June 2005).

2 We study the effect of ntal publc offerngs (IPOs) on the prces of other stocks lsted n a market. We focus on emergng markets n order to explore quanttatvely meanngful changes n asset supply. For each of 254 IPOs n 22 emergng markets, we measure excess returns on 17 ndustry portfolos n the market of ssuance durng the month of the IPO. We then regress the returns on the covarance between each ndustry and the ndustry of the IPO, as measured by returns on the 17 ndustres n hstorcal U.S. data. We fnd a sgnfcant negatve relaton between returns and the covarance wth the IPO ndustry. A strategy that takes a long poston n the ndustry wth the lowest covarance wth the IPO ndustry and a short poston n the ndustry wth the hghest covarance yelds approxmately 80 bass ponts over the month of ssuance of the typcal IPO. The effects are strong f the local market s poorly ntegrated wth nternatonal markets, whle they dsappear f the market s well ntegrated. Furthermore, the cross-sectonal gradent n response to the IPO s steeper when the new ssue s bg relatve to the local market captalzaton. Ths paper provdes drect evdence on the central role that market clearng and the supply sde play n asset prcng, supportng recent theoretcal and emprcal nterest on the ssue. The basc ntuton s the same as n Bansal, Fang, and Yaron (2005) and Cochrane, Longstaff, and Santa-Clara (2005), namely, that nvestors requre hgher expected returns for expandng sectors. As a sector grows wth an IPO, ts beta typcally goes up, and therefore ts prce has to fall n order to promse hgher expected returns. We add that ths ntuton s true also for sectors that have a hgh covarance wth the expandng sector. In fact, the covarance wth the growng sector governs the magntude of the prce change n the rest of the assets lsted n the market. It s not standard practce n asset prcng to talk about supply and demand because the mplct assumptons of perfectly elastc supply or demand domnate the lterature. Supply s generally assumed to adjust to swngs n demand, or n other words, supply s assumed to be perfectly elastc. Supply shocks are meanngless n such a world. The creaton of a new asset s 2

3 accompaned by the destructon of another asset (repurchases), so that n equlbrum prces stay determned by demand. In practce, ths rebalancng s not automatc, gvng rse to potental supply-sde effects on asset prces. Cochrane, Longstaff, and Santa-Clara (2005) study these effects n a framework where there are shocks to broad sectors, for nstance, stocks and bonds or the stock markets of two dfferent countres. However, the typcal IPO n the U.S. s too small to be comparable to these szeable supply shocks, so we can hardly expect to observe effects akn to the ones descrbed n that paper followng an IPO. For such small shocks, demand stll looks almost perfectly elastc, and, therefore, t elmnates any nterestng effect on prces. Our focus on emergng markets makes these nsghts applcable to the case of IPOs. These markets are small and not perfectly ntegrated wth nternatonal markets, makng IPOs relatvely bgger shocks. To some extent, these markets lve n autarky and, therefore, there s a market clearng condton for each of them. The second advantage of studyng emergng markets s that local demands for assets are expected to be more nelastc than n more developed markets, because the lmts to arbtrage are more strngent (Shlefer (1986), Shlefer and Vshny (1997)). Standard factors that make the demand more nelastc are, for example, restrctons to short-sales and the lack of close substtutes (Wurgler and Zhuravskaya (2002)). Both of these attrbutes are probably more pervasve n emergng markets, characterzed as they are by scarce lqudty and wdespread nsder ownershp (La Porta et al. (2000)). The exstence of greater lmts to arbtrage amplfes the effects of supply shocks beyond the frctonless demand-sde benchmark consdered n Cochrane, Longstaff, and Santa-Clara (2005). Throughout the paper, we reman agnostc about the determnants of the elastcty of demand. It s hard to dsentangle smple segmentaton from more sophstcated lmts to arbtrage because the underlyng causes of both are probably correlated. For nstance, the same lack of lqudty keeps a market segmented and at the same tme puts lmts on the opportuntes for 3

4 arbtrage. Our man nterest s n brngng attenton to the effects of supply shocks. Irrespectve of whether asset demand follows standard rsk-return theores or behavoral theores for lack of a better termnology the changes n prces after an IPO suggest that there s a role for supply that has not been suffcently studed. Our focus on supply shocks s new to the lterature. Prevous studes have mostly documented the effect of demand changes on prces. Bekaert and Harvey (2000) and Henry (2000) show that stock prces ncrease on average when an emergng market opens up to foregn nvestors. Openng up the market provdes a demand shock that nduces a change n the value of local assets. Ths strategy s n essence the same strategy that Harrs and Gurel (1986) and Shlefer (1986) use n the study of addtons to the S&P 500 a partcular segment of the larger U.S. market. A close paper to ours s that of Hong, Kubk, and Sten (2004), who document that market-to-book ratos are negatvely related to the rato of total book equty to total personal ncome across U.S. states. We can nterpret ths rato as a measure of a state s relatve asset supply. Our approach dffers n that, nstead of focusng only on varaton n the demand/supply balance derved from nvestors geographcal preferences (Coval and Moskowtz (1999)), we consder an experment where supply changes. We also explot wthn-market dfferences n returns and are thus better able to control for omtted varables. Our focus on the effect of a new ssue on the prces of other assets s related to the fndngs of Newman and Rerson (2004), who show that a very large ssuance of Deutsche Telekom depressed the prces of other European telecommuncatons bonds. Our paper dffers from thers n three man respects: we study stocks nstead of bonds; we document how the prce effect declnes as the cross-secton of assets covares less wth the IPO; and we use cross-country varaton from emergng markets. 4

