NOISE TRADERS IN EVENT STUDIES? THE CASE OF EQUITY CARVE-OUTS

Transcription

1 NOISE TRADERS IN EVENT STUDIES? THE CASE OF EQUITY CARVE-OUTS John R. M. Hand Kenan-Flagler Business School McColl Building, UNC Chapel Hill Chapel Hill, NC 27599, U.S.A. Abstract In this paper I hypothesize that the well documented positive mean excess stock return earned by parent firms when they announce they are carving out stock in a subsidiary is due to noise traders who optimistically misinterpret a carve-out s true value-irrelevance, rather than to the impounding of new value-enhancing information by sophisticated investors. I offer three pieces of evidence that support the noise trade view. First, parents experience a reliably negative mean excess stock return of 2.4% when the carved-out subsidiary goes public. This is in contrast with a 2.3% rise at the announcement date a median of only 45 trading days earlier. Second, proxies for the intensity of noise trader demand at the carve-out announcement explain 11% of the cross-sectional variation in parent excess returns at the subsidiary stock issue date. Third, the coefficients on the noise trader demand proxies in a cross-sectional regression whose dependent variable is the parent excess return at the issue date are individually and jointly equal to minus one times the coefficients obtained in an identical cross-sectional regression but whose dependent variable is the parent excess return at the announcement date. It appears that parent stock prices are optimistically mispriced by noise traders at the carve-out announcement, but that such mispricing is corrected when the subsidiary stock goes public. Keywords: Equity carve-outs, noise traders, market efficiency, event studies. JEL classification: G14; G32 First draft: May 5, 1997 This version: January 2, 2006 This work is supported by the Center for Finance & Accounting Research at UNC-Chapel Hill. I am grateful to Brad Barber, Mary Barth, Jennifer Conrad, Mark Lang, Charles Lee, Katherine Schipper, Henri Servaes, Terry Skantz, Steve Slezak, Jake Thomas, and participants at the 1996 Stanford Summer Camp and the 1996 Conference on Financial Economics & Accounting for related comments on an earlier related paper.

2 1. INTRODUCTION In his famous review, Fama (1991) argues that event studies offer the cleanest and most direct evidence on the extent to which capital markets are efficient because event studies come closest to allowing a break between the joint nature of market efficiency and equilibrium pricing. He concludes that the scientific evidence overwhelmingly demonstrates that with respect to firmspecific events, the adjustment of stock prices to new information is both fast and rational. One attractive aspect of the efficient market hypothesis is that it permits strong inferences to be made as to whether particular voluntary operating, investing or financing activities create or destroy firm value. The activity I examine in this paper is the financing event of an equity carve-out. In an equity carve-out, a firm s subsidiary makes an initial public offering (IPO) of the subsidiary s common stock. Unlike public offerings of seasoned stock, carve-outs have been seen as unusual financing events because several event studies have shown that on average they reliably increase, not decrease, parent stock prices at the initial announcement date. 1 Under the assumption of market efficiency, prior research has inferred that carve-outs therefore create value for their parents. In this study, I argue that this widely held inference is sensitive to the length of the event window, and whether noise trader demand affects stock prices or not. In a departure from the conventional viewpoint, I hypothesize that the positive mean excess parent stock return at the carve-out announcement is due to optimistic noise traders who misinterpret a carve-out s true value-irrelevance, rather than to the impounding into price of new value enhancing information by sophisticated investors. I further conjecture that the positive mispricing induced by noise traders is only temporary, and is corrected by sophisticated investors when the subsidiary stock is issued. In essence, the noise trader hypothesis proposes that the market is inefficient on the first event (the carve-out announcement), but efficient at the second (the subsidiary IPO date). 1 For 76 U.S. carve-outs undertaken between 1965 and 1983, Schipper and Smith (1986, 1989) report a reliably positive mean five-day announcement parent excess stock return of 1.8%. Using other samples and/or data up to 1997, similar announcement date findings and inferences concerning the valueenhancing properties of U.S. equity carve-outs for parents have been reported by Klein, Rosenfeld and Berenak (1991), Michaely and Shaw (1995), Slovin, Sushka and Ferraro (1995), Allen and McConnell (1998), Vijh (1997) and Vijh (2002). U.S. carve-outs after 1997 have been analyzed primarily in the context of small numbers of negative stubs (e.g., Lamont and Thaler, 2003). Positive announcement returns for German companies have been documented by Elsas and Löffler (2003) and Wagner (2005). The good news in equity carve-outs has been modeled within a Myers and Majluf (1984) framework by Nanda (1991). 2

3 I provide four separate strands of evidence in support of the noise-trader hypothesis. First, I document that parents experience a reliably negative mean excess stock return of 2.4% when the carved-out subsidiary goes public, in contrast with a 2.3% rise at the announcement date a median of only 45 trading days earlier. These contrasting results suggest that the market is inefficient at either the carve-out announcement, or the subsidiary IPO date, or both. Second, the mean parent excess return over the period spanned by the announcement and issue dates is 0.9% (t-statistic = 0.8). The lack of statistical significance of this announcement/issuance spanning stock return is consistent with the assumption under the noise trader hypothesis that carve-outs are on average value-irrelevant events. Consistent with this, I also find that there are no significantly non-zero stock price externalities felt by rivals to parents or carved-out subsidiaries at the announcement or subsidiary IPO dates. My third piece of evidence is that proxies for three components of the intensity of noise trader demand at the carve-out announcement together explain 18% of the cross-sectional variation in parent announcement date excess returns. While this could merely indicate that the proxies are capturing information used by sophisticated investors to rationally reprice parent stock prices, they also explain a remarkable 11% of the cross-sectional variation in subsequent parent issue date excess returns. This is inconsistent with market efficiency because the proxies are measured using information available at the carve-announcement. They are therefore stale by the issue date. Under market efficiency, the expected private return on stale public data is zero (Fama, 1970), so noise trader demand proxies measured at the announcement should not be able to explain any cross-sectional variation in parent excess returns at the subsequent date of the subsidiary IPO. Finally, I am unable to reject the hypothesis that the coefficients on the three noise trader demand proxies in a cross-sectional regression whose dependent variable is the parent excess return at the issue date are individually and jointly equal to minus one times those observed an identical cross-sectional regression but whose dependent variable is the parent excess return at the announcement date. That is, each source of mispricing in parent stock prices at the carve-out announcement appears to be exactly undone when the subsidiary stock is issued. Since I can identify no reason for why noise traders would exert systematic selling pressure on parent stock prices at the issue date, this suggests that the mispricing is being undone by sophisticated investors who, by definition, realize how and why noise trader demand at the announcement date 3

4 originally drove prices up. The regression results are all the more startling in the light of the fact that the simple product moment and rank correlations between parent excess returns at the announcement and issue dates are not reliably different from zero. The lack of any simple correlation juxtaposed with rich evidence of multivariate partial correlations indicates that mispricing induced by noise traders can be subtle and hidden from casual observation. I conclude that in the case of equity carve-outs, the evidence indicates that the positive mean parent excess return at the carve-out announcement is due to misplaced optimistic demand by noise traders, rather than the impounding into price of new, value-enhancing information by sophisticated investors. My findings can also be viewed as confirming Fama s (1991) view that the efficient market hypothesis is surely false in the sense that it will not always provide the best simplifying view of the world. Thus, in the case of event studies, there is always some positive probability that inferences strictly conditioned on market efficiency will be incorrect. Whether carve-outs are unique in this regard, or whether taking noise trader demand into account more generally can lead to inferences in event studies that differ from those under the efficient market hypothesis, seems a potentially fruitful area for further research. For example, in a narrow but very powerful way Lamont and Thaler (2003) show that the valuations of several notable carve-outs between 1998 and 2000 led to negative-stub situations where the value of the parent separate from its holdings in the carved-out stub were substantially negative. Lamont and Thaler conclude that while gross mispricings of this kind severely violate the law of one price (that is, the same asset cannot trade simultaneously at different prices), they do not represent exploitable arbitrage opportunities because of the very high costs of shorting the subsidiary. In the more general situation examined in my study, the mispricing involved is much smaller, but the underlying optimism on the part of noise traders may be similarly leading to a departure of prices from those predicted under the efficient market hypothesis. The remainder of the paper is as follows. Section II details the sampling procedure and describes selected characteristics of parents and subsidiaries. Section III replicates the announcement date event study result of prior research, and establishes the existence of a new parent-pricing anomaly at the subsidiary issue date. Section IV then develops the noise trader hypothesis and its predictions concerning the separate and joint behavior of parent excess returns at the announcement and subsidiary issue dates. Section V enumerates the results of the regression-based tests of the noise trader hypothesis. Section VI concludes. 4

5 2. DATA My analysis is based on 265 equity carve-outs taken from Securities Data Corp. s Worldwide New Issues data base. An initial screen that required SDC s spinoff code be set to Yes and the units code be set to No yielded 677 equity carve-outs. 2 Of these, 301 had parents with a valid CRSP IPERM identifier. After further exclusions for various reasons, the sample consists of 265 carve-outs undertaken between 1981 and This compares with 336 carve-outs over analyzed by Vijh (2002). Table I reports descriptive statistics on selected characteristics of parents and carved-out subsidiaries. 4 Carve-outs are undertaken by firms of widely varying sizes, with the market value of total parent equity ranging between $1.7 million and $75 billion. Although typically less than half the size of the consolidated entity, the market value of carved-out subsidiaries common equity nevertheless ranges from $3.4 million to $59 billion. The median dilution in the parent s pre-carve-out interest in the subsidiary due to the IPO is 24.3%, where dilution is defined as the percentage of subsidiary shares held by the parent prior to the carve-out less the percentage held by the parent after the carve-out. In 64% (14%) of carve-outs, only primary (secondary) common shares are offered to the public. Of parents, 78% (22%) are traded on the NYSE or AMEX (NASDAQ), and 44% (56%) of subsidiaries begin trading publicly on the NYSE or 2 SDC s spinoff code is #438. It is set to Yes when the issue is deemed by SDC to occur when a company decides to distribute shares representing ownership in a division or subsidiary of the company that will now trade separately from its former parent. Transactions labelled spinoffs in the Worldwide New Issues Database are therefore public offerings, not true spinoffs (pro rata distributions of subsidiary stock to parent shareholders). SDC s units code is #940. It is set to Yes when the offering is for units. Since a unit represents a combination of securities such as common stock, debt, preferred stock, and warrants, unit public offerings may have very different economic characteristics and be issued by firms for different reasons than all-equity carve-outs. I therefore excluded all subsidiary IPOs that were units. 3 The sample currently ends in 1995 because the data were originally collected in The following criteria had to be satisfied for an observation in the initial group of 301 cases to be included in the final sample. The subsidiary had to have a CRSP IPERM identifier; the parent had to be a U.S. company and have undertaken the carve-out after 1/1/81; there had to be either an announcement of the carve-out found through Lexis/Nexis or a filing of the subsidiary stock with the SEC; the parent had to be traded on either the NYSE, AMEX, or NASDAQ for one year prior to the initial announcement; the parent stock price at both the announcement and issue dates had to be at least $1.25 per share; and finally, the carve-out proceeds could not exceed the total market value of the parent s equity just prior to the carve-out announcement. In total, these criteria eliminated just 36 of the 301 cases (12%). 4 Since SDC s data on carve-outs covers the period , the sole use of SDC to identify a sample of carve-outs means that the sample encompasses a different time period than some of prior studies. Although the sample is larger than is any prior published work, to the extent that SDC omits carve-outs for whatever reasons, the sample composition could differ from that in earlier research. 5