5 The next sectons present a prelmnary motvaton (Secton 1) and a descrpton of the methodology and the data (Secton 2). The results follow n Secton 3. We then conclude. 1. The Effect of a New Issue on the Prces of Other Assets: A Mean-Varance Approach Assume that the CAPM holds and that each market s n autarky. Expected returns on asset are descrbed by the followng equaton: E r ) r = β [ E( r ) r ]. (1) ( f m f Expected excess returns are equal to the beta of the asset tmes the local market rsk premum. Under standard assumptons, Merton (1980) shows that the market rsk premum can be wrtten as 2 E( r m ) r f = γσ m. (2) The parameter γ s the coeffcent of relatve rsk averson of a representatve nvestor, 2 and σ m s the varance of the market return. Usng the defnton of market beta and substtutng equaton (2) nto (1), we get E r ) r = γ Cov( r, r ). (3) ( f m Now assume that a new asset (the IPO) s ntroduced n the market. The market ntally has =1 n assets, so the IPO s asset n+1. We refer to the market wth n assets as market 0, and to the market wth n+1 assets as market 1. The weght of asset n market 0 s denoted byω, 0 (analogously for market 1). We assume that the number of shares s constant and, therefore, that any change n the market weght comes from a change n prce. Wth the ntroducton of the IPO, the covarance n the rght-hand sde of equaton (3) changes, therefore changng expected 5

6 returns. Assumng, for smplcty, that the rsk-free rate stays constant, we can express the change n expected returns as: n ΔE( r ) = γ ω Cov( r, r ) γ ( ω ω ) Cov( r, r ). (4) po po j= 1 j,0 j,1 j Equaton (4) has two opposng terms. In order to smplfy the nterpretaton, frst consder the case of an asset that has zero covarance wth the orgnal n assets, but a non-zero covarance wth the IPO. In market 0, the expected return on ths asset s the rsk-free rate the asset has no systematc rsk. The change n expected return on ths asset corresponds only to the frst term n equaton (4). If the covarance wth the IPO s postve, the asset receves a rsk premum after the IPO; f the covarance s negatve, the asset s a good hedge aganst the fluctuatons of the IPO and t receves a rsk dscount. The magntude of the effect s nfluenced by the weght of the IPO n the market, ω po, and by the prce of rsk gven by the nvestor s rsk averson. The second term n equaton (4) tends to offset the effect of the frst term. The ntuton s the followng. From the frst term we know that an asset that covares postvely wth the IPO receves a hgher expected return, a lower prce, and consequently a lower weght n the market (ceters parbus). Therefore, assets wth postve IPO covarance see ther market weght declne accordng to the frst term. But the decrease n market weght leads mechancally to a lower covarance of these assets wth the new market and a lower rsk premum, dampenng the prevous ncrease n rsk premum. Ths second effect s lkely to be of second order except for extreme cases. The extreme cases are smlar to the examples n Cochrane, Longstaff, and Santa-Clara (2005), where an ncrease n the market share of an asset lowers ts expected returns, for nstance, when the share s close to one. Bansal, Fang, and Yaron (2005) also regard these cases as not emprcally relevant. 6

7 We focus on the mpact of Cov r, r ) throughout the paper, so, f anythng, the second term n ( po equaton (4) bases our emprcal strategy aganst fndng a result. A smple example that shows the lnear dependence of changes n expected returns wth respect to the covarance s the case of the entre market, 2 Δ E( r ) = γ ω [ Cov( r, r ) σ ]. (5) m po m po m We consder varatons n the mpact of the covarance as the sze of the IPO ( ω ) changes. We also study the mpact of market segmentaton, whch can be understood as another way of varyng the sze of the IPO relatve to the market. In a less segmented market the relevant market captalzaton ncludes foregn assets, whch amounts to sayng that ω po shrnks. In the extreme case of a fully ntegrated market (that s, where the world market s the reference for the CAPM as n Karoly and Stulz (2003)), any IPO necessarly has a neglgble sze, and therefore the change n expected returns n equaton (4) s zero. In ths analyss we assume that the IPO creates a new source of wealth n the economy. In a mean-varance graph (see Fgure 1), the addton of the new asset modfes the effcent fronter and therefore the market for rsky assets. In such case t s clear that the IPO has a potental effect on other asset prces. On the other hand, Wllen (2005) fnds that the ntroducton of an asset n zero net supply (that s, an asset that s not new wealth) leaves the prces of other rsky assets unchanged. But even then, addng an asset that s not new wealth can affect other prces f we consder further frctons. For nstance, an IPO leads to changes n prces n the case of a prvately held company whose owners were formerly lqudty constraned. po 7

8 2. Event Study around IPO Dates A. Data Sources Stock prces come from the Emergng Markets Database (EMDB). We use dollar prces as of the end of the month. We do not use daly data because many stocks are traded only sporadcally n emergng markets, and we thus often observe zero daly returns. We form 17 value-weghted ndustry portfolos n each country, followng the ndustral classfcaton of Fama and French. 1 We defne the market return as the value-weghted return on the EMDB stocks n the country durng the month. The IPO data come from Thomson Fnancal s SDC Platnum. We start wth all common equty prmary IPOs. We then restrct the sample to the ssues where the frm s lstng n ts home market. The sample excludes events ntated by frms already lsted (frms ssung ether a new class of stock or n other markets). IPOs are ncluded only f the amount s larger than $20 mllon. Ths leaves out data of debatable qualty and retans ssues more lkely to have a materal mpact on prces. In order to keep the dentfcaton of the events as clean as possble, we use IPOs that are ssued n a month n whch no other IPO larger than $20M s lsted n the same country. From ths data set we keep the ssuance date, the dollar amount of the IPO, and the ssung frm s country and ndustry. After matchng the data sets, we end up wth 254 IPOs n 22 dfferent emergng markets, correspondng to the sample perod. Table AI n the appendx provdes summary statstcs. 1 The defnton and returns assocated wth these portfolos n the U.S. can be found on Ken French s webpage ( We match SDC s SIC and EMDB s GICS classfcaton to the classfcaton used by Fama and French. We also perform tests wth Fama-French s 48-ndustry classfcaton and obtan smlar results. The panel looks more unbalanced n that case because of mssng ndustres n some countres and perods. 8