6 AMEX (NASDAQ). Carve-outs are also large financing transactions, generating an average of $149 million of net proceeds. They are underpriced an average of 7.0%. The median trading days between the announcement of the parent s intention to undertake a carve-out and the subsidiary stock issue date is 45 days. Finally, the number of carve-outs undertaken per year (classified by the year of the subsidiary IPO) varies between five in 1982 and 32 in EVENT STUDIES: EXCESS STOCK RETURNS AT AND AROUND THE INITIAL CARVE-OUT ANNOUNCEMENT AND THE SUBSIDIARY IPO DATES 3.1 Parent Excess Returns The empirical analysis begins by replicating the event study finding in prior research that carve-out announcements are associated with a positive mean excess parent stock return. I then describe the behavior of parent stock prices when the subsidiary IPO takes place. The date of the subsidiary IPO has not been examined in carve-out research, but the parallel or equivalent date for seasoned equity offerings has been the subject of several studies (Mikkelson and Partch, 1986; Barclay and Litzenberger, 1988; and Kadlec, Loderer and Sheehan, 1994). So as to provide a medium-term stock price performance context within which to embed the event studies, I calculate mean cumulative daily parent excess returns earned by parents beginning one year prior to the carve-out announcement and ending one year after the subsidiary IPO. These are shown with the thick solid line in Figure I. 5 Table II details the daily mean and median parent excess stock returns at and immediately around the announcement and issue dates. Event day ANN<0> for the carve-out announcement is defined as the earlier of the first public disclosure of the parent s intention to undertake the carve-out and the filing of the subsidiary s common stock with the SEC. 6 Event day ISS<0> for the subsidiary stock issue date is the first day the subsidiary s stock is reported on CRSP. For both events, parent daily excess returns are calculated from a shrinkage-type market model whose parameters are estimated over 5 Since the length of the period beginning immediately after the carve-out announcement and ending immediately be fore the subsidiary issue date varies across parents, for graphical purposes I interpolate returns across 50 event and trading days (the sample median is 45 trading days). 6 When it preceded the SEC filing date, the first public disclosure of the parent s intention to undertake the carve-out was usually made in a press release. Such a disclosure was found for 36% of the sample, and in each case was identified by searching Lexis/Nexis (library = BUSFIN, file = ALLNWS) for relevant stories for the six months prior to the SEC filing date. 6

7 the pre-announcement window ANN< 450, 251> using the value-weighted market. 7 The market is defined as the value-weighted union of all NYSE, AMEX and NASDAQ firms. 8 As in any event study, the appropriate length of an event window is uncertain. The longer the window, the more likely it is that it will fully capture price changes due to the event and accommodate minor misdatings of the event by the researcher. However, the longer the window, the more likely it is that the effects of contaminating information and other noise will be inadvertently included. The criteria that I adhered to were as follows. For each of the carveout announcement and subsidiary IPO events, event day <0> is always included. Other days within the window < 10,10> are only included if the day exhibits either a reliably nonzero mean excess return, or an abnormally high or low percentage of positive excess returns. In addition, each day satisfying the above condition had to be adjacent to another day on which one or more of the above criteria were satisfied, beginning with event day <0>. The application of these criteria led to the event windows being defined as ANN< 1,1> for the carve-out announcement and ISS<0,4> for the subsidiary stock issue. Each is highlighted in Table II. The end point for the post-announcement period ANN<2,10> and the beginning point for the pre-issue period ISS< 10, 1> were set relatively close to day <0> to avoid overlap in the post-announcement and pre-issue windows. Since the number of trading days between the announcement and subsidiary IPO ranges between 20 and 251, none of the sample had overlapping post-announcement and pre-issue windows. 7 Each parent s intercept and slope parameters are shrunk by setting them equal to the cross-sectional sample means. Mean and median market model parameter estimates are always very similar. Shrinkage discards the potentially useful information contained in individual market model parameter estimates, but does not induce extreme cases of parameter estimation error into calculations of excess stock returns (particularly those cumulated over medium- and long-run periods). Since parameter estimation noise can be substantial, shrinkage-based excess returns are likely to increase the power of test statistics based on the cross-sectional sample moments of excess returns. For example, the largest (smallest) estimated intercept among the sample of 265 parents is a whopping 1.19% ( 1.15%) per trading day. Annualized on a purely additive basis over 250 trading days, these extremes cumulate to one-year excess returns of 298% ( 288%). Annualized on a buy-and-hold basis, they cumulate to 1,825% ( 97%). Setting aside market adjustments, it is difficult to believe that such amounts are unbiased estimates of any firm s annual expected return. However, no major inferences of the paper are affected by the choice of a shrinkagebased approach in lieu of the more standard non-shrinkage method. 8 Using value-weighted market returns avoids the approximately 6% per year negative bias imparted by compounding daily excess returns derived from an equally-weighted market index (Canina, Michaely, Thaler and Womack, 1998). However, the results in Table II and inferences of the paper are insensitive to a variety of alternative measurement methods, including buy-and-hold returns and excess returns computed using standard non-shrunk market model parameters. 7

8 The most striking features of Figure I are the sharp upward and downward jumps over the event windows ANN< 1,1> and ISS<0,4>. In panels A and B of Table II, I describe the mean and median parent excess stock returns immediately around the announcement and issue dates, from which I note the following points. First, I replicate the result present in all prior research in that I find a reliably positive parent stock price reaction at the announcement by the parent of its intention to carve out the subsidiary (panel A). The mean parent excess return for my sample is 2.3% (t-statistic 9 = 6.5); the median parent excess return is 1.4%; and 61.1% of individual excess returns are positive (t-statistic relative to a null of 50% = 3.6). Second, I document in panel B that there is also a negative mean excess return of 2.4% at the subsidiary stock issue date (tstatistic = 6.8). This result is new in the equity carve-out literature. The median parent excess return is 1.6% and 32.1% of individual excess returns are positive (t-statistic relative to a null of 50% = 5.8). Third, as indicated by the results of a battery of sensitivity tests reported in panel C, the issue date result is robust to time and a variety of parent characteristics. These include the quality of the underwriter used in the subsidiary IPO, the mix of primary and secondary shares issued, the number of trading days between the announcement and issue dates, and alternative measures of excess returns. 10 Finally, the market s reaction at and around the subsidiary IPO date is quite different to that found at the issue date for seasoned equity offerings. Mikkelson and Partch (1986), Barclay and Litzenberger (1988), and Kadlec, Loderer and Sheehan (1994) in aggregate report a distinct stock price drop of around 1.5% for NYSE- and AMEX-listed seasoned equity offerings and 3.5% for offerings on the NASDAQ immediately prior to and on the date seasoned stock is issued. The price drops are transitory and are completely reversed out over either the 5- or 25-day periods immediately after the issue, depending on the study. In sharp contrast, carve-outs experience no such price rebound. The fall in the carve-out parent s stock price begins at the issue date, not before as with seasoned equity offerings, and is not undone in the immediate post-announcement period. 9 All t-statistics are based on the simple cross-sectional variance in the underlying distribution. Crosssectional t-statistics are generally more robust to violations of normality and more conservative than alternatives such as those proposed by Patell (1976) and Jaffe (1974). However, identical inferences are obtained from these alternatives. 10 For those very few carve-outs where the subsidiary is jointly-owned, both parents are included in the event study. However, there are even fewer carve-outs where more than one parent of a jointly-owned 8

9 The negative mean parent excess return at the issue date as compared to the positive mean parent excess return at the carve-out announcement implies that there is predictable bad news for parent stock prices following soon after unexpected good news. Since it is very rare for a parent to announce but not execute a carve-out, the probability that the subsidiary will complete the carve-out is close to one. 11 The negative mean parent excess return at the subsidiary stock issue date is therefore inconsistent with market efficiency, since market efficiency requires that public information at time t (such as the carve-out announcement) cannot expect to earn an abnormal private return beyond time t. 3.2 Excess Returns Earned by Rivals to Parents and Carved-Out Subsidiaries As suggested in Figure I, average parent excess returns over the period spanned by the announcement and issue dates are close to zero. The mean parent excess return is 0.9% (tstatistic = 0.8) over a window whose median length is 50 trading days. This is consistent with the noise trader proposition that carve-outs are on average value-irrelevant events for parents. As a cross-check on this inference, in Figure I and Table III I report the mean daily excess returns earned by portfolios of rivals to parents and rivals to carved-out subsidiaries at and around both the carve-out announcement and subsidiary stock issue dates. If parent excess returns at the carve-out announcement and/or subsidiary IPO dates reflect value-relevant information, and if this value-relevant information at least in part pertains to the industries comprising the pre-carve-out parent, then I might expect to observe externalities on the stock prices of rivals to the parent and/or the carved-out subsidiary at these dates. Externalities of this kind have been found in other event studies, such as going-private bids (Slovin, Sushka and Bendeck, 1991), Chapter 11 announcements (Lang and Stulz, 1992), stock repurchases (Hertzel, 1991), and commercial bank common stock issues (Slovin, Sushka and Polonchek, 1992). Since both before and after the carve-out the parent can be considered as being comprised of two firms (in potentially different industries), I compare parent excess returns to those of rivals to the parent-alone and the carved-out subsidiary. Rivals to parents-alone (subsidiaries) subsidiary is publicly traded. Excluding these from the event-study does not materially affect either the mean parent excess returns reported in Table II or the inferences based on them. 11 When the same criteria that were applied to SDC s Completed Issues File were applied to SDC s Withdrawn Issues File (i.e., SPIN be set to Yes and UNITS be set to No ), only 14 cases were obtained. This compares to 677 from the Completed Issues File. However, to the extent that SDC omits carve-outs for whatever reasons, the 14 withdrawn carve-outs may not be exhaustive. 9