9 B. Basc Regresson and Identfcaton We conduct an event study around the date of the lstng of new ssues. Tradtonally, event studes n the fnance lterature have focused on outcomes of the frm affected by or ntatng the event. We nstead concentrate on the evoluton of the stock prce of the other frms n the same natonal market. The regressons we estmate are of the followng type: R c usa usa j c = α + β cov( R, R ) + ε, (6) where the dependent varable s the return on ndustry n country c durng the month of ssuance of an IPO n country c. The vector α represents a set of IPO fxed effects. The man ndependent varable measures the covarance of returns between ndustry and ndustry j to whch the new ssue belongs. A negatve estmate for β s consstent wth our hypothess that hgh IPOcovarance stocks see ther prces declne relatve to other stocks as an IPO enters the market. Ths s a reduced-form regresson, so there s no drect mappng between the estmate for β and parameters n equaton (4) such as the rsk averson coeffcent. We compute the covarance of returns between each par of ndustres wth U.S. monthly excess returns on Fama-French s 17-ndustry portfolos from 1974 to Table AII n the appendx presents the 17x17 covarance matrx wth the 153 dfferent covarances. A countryspecfc covarance between ndustres was computed, although t s mperfect because of the dramatc changes n market structure and the lack of a long tme seres. Thus, t s not a good dea to then run a regresson wth these covarances because, on top of beng nosy, they are endogenous. The need for an exogenous measure of covarance can be understood by notng that the model presented n the prevous secton s a partal equlbrum model. Frst moments, or expected returns, are derved from second moments of returns that are taken as gven. In realty, 9

10 second moments are equally endogenous. Therefore, equaton (4) cannot be estmated drectly as a regresson wthout thnkng further about the dentfcaton problem. Our whole approach hnges on the dea that supply and demand matter for asset prces, and therefore that covarances respond to the local structure of the market. In other words, covarances have embedded n them the characterstcs of the segment where they are traded. For example, stocks added to the S&P500 exhbt changes n ther degree of comovement wth other stocks nsde and outsde the ndex (Barbers, Shlefer, and Wurgler (2005)). One example can llustrate the potental correlaton between ε c and the covarance computed wth local data. Partcularly n emergng markets, frms are usually organzed n conglomerates because of the poor development of fnancal ntermedares. A hgh covarance between frms can n part reflect the exstence of these nternal captal markets (Lamont (1997)). In such case, an IPO can sgnal an allevaton of fnancal constrants for a whole set of frms wthn a conglomerate. In ths example, the extent of nternal captal markets s the omtted varable that s hdden n the error term and s correlated wth the local covarance. Unfortunately, measurng these nter-frm lnks s vrtually mpossble, at least for a broad sample lke the one we study. Our dentfyng assumpton s that the covarances n the U.S. capture the exogenous component of the covarances n each country. The U.S. market s a well-dversfed, nternatonally-ntegrated market, wth many arbtrageurs, and consequently a market where the covarances are potentally closer to fundamental measures of rsk or behavoral degrees of substtutablty between assets that do not rely on a partcular market structure. Gven that we explot wthn-ipo, cross-ndustry varaton n the data, we just need the rankng of the nterndustry covarances to be relatvely stable across countres. For nstance, Morck, Yeung, and Yu (2000) show that stocks n less-developed markets tend to be more correlated, leadng mechancally to hgher covarances (ceters parbus). However, even f all covarances are hgher 10

11 n some markets, our varable s vald as long as the rankng of comovement across ndustres does not change dramatcally. We concentrate on the wthn-country varaton n the effect of the covarance by ncludng IPO fxed effects that absorb the market-wde prce fluctuaton or any change n the rsk-free rate. Beng able to control for unobserved characterstcs consttutes a major advantage of our emprcal desgn, because the results are robust to omtted varables that vary along any combnaton of the country, year, and IPO-ndustry dmensons. In partcular, we sheld ourselves from the potental bases due to market tmng n new ssues (Rtter (2003)) by focusng on the cross-secton of prce changes rather than the market prce change. The need to control for country heterogenety seems crtcal gven the evdence on cross-country dfferences n valuatons (La Porta et al. (2002)) and IPO underprcng (Ljungqvst (2004)), and the fact that these dfferences are not fully explaned. We also take nto account the fact that returns n the same country are potentally correlated across stocks and through tme by allowng the resduals to be clustered wthn a country. 3. Emprcal Results A. Asset Prces Fall as the Covarance wth the IPO Increases Table I presents the results from the regresson n (6) usng returns on the month of the IPO. We measure abnormal returns n two ways n ths table. Frst, we smply subtract the market return, whch we call the market-adjusted return. Gven the IPO fxed effects, t s equvalent to run regressons wth market-adjusted returns or raw returns. Second, we compute the return n excess of a market-model return estmated wth data from month t-30 to month t-7, where t s the month of the IPO. We lose approxmately 10 percent of the observatons wth the second method because t requres a longer tme seres for each ndustry. 11