10 are defined as firms with the same primary 4-digit SIC code as the parent-alone (subsidiary) one year prior to the carve-out announcement. Where possible, I select three rivals per parent-alone and subsidiary. 12 Table III indicates that the mean excess stock returns experienced by rivals to parentsalone and subsidiaries at both the carve-out announcement and subsidiary stock issue dates are not reliably different from zero. I interpret this as consistent with sophisticated investors determining that there are no economically significant and unexpected systematically positive or negative real valuation implications arising from parents carve-out announcements THE NOISE TRADER HYPOTHESIS FOR THE ISSUE DATE ANOMALY 4.1. Hypothesis Development Logically, there are four possible interpretations of the announcement and issue date mean parent excess returns. Either the market is efficient on both dates, on neither date, on only the announcement date, or on only the issue date. The reliably negative mean issue date return rules out the first alternative. With regard to the second alternative, I assume that the probability that the market is inefficient on both dates is lower than the probability that it is inefficient on only one of the two dates. Although it is possible that the market is efficient on the announcement date but inefficient on the issue date, I see this third alternative as improbable. By definition, market inefficiencies are caused by unsophisticated investors, who by virtue of their unsophistication are most likely to affect parent stock prices when they observe data that is both highly salient and parent-specific. However, when the subsidiary IPO takes place, the financial press almost always focuses its attention on the subsidiary, not the parent. While parent and subsidiary equity values are clearly linked through the parent s (diluted) post-carveout ownership in the subsidiary, implying that a drop in parent stock price might arise from subsidiaries on average being overpriced at the issue date, 68% of equity carve-outs are 12 SIC codes were obtained from annual issues of the Standard & Poor s Register. The mean number of valid 4-digit SIC matched rivals per parent-alone is 2.8 relative to a maximum of This inference may have low power for two reasons. First, there may truly be industry-specific information conveyed by carve-out announcements and subsidiary stock issuances, but the valuation implications of the information may vary from rival to rival and be mean zero. Second, the value-relevant information captured in parent stock price revisions at the announcement and subsidiary IPO dates may be entirely parent-specific, and have no industry-wide component to it. 10

11 underpriced. Moreover, if the inefficiency lay at the issue date, one might expect a subsequent rebound in the parent s stock price. This is not supported by panel B of Table II. The seemingly counterintuitive explanation for the issue date anomaly that I therefore develop and test in this paper is that the market inefficiency lies not at the issue date but at the earlier carve-out announcement. I hypothesize that the mean revaluation in parent stock prices assessed by sophisticated investors at the carve-out announcement is zero, and that the observed positive mean parent announcement date excess return is due to misplaced noise trader demand. As defined by Black (1986), noise traders are investors who trade on noise as if it were real information. Since he assumes that both noise traders and sophisticated investors are present in the market, Black conjectures that the price of a stock reflects both noise and information, and that because of this, prices will tend to converge over time toward fundamental value. I therefore further propose that demand by noise traders who misinterpret the value-relevance of equity carve-outs leads to only a temporary overvaluation of the parent s stock price at the carveout announcement, and that this overvaluation is corrected by sophisticated investors at the subsidiary stock issue date. I do not model why sophisticated investors find it optimal to undo the overvaluation at, rather than before, the subsidiary issue date. Rather, I make the prediction conditional on the patterns in mean parent excess returns observed in Figure I and Table II. While the aggregate implications of noise traders have been modeled by De Long, Shleifer, Summers and Waldmann (1989, 1990, 1991), there have been few empirical tests aimed at assessing the direction and/or magnitude of the impacts of noise trading in individual securities. This may bear out Black s original skepticism as to the ability of conventional empirical tests to distinguish the noise trader hypothesis from conventional asset pricing models. The area that has been used most as a test vehicle in finance is the closed-end fund puzzle (Lee, Shleifer and Thaler, 1991; Pontiff 1996). In the accounting literature, Hand s (1990) extended functional fixation hypothesis is also an application of the noise trader concept. 4.2 Proxies for the Components of Noise Trader Demand I develop proxies for three components of noise trader demand at the carve-out announcement. Each aims to measure a portion of the extent to which noise traders misperceptions of the value-relevance of carve-outs may differ from the rational revaluation, which I hypothesize to be zero. The proxies are the expected carve-out proceeds relative to the 11

12 size of the parent, abnormal parent trading volume at the carve-out announcement, and whether the parent trades on the NASDAQ or the NYSE + AMEX. I argue that noise traders are more likely to view bigger carve-outs as more value-enhancing to parents; more likely to impact parent price the larger the abnormal volume they generate; and more likely to be concentrated in NASDAQ than in NYSE or AMEX stocks Size of the carve-out relative to the size of the parent The first proxy for the extent of noise traders misperceptions is the size of the carve-out relative to the market capitalization of the parent. I conjecture that noise traders will erroneously believe that parent shareholder wealth will be increased by the carve-out, and moreover that the bigger the carve-out the more will be the wealth increase, for two reasons. First, noise traders may mistakenly suppose that parent shareholder wealth will be increased by a carve-out because carve-outs have been (and still are) often mislabeled as spin-offs in the financial press. Spin-offs are more common events that are well known to increase parent stock prices by an average of 3% (Schipper and Smith, 1983; Vijh, 1994). Second, carve-outs almost always trigger accounting or paper gains for parents in the quarter and year in which the carve-out is undertaken because the price at which the subsidiary s shares are issued almost always exceeds the subsidiary s pre-carve-out book value per share. Under current accounting rules, the carve-out effectively enables the parent to mark a portion of its historical cost interest in the subsidiary to market value. In approximately 80% of carve-outs the parent chooses to recognize the (often substantial) accounting gain in its net income rather than have it bypass the income statement by taking it directly to shareholder equity. 14 Since they are unsophisticated, noise traders may fail to realize the book nature of carve-out gain, and instead treat the gain as they would ordinary income. 15 My proxy for the gain that would be observable by unsophisticated investors at the carveout announcement is the expected gross proceeds. This is because the ex post accounting gain is 14 Since 1983 parents have been allowed to take the gain either directly into net income on a before- or after-tax basis or directly to shareholders equity, a choice that is unique in U.S. GAAP. Hand and Skantz (1997) document that 81% of the time the gain is taken to net income and that the gain is a median of 59% of annual parent pre-gain net income in the carve-out year. 15 This view does not require that an estimated carve-out gain actually be disclosed at the announcement date, only that with some positive probability the knowledge that the transaction is an equity carve-out 12

13 empirically highly correlated with the expected gross proceeds from the carve-out. The proxy, denoted EPRCDS, is defined as the relative size of the expected gross carve-out proceeds scaled by the market value of the parent s equity two days prior to the carve-out announcement. The expected gross carve-out proceeds is defined as the number of subsidiary shares filed to be issued globally multiplied by the average of the expected high and low offer prices per share Abnormal trading volume at the carve-out announcement The second component of noise trader demand for which I construct a proxy is abnormal parent trading volume in the carve-out announcement window ANN< 1,1>, denoted ABVOL_ANN. I propose that the likelihood that noise traders impact a parent s stock price at the carve-out announcement is increasing in the extent to which their relative share of trading volume rises on that date, and that this rise is captured by, or is positively related to, ABVOL_ANN. While abnormal trading volume following the release of new information could be due to sophisticated investors, to the extent that fully rational individuals holding common information sets do not benefit from trading and therefore will not trade (Milgrom and Stokey, 1982), this concern is dampened. Abnormal volume is defined as actual less expected parent volume, deflated by the total number of parent shares outstanding, where expected daily volume is estimated as mean parent volume over the pre-announcement period ANN< 200, 11> Parent exchange listing The third proxy aimed at capturing the valuation impacts of noise trader demand is whether the parent trades on the NASDAQ or NYSE + AMEX. I use this exchange listing dummy variable to proxy for the probability that the marginal investor is a noise trader. I conjecture that relative to NYSE and AMEX listed securities, NASDAQ listed stocks are likely to be on average smaller, less followed by analysts, and less frequently held by institutional investors. I therefore conjecture that the probability that the marginal investor is a noise trader leads to an expectation that annual parent net income will be substantially increased in the carve-out year. A recent article in the Economist even highlighted the profit from gain on issuance (Economist, 4/12/97). 16 Following Gould and Kleidon (1994), volume is adjusted for assumed double-counting of trades between dealers. Parent volume is multiplied by ADJ = 0.8 (0.5) if the shares are traded on the NYSE or AMEX (NASDAQ). 13

14 will be higher for NASDAQ-listed stocks than for NYSE- or AMEX-listed stocks. 17 The dummy variable PAREX is therefore set to one if the parent s common stock trades on the NASDAQ at ANN< 2>, and zero if on the NYSE or AMEX. 4.3 Regressions The tests of the noise trader hypothesis are structured through two sets of cross-sectional regressions, described in matrix notation in equations (1) and (2) below: ER_ANN = α ann + NT_ANN. β ann + SOPH_ANN. γ ann + ε (1) ER_ISS = α iss + NT_ANN. β iss + SOPH_ANN. γ iss + SOPH_ISS. π + v (2) ER_ANN and ER_ISS are parent excess returns over the carve-out announcement and subsidiary stock issue event study windows ANN< 1,1> and ISS<0,4>, respectively. NT_ANN and SOPH_ANN are proxies for the components of noise trader and sophisticated investor demand at the carve-out announcement, respectively, and SOPH_ISS is a set of proxies for sophisticated investor demand at the subsidiary stock issue date. Per section III.B, the matrix NT_ANN is comprised of the vectors EPRCDS, ABVOL_ANN, and PAREX. 4.4 Primary predictions The noise trader hypothesis proposes that demand by noise traders who positively misinterpret the true value-irrelevance of equity carve-outs leads to a temporary overvaluation of the parent s stock price at the carve-out announcement that is then corrected at the subsidiary stock issue date by sophisticated investors. In terms of equations (1) and (2), this translates into the following key parameter predictions: 17 In the accounting literature, Hand (1990) has proxied for the probability that the marginal investor is unsophisticated using a continuous measure, namely the proportion of a firm s stock held by noninstitutional holders. I use a dummy variable because the sample period is 15 years long, over which time the average proportion of stocks held by noninstitutional holders has dropped quite substantially. Hand s sample period was only the 3-year period