12 The coeffcent of the covarance wth the IPO ndustry s negatve and sgnfcant at the 5 percent level wth both defntons of abnormal returns. The coeffcent n the regresson wth market-adjusted returns mples that a one-standard-devaton ncrease n the covarance reduces prces by 40 bass ponts. In order to put ths number nto perspectve, consder that HML (the book-to-market factor of Fama and French (1993)) gves an average premum of 40 bass ponts per month. An alternatve way of quantfyng these magntudes s to use as ndependent varable the rankng of each ndustry n terms of ts covarance wth the IPO ndustry. Usng the rankng s a way of controllng for possble non-lneartes n the effect of the covarance. The results n Table I ndcate that movng one place closer to the IPO n the rankng lowers prces by 5.7 bass ponts (6.3 bass ponts when usng market-model abnormal returns). We fnd small changes n prces, whch mply even smaller changes n expected returns. Ths can easly be seen from the Gordon growth model for the prce-dvdend rato: P/D = 1/(r-g). Assume that the P/D rato s 20. For gven dvdends, a change n prces of 40 bass ponts mples a change of only 2 bass ponts per month n expected returns. A back-of-the-envelope calbraton of our model gves smlar magntudes. Take the frst term n equaton (4), whch s our man focus, and consder the effect of a one-standard-devaton ncrease n the IPO covarance. Assume that the rsk averson coeffcent s 100, consstent wth the equty premum evdence, and that the IPO has the average sze n the sample (0.25 percent of the country s market captalzaton; see Table AI). Multplyng these terms gves the result that the change n expected returns s 1.5 bass ponts per month. We do not perform the tests wth expected returns nstead of prces because these tests would most lkely lack power. The varance of returns s just too large relatve to the sze of the effect that we document. Sometmes the changes n prce are reported n terms of demand elastctes, partcularly n the lterature on ndex addtons. If we assume that the IPO has the average sze n the sample, 12

13 a 0.40 percent change n prces mples an elastcty of Ths number s wthn the range of prevous estmates n the lterature (see Wurgler and Zhuravskaya (2002) for a survey). Fgure 2 summarzes the basc result graphcally. For each IPO we compute the marketadjusted return on each of the 17 ndustres n the country durng the month of ssuance. We then rank the ndustres from 1 to 17 accordng to the U.S. covarance wth the IPO ndustry (wth 1 beng the ndustry wth the lowest covarance). Fnally, we average the returns across all IPOs for each rankng poston. These returns are then plotted aganst the rankng, along wth a regresson lne. Ths fgure shows a strong negatve relatonshp between returns and the covarance of dfferent ndustres wth the IPO ndustry. It s clear that the effect does not come from a few outlers, but s a robust feature of the data. In partcular, the effect s not derved from the dfference between the same ndustry of the IPO versus the mpact on other ndustres. The samendustry data ponts (almost always correspondng to rankng poston 17) can be dscarded and a smlar relatonshp holds. The lower panel of Fgure 2 shows the same average returns by rankng n the months before and after the IPO. The negatve slope s no longer there. Whle durng the month of the IPO the coeffcent on the rankng s sgnfcantly negatve (at the 2 percent level) and explans 30 percent of the varaton n excess returns, t s nsgnfcant and explans only 10 percent of the varaton durng the months before and after the ssuance. Fgure 2 suggests that the U.S. covarance s not just pckng up some permanent dfference n expected returns between ndustres. In Fgure 3, we show our basc result n yet another way. For each IPO we compute separately the market-adjusted return on ndustres above and below the medan of the IPO covarance. We then plot the entre dstrbuton of returns for both groups of ndustres. The dfference n means of the two dstrbutons s qute apparent n the month of the IPO. A Kolmogorov-Smrnov test easly rejects the null hypothess of equalty of dstrbuton functons, 13

14 wth a p-value lower than 1 percent. Once agan, the effect s not present n the months before and after the IPO (Fgure 3B). For each of these months the test fals to reject the null (at p-values of 46 percent and 75 percent for the prevous and subsequent month, respectvely). Table II confrms that the effect of the IPO covarance s exclusve to the month of the ssue by showng the results of the same basc regresson for the months before and after the IPO. In both cases the covarance s not sgnfcant and the coeffcents are much smaller (n magntude) than durng the month of the IPO. As an example of the basc effect, consder the prce mpact of the typcal IPO n Transportaton. Fgure 4 plots market-adjusted returns aganst the covarance rankng as n Fgure 2, but only for the IPOs n the transportaton ndustry. As ntuton suggests, the returns of transportaton covary sgnfcantly more wth the steel ndustry than wth the food ndustry (the covarances are 0.29 percent and 0.19 percent, respectvely). IPOs n transportaton turn out to be assocated wth a negatve change of more than 50 bass ponts n the prce of the steel ndustry relatve to the food ndustry. Put dfferently, when an IPO n the transportaton ndustry occurs, a portfolo that shorts the local steel ndustry and buys the food ndustry generates a return of more than 50 bass ponts over the month of the IPO. These numbers closely match the ones n the benchmark regresson. An advantage of our methodology s that the event s not ntated by the frms for whch we measure the change n the stock prce. In fact, the ndustry returns n the dependent varable do not nclude the return on the ssung frm. In prncple, the decson to ssue equty can convey nformaton about the future prospects of the frm (Myers and Majluf (1984)) and can also drectly affect future cash flows f credt constrants are mportant. It can be argued that these effects are relevant not only for the ssung frm but also for other frms n the same ndustry or close compettors (Chevaler (1995), Phllps (1995)). The drect effect on cash flows or the nformaton sgnaled about cash flows can affect the demand for assets and blur the effects of 14

15 changes n expected returns that we pont to n the model. We checked that the results are not drven by the competton wthn the same ndustry of the IPO by runnng regressons that exclude that ndustry, and we obtaned the same results as before. In terms of the nformatonal story, we thnk t s hard to argue that a sngle IPO conveys nformaton not prevously known to markets about the cash flows of all other ndustres n a country, and that the nformaton s correlated wth the covarance of returns measured n the U.S. One caveat to our approach s that we measure prce changes around the date of ssuance of the new stock and not around the announcement date. If arbtrage s frctonless, the prce effect should be observed when the ssue s announced. At the date of announcement, arbtrageurs should sell-short stocks of ndustres wth hgh IPO covarance and should go long n ndustres wth low IPO covarance. Unfortunately, we do not have a practcal way of dentfyng the announcement date because of the very nature of the process of publc offerngs. There s no certanty about the ssuance when management announces plans to do t or fles for t; rather, the probablty of ssuance grows slowly n tme and reaches ts peak only on the actual date of lstng. In other words, there s substantal rsk n the strategy suggested above, and ths deters arbtrageurs from pursung t (De Long et al. (1990)). Measurng returns around the month, and not the day, of the IPO lkely mtgates ths concern. In any case, f the effects are concentrated around the announcement date and not the ssuance date, then t s more dffcult for us to fnd emprcally the results we document. 2 B. Other Factors n the Cross-Secton of Stock Returns There s the possblty that the IPO covarance proxes for some of the factors that are usually consdered n cross-sectonal regressons of stock returns, such as the market-to-book rato or sze. In Table III we study the effect of ncludng alternatve factors. We frst consder the 2 Newman and Rerson (2004) document prce effects both at announcement and ssuance n ther study on bonds. 15

16 factors used by Fama and French (1992), whch are the log of market equty (ME), the log of the market-to-book rato (P/B), the prce-earnngs rato when earnngs are postve (P/E(+)), and a dummy for those observatons wth negatve earnngs (E<0). These varables are measured 12 months pror to the IPO for each country-ndustry par. Out of these four factors, the prceearnngs rato s the only one that enters sgnfcantly and wth the expected negatve sgn. The market-to-book rato has the rght sgn, but t s not sgnfcant. Sze s not sgnfcant ether, and t has the wrong sgn when compared wth what s found n the U.S. The IPO covarance survves all of these controls n terms of magntude and sgnfcance; hence, a hgh covarance wth an IPO s not smply an ndcaton of small sze or hgh market-to-book value (whch s probably ndcatve of hgh growth opportuntes). Two other nterestng factors are lqudty and momentum. Turnover s a proxy for lqudty rsk, whch may be a partcularly dscouragng factor for foregn nvestors consderng nvestng n emergng markets (Bekaert and Harvey (2003), Lesmond (2005)). We defne turnover as the average over the 12 months pror to the IPO of value traded dvded by market captalzaton of each ndustry n each country. However, the coeffcent on turnover s not sgnfcant and t has the wrong sgn (that s, negatve). Momentum, nstead, s a robust predctor of returns. We measure momentum as a dummy varable that takes the value of one when the cumulatve market-adjusted return over months t-6 through t-1 s postve; or n other words, when the ndustry under consderaton s a wnner n the 6 months pror to the IPO. 3 As seen n Table III, the momentum effect s very strong. Wnners n the past 6 months earn, on average, an extra 1 percent durng the IPO month. Even after ncludng momentum, however, the coeffcent on the IPO covarance remans sgnfcant at the 5 percent level, and ts magntude s only slghtly reduced. 3 We also tred the orgnal defnton of momentum n Jegadeesh and Ttman (1993), whch goes from month t-12 to month t-2. It was less robust than the defnton we use here, and t does not affect the coeffcent on the IPO covarance. 16

17 IPOs tend to cluster around hot markets, that s, after a successon of postve returns. It has been suggested that managers explot temporary wndows of opportunty provded by market msprcng (Rtter (2003)). Under ths hypothess, we should observe IPOs clustered n ndustres wth postve momentum. However, f ths s the case, ndustres wth a hgh covarance wth the IPO wll share the momentum and the hgh returns of the IPO ndustry. We show, nstead, that hgh covarance ndustres have unusually low returns durng the month of the IPO. In other words, a contagous IPO-ndustry momentum works aganst the negatve effect of the hgh IPOndustry covarance. The results n Table III suggest that sharng the postve momentum of the IPO ndustry s not enough to overturn the negatve effect of the market-clearng consderatons. C. IPO Sze and Market Segmentaton In a deep market lke the U.S., probably no IPO s bg enough to have a sgnfcant effect on all other stocks. Emergng markets, nstead, are much smaller n terms of total market captalzaton and number of nvestors. The sze of the average (medan) IPO n our sample s $98 ($43) mllon, whle the average (medan) market captalzaton s just $91 ($80) bllon. Perhaps more mportant s the fact that, gven the prevalence of government and nsder control (La Porta et al. (2000)), market captalzaton substantally overstates the value of stocks that are actually traded n these markets. Just to gve a sense of the magntude of the correcton needed to account for ths problem, the free-float market captalzaton s only 14 percent of total captalzaton n Chle. If we assume that ths number s the same for all countres, then the average IPO represents just below 1 percent of the respectve market free float. As seen n equaton (4), a bgger IPO amplfes the effect of the IPO covarance. The IPO fxed effects absorb any drect mpact of sze, but sze can stll nteract wth the covarance. In Table IV we splt the sample n three, accordng to the dollar amount of the IPO relatve to the total market captalzaton. The coeffcent on the IPO covarance ncreases (n magntude) as we 17