15 H1: α ann = 0 H2: β ann > 0 H3: α iss = 0 H4: β iss = β ann Noise traders at the carve-out announcement are predicted to misperceive bigger carveouts as more value-enhancing (β ann,2 > 0); have more impact on a parent s stock price the larger is parent abnormal volume (β ann,1 > 0); and be more concentrated in NASDAQ than in NYSE or AMEX stocks (β ann,3 > 0). Since the mean revaluation in parent stock prices assessed by sophisticated investors at the carve-out announcement is assumed to be zero, if noise trader demand is adequately proxied by NT_ANN, then the mean residual parent excess return will be zero (α ann = 0). If mispricing induced by noise traders at the carve-out announcement is undone by sophisticated investors at the subsidiary IPO, then β iss,i = β ann,i i. Finally, if noise traders exert price pressure on the issue date that is mean zero, and sophisticated investors correcting of earlier mispricing is fully captured by NT_ANN, then the mean residual parent excess return will be zero (α iss = 0). 4.5 Secondary predictions New information at the carve-out announcement The noise trader hypothesis permits stock price to reflect noise and true information. The matrix of proxies for rational revaluations in the parent s stock price is denoted SOPH_ANN. If valid, the elements of SOPH_ANN should be able to explain cross-sectional variation in parent excess returns at the announcement date but not at the issue date, since at the subsidiary stock issue date SOPH_ANN is stale and therefore of no value to sophisticated investors. I construct SOPH_ANN to be a matrix of four dummy variables: SUBEX, DUM_80%, DUM_50%, and DUM_ALLSEC. SUBEX is set to one if the subsidiary first trades on the NASDAQ; DUM_80% (DUM_50%) are set to one if the parent holds less than 80% (50%) of outstanding subsidiary shares after the IPO; and DUM_ALLSEC is set to one if no primary subsidiary shares are offered in the IPO. I propose that listing the subsidiary shares on the NASDAQ may be a weaker signal of management s confidence in the economic value created by the carve-out than is listing on the NYSE or AMEX (γ ann,1 < 0). In addition, I hypothesize that 15

16 reducing the parent s fractional ownership of the subsidiary below 80% (thereby foregoing tax benefits typically associated with filing a consolidated tax return) and/or reducing the parent s fractional ownership below 50% (thereby permitting certain kinds of off-balance sheet financings) will be perceived as negative signals by sophisticated investors. 18 Finally, I suggest that a carve-out in which some or all of the offered shares are secondary stock is a negative signal to sophisticated investors (γ ann,4 < 0). Secondary shares differ from primary shares in that the latter are newly issued, whereas the former are already issued and most likely held by the parent. A carve-out in which some or all of the shares are secondary stock may be bad news for parent stockholders because the sale of secondary shares may trigger a tax liability for the parent. This will be especially acute for carve-outs in which all the shares sold to the public are secondary. Formally, the predictions corresponding to the discussion in this section are that: H5: γ ann < 0 H6: γ iss = New information at the subsidiary stock issue date Equation (2) contains one further set of predictions. Over the subsidiary stock issue event study window ISS<0,4>, new information may become available that is useful in the rational revaluation of the parent s stock price. The proxy for this new information, denoted by the matrix SOPH_ISS, consists of the three variables ABVOL_ISS, SUB_UPX, and SUB_ER. ABVOL_ISS is abnormal parent trading volume over ISS<0,4>. Since the issue date is the first opportunity for trading in the stock of the new public entity, portfolio managers may wish or need to rebalance their portfolios by reallocating some of their pre-issue investment in the parent into the newly floated subsidiary. This may lead to selling pressure on the parent s stock associated with abnormal trading volume in the parent s stock. In terms of equation (2), I therefore predict that the coefficient on ABVOL_ISS will be negative (π 1 < 0). A second piece of information that becomes publicly available over ISS<0,4> is the subsidiary excess stock return over the same period. This should positively correlate with the change in the parent s stock price because the parent retains a fractional interest in the 18 When the parent s fractional ownership in the subsidiary falls below 50%, it reports only its net investment for financial purposes. Thus, a parent with a highly leveraged subsidiary can remove the subsidiary s debt from the overall entity s balance sheet when its interest in the subsidiary falls below 50%. 16

17 subsidiary. To increase the power of the tests, I divide the subsidiary excess stock return over the window ISS<0,4> into two parts: the degree to which the subsidiary is underpriced as of the end of ISS<0>, denoted SUB_UPX, and the subsidiary excess return over ISS<1,4>, denoted SUB_ER. I predict that the coefficient on SUB_ER will be positive (π 2 > 0). SUB_UPX is defined as the difference between the subsidiary price per share at the end of the first day of public trading and the offer price per share, as a percentage of the offer price per share. Despite its size and maturity, the IPO literature does not provide an unambiguous prediction as to whether larger underpricing will be associated with an increase or a decrease in the parent stock price (Ibbotson and Ritter, 1995). 19 I therefore make no prediction for π 3, the coefficient on SUB_UPX. 5. EMPIRICAL TEST RESULTS ON NOISE TRADER PREDICTIONS 5.1 Regression Results at the Carve-Out Announcement The results of estimating equation (1) are reported in panel A of Table IV. They uniformly conform with predictions H1 and H2 from the noise trader hypothesis. Thus, consistent with H1, the estimated intercept α ann is never reliably different from zero. Consistent with H2, in both regressions R1 and R2 each β ann,i is reliably positive. In total, the proxies for noise trader demand in regression R1 explain a substantial 18% of the cross-sectional variation in parent excess returns at the carve-out announcement. However, the data are inconsistent with hypothesis H5. The estimated coefficients on the dummy variable proxies for rational revaluations in the parent s equity due to new value-relevant information released in the carve-out announcement are individually and jointly insignificantly different from zero. While these results might arise because proxies in SOPH_ANN are measured with enormous error, I instead interpret them as indicating that the elements of SOPH_ANN are simply uninformative to sophisticated investors. 19 Some view underpricing as a cost created institutionally by the IPO process itself, while others view underpricing as a signal of the quality of the firm undertaking the IPO. The former (latter) view predicts that larger underpricing will be negatively (positively) associated with parent issue date excess returns. 17