18 move from small to bg IPOs. In fact, the covarance effect s sgnfcant n the thrd of the sample that corresponds to the relatvely bg IPOs, but not n the other two sub-samples. A second source of varaton n sze comes from the segmentaton of the market. Segmentaton determnes the extent of the demand for assets. For nstance, nvestors from all over the world are potental partcpants n a perfectly-ntegrated market. We present two alternatve measures of segmentaton n Table V. These measures vary across countres and through tme, as opposed to other nsttutonal features that vary almost exclusvely across countres. The decade under consderaton s a perod of substantal changes n the segmentaton of emergng markets, so we prefer these tme-varyng measures (Bekaert and Harvey (1995)). Our frst measure corresponds to the rato of the nvestable IFC ndex to the global IFC ndex (Bekaert (1995)). Ths rato, whch s avalable at the monthly frequency, shows the fracton of market captalzaton n whch foregners can potentally nvest. In the top panel of Table V we splt the sample n three accordng to ths rato. The coeffcent on the IPO covarance ncreases (n magntude) as we move to more segmented markets. As seen n the frst column, t s not sgnfcant n well-ntegrated emergng markets. The mddle panel of Table V presents results when the sample s splt accordng to market turnover. Low lqudty can be a deterrent to foregn nvestors and an mportant cause of segmentaton. As expected, the effect of the IPO covarance s strong n less lqud markets, but mssng n the most lqud ones. So far, we have focused on varaton wthn emergng markets. In the bottom panel of Table V we compare emergng markets as a group wth those markets that are more developed and well-ntegrated accordng to the IFC classfcaton. To conduct ths exercse, we gather stockprce data from Datastream to buld the ndustry portfolos of 37 countres snce We then match the returns to the SDC IPO data as before, and run the benchmark regresson separately for emergng and developed countres. In the sample of emergng markets the results are comparable 18

19 n magntude and sgnfcance wth the results of our benchmark sample. As expected, there s no effect of the IPO covarance n developed markets. We consder ths as just a robustness exercse, because the number of stocks n Datastream s much smaller than n EMDB, and because we can form only equal-weghted portfolos snce data on shares outstandng are not avalable. D. Other Measures of Substtutablty between Assets The model presented n the ntroducton s a standard model of a rsk-return tradeoff. The IPO changes the covarance of each asset wth the market, whch s the measure of rsk, and therefore t commands a change n expected returns. Instead of focusng on tradtonal measures of rsk, Barbers and Shlefer (2003) suggest that nvestors use easly observable characterstcs such as sze, the book-to-market rato, or the ndustry, to classfy assets. We can then speculate that, when a new asset appears, nvestors adjust ther portfolos to reflect ther desred exposure to the dfferent styles wthn the market. Those assets that have a style smlar to that of the IPO are substtuted away more strongly than other assets. For example, an IPO can crowd out and lower the prce of other stocks wth smlar book-to-market values. Ths effect can potentally wpe out or complement the effect of the IPO covarance prevously dentfed. To explore ths ssue, we classfy assets accordng to the book-to-market rato and sze of each ndustry relatve to the IPO ndustry n the month pror to the realzaton of the return. We say that an ndustry s close to the IPO f the absolute dfference n the book-to-market rato between the two s small, and proceed analogously for sze. Teo and Woo (2004) also use categores based on sze and book-to-market values n ther tests of style nvestng. In Table VI we show that the prces of ndustry portfolos that are close to the IPO ndustry n terms of book-to-market value and sze fall relatve to other ndustres. Ths effect s agan lmted to the month of the IPO. The effect of sze s more robust, and n fact makes the book-to-market varable nsgnfcant when both are ncluded n the regresson. The IPO 19

20 covarance s stll sgnfcant and ts coeffcent s of smlar magntude to the one n our benchmark regresson. In Table VII we see that the effects are concentrated n markets wth medum and hgh levels of segmentaton. In prncple, the mpact of ths second class of substtutablty measures s not necessarly expected to be stronger n more segmented markets. Style nvestng can affect nternatonal nvestors as well as local nvestors. However, our evdence suggests that style nvestng may be even worse n markets domnated by local nvestors. One problem wth testng style nvestng s that the defnton of style s always debatable. For nstance, followng the methodology for the IPO covarance, we also try the measure of book-to-market closeness wth hstorcal book-to-market ratos for U.S. ndustres. However, ths measure s never sgnfcant. We can argue that t s smply not a good measure of asset style, or that t s not relevant for the partcpants n the market. In any case, ts ncluson does not affect the coeffcent of the IPO covarance (results not reported). E. Volume Traded As a fnal step, we examne the volume traded durng the month of the IPO and the months around t. Table VIII shows that the IPO covarance sgnfcantly predcts hgher volume n the month of ssuance, both when measured as dollar volume and when measured as number of shares traded, and does so even after controllng for the hgh autocorrelaton by ncludng lagged volume (Lo and Wang (2000)). The relatonshp s, n general, nsgnfcant for the prevous and followng months (wth one excepton, where the covarance comes n sgnfcantly at only the 10 percent level). The evdence on volume, taken together wth the evdence on prce changes, suggests that the ndustres that covary hghly wth the IPO experence more sellng pressure than other ndustres as nvestors rebalance ther portfolos. However, other possbltes cannot be ruled out completely wthout more detaled data on order flows. 20

21 4. Conclusons Ths paper shows emprcally that changes n asset supply have a sgnfcant mpact on the prces of assets n a market. Therefore, the constrants mposed by market clearng should not be gnored, as also suggested by Cochrane, Longstaff, and Santa-Clara (2005). We measure the change n supply through IPOs and focus on mperfectly ntegrated emergng markets. The supply shock has a cross-sectonal prce mpact that s nversely related to the covarance of returns between each ndustry and the IPO s ndustry. If one consders Fama and French s 17-ndustry classfcaton, sellng the closest ndustry and buyng the most dstant ndustry gves a spread of approxmately 80 bass ponts n the month of the IPO. 21