18 5.2 Regression Results at the Subsidiary IPO Date A legitimate criticism of the noise trader hypothesis, were it only to be tested through equation (1), is that ABVOL_ANN, EPRCDS, and PAREX are not proxies for noise trader demand at all. Rather, they are in actuality excellent proxies for the rational revaluations in the parent s equity due to real information in the carve-out announcement. This concern can be directly addressed by noting that since ABVOL_ANN, EPRCDS, and PAREX become public information at the carve-out announcement, they are stale by the time of the subsidiary IPO. The inferential power thereby obtained by estimating equation (2) in addition to equation (1) is that the efficient market hypothesis dictates that α iss = β iss,1 = β iss,2 = β iss,3 = 0. Since SOPH_ANN is also stale by the subsidiary IPO date, the efficient market hypothesis further predicts that γ iss = 0. In contrast, if as proposed by the noise trader hypothesis the market is inefficient at the carve-out announcement but efficient at the issue date, and the mispricing induced by noise traders at the carve-out announcement is undone by sophisticated investors at the issue date, then according to H3 and H4, α iss = 0 and β iss = β ann. The noise trader hypothesis also makes the prediction that γ iss = 0. This is because noise traders are assumed to not react to non-salient data such as that contained in SOPH_ANN. Parameter estimates for equation (2) are reported in panel B of Table IV. The results are strongly consistent with all noise trader predictions. First, as predicted by H3, in regressions R3, R6 and R7 where the matrix of noise trader demand proxies NT_ANN is controlled for, the estimated intercept α iss is never reliably different from zero. Second, as predicted by H4, regression R3 demonstrates that the coefficient estimates on NT_ANN are all reliably negative when NT_ANN is considered alone. Third, in total the stale elements of NT_ANN explain an astonishing 11% of the cross-sectional variation in parent excess returns at the issue date. Fourth, the p-values on the three individual restrictions β iss,1 = β ann,1, β iss,2 = β iss,2, and β iss,3 = β ann,3 across regressions R1 and R3 are 0.16, 0.55 and 0.54, respectively. The p-value on the joint restriction that the vector β iss is the negative of the vector β ann is As is suggested by the very similar coefficient estimates obtained in regressions R3 versus R6 and R7, these inferences do not change when the independent variables SOPH_ANN and SOPH_ISS are included. The finding that β iss = β ann is all the more startling in the light of the fact that the product moment and rank correlations between parent excess returns at the announcement and issue dates are not reliably different from zero. The Pearson product moment correlation is 18

19 0.09 (2-tailed p-value of 0.07), while the Spearman rank correlation is 0.05 (2-tailed p-value of 0.23). The lack of any simple correlation juxtaposed with rich evidence of multivariate partial correlations indicates that mispricing induced by noise traders can be subtle and readily hidden from casual observation. Regressions R4 and R6 are generally consistent with the predictions in H6. With the exception of the observed negative coefficient on SUBEX, the proxies for sophisticated investor demand at the carve-out announcement have no individual explanatory power at the subsequent issue date. Moreover, the p-value on the vector hypothesis γ ann = 0 is 0.14, strictly speaking rejecting H6. Insofar as hypothesis H5 is concerned, regressions R5 and R6 report the coefficient estimates on SOPH_ISS, the proxies for sophisticated demand at the issue date. In both cases, only π 2, the coefficient estimate on the carved-out subsidiary s excess return SUB_ER over ISS<1,4>, is significant and in the predicted direction. 6. CONCLUSIONS In this paper I have asked the controversial question of whether accounting for noise trader demand in event studies can yield valuation inferences that differ from those obtained under the efficient market hypothesis. I have proposed that in the case of the uniformly held view from event studies that equity carve-outs create value for their parents, the answer seems to be yes. In particular, I hypothesized that the positive mean excess stock return earned by parents at the carve-out announcement documented by prior research is due to noise traders who optimistically misinterpret a carve-out s true value-irrelevance, rather than to the impounding into price of new value enhancing information by sophisticated investors. I offered three key pieces of evidence in support of this hypothesis. First, I found that parents experience a reliably negative mean excess stock return of 2.4% when the carved-out subsidiary goes public. This is in contrast with the 2.3% rise at the announcement date a median of only 45 trading days earlier. The mean excess return for the (on average) 45-day window spanning the announcement and issuance is a statistically insignificant 0.9%. Second, I found that proxies for the intensity of three components of noise trader demand at the carve-out announcement together explain 11% of the cross-sectional variation in parent excess returns at the subsequent subsidiary stock issue date. Finally, I was unable to reject the hypothesis that the coefficients on the three noise trader demand proxies in a cross-sectional regression whose 19