22 References Bansal, Rav, Ed Fang, and Amr Yaron, 2005, Equty captal: A puzzle? Workng paper, Fuqua School of Busness. Barbers, Ncholas, and Andre Shlefer, 2003, Style nvestng, Journal of Fnancal Economcs 68, Barbers, Ncholas, Andre Shlefer, and Jeffrey Wurgler, 2005, Comovement, Journal of Fnancal Economcs 75, Bekaert, Geert, 1995, Market ntegraton and nvestment barrers n emergng equty markets, World Bank Economc Revew 9, Bekaert, Geert, and Campbell Harvey, 1995, Tme-varyng world market ntegraton, Journal of Fnance 50, Bekaert, Geert, and Campbell Harvey, 2000, Foregn speculators and emergng equty markets, Journal of Fnance 55, Bekaert, Geert, and Campbell Harvey, 2003, Emergng markets fnance, Journal of Emprcal Fnance 10, Chevaler, Judth, 1995, Captal structure and product-market competton: Emprcal evdence from the supermarket ndustry, Amercan Economc Revew 85, Cochrane, John, Francs Longstaff, and Pedro Santa-Clara, 2005, Two trees, Workng paper, Chcago GSB. Coval, Joshua, and Tobas Moskowtz, 1999, Home bas at home: Local equty preference n domestc portfolos, Journal of Fnance 54, De Long, J. Bradford, Andre Shlefer, Lawrence H. Summers, and Robert J. Waldmann, 1990, Nose trader rsk n fnancal markets, Journal of Poltcal Economy 98, Fama, Eugene, and Kenneth French, 1992, The cross-secton of expected stock returns, Journal of Fnance 47,

23 Fama, Eugene, and Kenneth French, 1993, Common rsk factors n the returns on stocks and bonds, Journal of Fnancal Economcs 33, Harrs, Lawrence, and Etan Gurel, 1986, Prce and volume effects assocated wth the new S&P 500 lst: New evdence for the exstence of prce pressures, Journal of Fnance 41, Henry, Peter Blar, 2000, Stock market lberalzaton, economc reform, and emergng market equty prces, Journal of Fnance 55, Hong, Harrson, Jeffrey Kubk, and Jeremy Sten, 2004, The only game n town: Stock-prce consequences of local bas, Workng paper Prnceton Unversty. Jegadeesh, Narasmhan, and Sherdan Ttman, 1993, Returns to buyng wnners and sellng losers: Implcatons for stock market effcency, Journal of Fnance 48, Karoly, Andrew, and Rene Stulz, 2003, Are fnancal assets prced locally or globally? Handbook of the Economcs of Fnance, George M. Constantndes, Mlton Harrs, and Rene Stulz, eds., Elsever-North Holland. Lamont, Owen, 1997, Cash-flow and nvestment: Evdence from nternal captal markets, Journal of Fnance 52, La Porta, Rafael, Florenco Lopez-de-Slanes, Andre Shlefer, and Robert Vshny, 2000, Investor protecton and corporate governance, Journal of Fnancal Economcs 58, La Porta, Rafael, Florenco Lopez-de-Slanes, Andre Shlefer, and Robert Vshny, 2002, Investor protecton and corporate valuaton, Journal of Fnance 57, Lesmond, Davd, 2005, Lqudty of emergng markets, Journal of Fnancal Economcs 77, Ljungqvst, Alexander, 2004, IPO underprcng: A survey, Handbook n Corporate Fnance: Emprcal Corporate Fnance, B. Espen Eckbo, ed., Elsever-North Holland. 23

24 Lo, Andrew, and Jang Wang, 2000, Tradng volume: Defntons, data analyss, and mplcatons for portfolo theory, Revew of Fnancal Studes 13, Merton, Robert C., 1980, On estmatng the expected return on the market: An exploratory nvestgaton, Journal of Fnancal Economcs 8, Morck, Randall, Bernard Yeung, and Wayne Yu, 2000, The nformaton content of stock markets: Why do emergng markets have synchronous stock prce movements? Journal of Fnancal Economcs 58, Myers, Stewart, and Ncholas Majluf, 1984, Corporate fnancng and nvestment decsons when frms have nformaton that nvestors do not have, Journal of Fnancal Economcs 13, Newman, Ygal, and Mchael Rerson, 2004, Illqudty spllovers: Theory and evdence from European telecom bond ssuance, Workng paper Stanford GSB. Phllps, Gordon, 1995, Increased debt and ndustry product markets: An emprcal analyss, Journal of Fnancal Economcs 37, Rtter, Jay, 2003, Investment bankng and securtes ssuance, Handbook of the Economcs of Fnance, George M. Constantndes, Mlton Harrs, and Rene Stulz, eds., Elsever-North Holland. Shlefer, Andre, 1986, Do demand curves for stocks slope down? Journal of Fnance 41, Shlefer, Andre, and Robert Vshny, 1997, The lmts of arbtrage, Journal of Fnance 52, Teo, Melvn, and Sung-Jun Woo, 2004, Style effects n the cross-secton of stock returns, Journal of Fnancal Economcs 74, Wllen, Paul, 2003, New fnancal markets: Who gans and who loses, Economc Theory. Wurgler, Jeffrey, and Ekaterna Zhuravskaya, 2002, Does arbtrage flatten demand curves for stocks? Journal of Busness 75,

25 Table I The Effect of an IPO on the Stock Returns of Other Industres durng the Month of the IPO Ths table shows the results from the followng regresson: c usa usa c R = α + β cov( R, R ) + ε. In the left panel, the dependent varable s the return of ndustry n country c n excess of the local market return durng a month (market-adjusted returns). The local market s defned as the value-weghted sum of all stocks n that country and month reported n the EMDB database. In the rght panel, the excess return s computed wth a market model estmated between months t-7 and t-30. Results are shown for month t, whch s the month of the IPO. The ndependent varable s the covarance between ndustry and ndustry j, whch s the ndustry of the IPO. Ths covarance s estmated wth monthly ndustral returns from U.S. stocks between 1973 and The ndustry defntons correspond to the 17 groups of SIC codes defned on Ken French s webste. The results are also shown usng the rank of the covarance of each ndustry wth a gven IPO ndustry. The covarance rank ranges from 1 to 17. The coeffcent on the covarance rank s multpled by 1000, so t s nterpreted as bass ponts lost (or ganed) when movng one place n the rankng. The IPO fxed effects (α n the equaton above) are not reported. Detals on the selecton of IPOs are provded n the text. Returns n the dependent varable are truncated at the 1% and 99% levels. Robust standard errors clustered by country are reported below the coeffcents. Sgnfcance (two-sded): ***1%, **5%, *10%. j Dependent Varable n IPO Month Market-Adjusted Return Market Model Abnormal Return Covarance wth IPO ndustry ** ** Covarance Rank * ** N Observatons N IPOs R

26 Table II The Effect of an IPO on the Stock Returns of Other Industres: The Prevous and the Followng Months Ths table shows the results from the followng regresson: c usa usa c R = α + β cov( R, R ) + ε. The dependent varable s the return of ndustry n country c n excess of the local market return durng a month (market-adjusted returns). The local market s defned as the value-weghted sum of all stocks n that country and month reported n the EMDB database. Results are shown for month t, whch s the month of the IPO, and for months t-1and t+1. The ndependent varable s the covarance between ndustry and ndustry j, whch s the ndustry of the IPO. Ths covarance s estmated wth monthly ndustral returns from U.S. stocks between 1973 and The ndustry defntons correspond to the 17 groups of SIC codes defned on Ken French s webste. The results are also shown usng the rank of the covarance of each ndustry wth a gven IPO ndustry. The covarance rank ranges from 1 to 17. The coeffcent on the covarance rank s multpled by 1000, so t s nterpreted as bass ponts lost (or ganed) when movng one place n the rankng. The IPO fxed effects (α n the equaton above) are not reported. Detals on the selecton of IPOs are provded n the text. Returns n the dependent varable are truncated at the 1% and 99% levels. Robust standard errors clustered by country are reported below the coeffcents. Sgnfcance (two-sded): ***1%, **5%, *10%. j Dependent Varable: Market-Adjusted Return Month Relatve to IPO Prevous Month Month of IPO Followng Month Covarance wth IPO ndustry ** Covarance Rank * N Observatons N IPOs R

27 Table III The Effect of an IPO on the Stock Returns of Other Industres: Alternatve Cross-Sectonal Factors Ths table shows the results from the followng regresson: c usa usa c c R = α + β cov( R, R ) + X + ε. The dependent varable s the return of ndustry n country c n excess of the local market return durng a month (market-adjusted returns). The local market s defned as the value-weghted sum of all stocks n that country and month reported n the EMDB database. Results are shown for month t, whch s the month of the IPO. The set of ndependent varables ncludes the covarance between ndustry and ndustry j, whch s the ndustry of the IPO. Ths covarance s estmated wth monthly ndustral returns from U.S. stocks between 1973 and The ndustry defntons correspond to the 17 groups of SIC codes defned on Ken French s webste. The other ndependent varables, represented by X c n the equaton above, nclude the log of the market equty (ME), the log of the prce-to-book rato (P/B), the prce-earnngs rato f earnngs are postve (P/E(+)), a dummy for negatve earnngs (E<0), the value of traded shares as a fracton of market captalzaton averaged over the past 12 months (turnover), and a dummy for those ndustres that have postve accumulated market-adjusted returns n the 6 months pror to the IPO (momentum). The frst 4 control varables mentoned are measured 12 months pror to the IPO. The IPO fxed effects (α n the equaton above) are not reported. Detals on the selecton of IPOs are provded n the text. Returns n the dependent varable are truncated at the 1% and 99% levels. Robust standard errors clustered by country are reported below the coeffcents. Sgnfcance (two-sded): ***1%, **5%, *10%. j Dependent Varable: Market-Adjusted Return n IPO Month Covarance wth IPO ndustry ** * ** * ** ** Log(ME) Log(P/B) P/E(+) *** *** E<0 Dummy Turnover Momentum ** * N Observatons N IPOs R

28 Table IV The Effect of an IPO on the Stock Returns of Other Industres: Sub-Samples Accordng to the Sze of the IPO Ths table shows the results from the followng regresson: c usa usa c R = α + β cov( R, R ) + ε. The dependent varable s the return of ndustry n country c n excess of the local market return durng a month (market-adjusted returns). The local market s defned as the value-weghted sum of all stocks n that country and month reported n the EMDB database. Results are shown for month t, whch s the month of the IPO. The ndependent varable s the covarance between ndustry and ndustry j, whch s the ndustry of the IPO. Ths covarance s estmated wth monthly ndustral returns from U.S. stocks between 1973 and The ndustry defntons correspond to the 17 groups of SIC codes defned on Ken French s webste. The sze of an IPO s the proceeds from the IPO dvded by the total market captalzaton of the country n the month of the IPO (excludng the IPO tself). The sample s splt nto three groups (small-medum-bg) accordng to the 33 rd and 66 th percentle of the IPO sze. The IPO fxed effects (α n the equaton above) are not reported. Detals on the selecton of IPOs are provded n the text. Returns n the dependent varable are truncated at the 1% and 99% levels. Robust standard errors clustered by country are reported below the coeffcents. Sgnfcance (two-sded): ***1%, **5%, *10%. j Sze of the IPO Relatve to the Local Market Small Medum Bg Covarance wth IPO ndustry *** N Observatons N IPOs R