Do Tracking Stocks Reduce Informational Asymmetries? An Analysis of Liquidity and Adverse Selection

Transcription

1 Do Tracking Stocks Reduce Informational Asymmetries? An Analysis of Liquidity and Adverse Selection John Elder* Pankaj K. Jain and Jang-Chul Kim JEL Classifications: G14, G34 Keywords: Tracking stock, bid-ask spreads, restructuring. * Elder and Kim are from North Dakota State University, Department of Finance, PO Box 5137, Fargo, ND Phone: ; john.elder@ndsu.nodak.edu and jckim@memphis.edu, and Jain is from the Fogelman College of Business and Economics, The University of Memphis, Memphis, TN Phone: (901) ; Fax: (901) ; pjain@memphis.edu.

2 Do Tracking Stocks Improve Informational Asymmetries? An Analysis of Liquidity and Adverse Selection Abstract A firm s announcement that it intends to restructure based on tracking stock is usually associated with a significantly positive stock price reaction, at least in the short-run. Typically, this reaction has been attributed to expected reductions in a diversification discount, via reduced agency costs and/or reduced informational asymmetries. The existing literature has investigated this latter hypothesis the impact of tracking stocks on informational asymmetries by focusing on the behavior of equity analysts: the number following the specified firms and the accuracy of their forecasts. The results thus far have been inconclusive. In contrast, we focus on the behavior of market makers. In particular, we analyze the liquidity provided by market makers, as measured by the bid-ask spread, when a firm announces an intention to issue a tracking stock and when a firm actually implements the tracking stock. Our results provide additional support to the growing evidence that restructurings based on tracking stock are not effective at reducing informational asymmetries in diversified conglomerates. 2

3 1. Introduction Tracking stock is a unique form of corporate restructuring in which a diversified conglomerate creates a new class of shares whose value is linked to a particular business segment. An important feature of a restructuring based on tracking stock is that additional financial disclosures are required, whereby the parent firm (the general division ) and the tracked segment (the business group ) file separate financial statements with Securities and Exchange Commission. It has been suggested that these additional disclosures may improve the information environment, thereby reducing informational asymmetries among investors, as well as the informational asymmetries between management and investors. There are both theoretical and institutional justifications, however, to suggest that the net impact on the informational asymmetries between investors is ambiguous. This paper details the issues involved and investigates the effect of restructuring based on tracking stock on informational uncertainties among investors by utilizing market microstructure-based tools. In particular, we examine changes in the liquidity provided by market participants during the announcement and issuance of tracking stocks. Although recent research has shown that, ex-post, tracking stocks have under-performed usual benchmarks (c.f. Billet and Vijh (2002) and Chemmanur and Paeglis (2000)), initial announcements of an intention to issue a tracking stock have tended to increase firm value in the short term. Billet and Mauer (2000), D Souza and Jacob (2000) and Elder and Westra (2000) document positive abnormal returns between 2 to 4% in the days surrounding such announcements. These gains have typically been attributed to expected reductions in a diversification discount through reduced informational asymmetries and/or reduced agency costs. For example, Zuta (2000) finds that conglomerates with tracking stocks tend to have lower diversification discounts than comparable firms, while Billet and Mauer (2000) find that such firms tended to have lower diversification discounts before the tracking stock is issued. Harper and Madura (2002) find evidence to suggest that the tracking stock structure reduces agency costs in conglomerates. 1

4 Several studies have investigated the impact of tracking stocks on informational asymmetries. The usual premise is that, since the SEC requires the disclosure of additional financial statements detailing the performance of the general division as well as the tracked business group, analysts can better focus on the performance of each segment. This would tend to increase both the number of analysts following the firm, and, since analysts tend to specialize in particular industries, the accuracy of their forecasts. Both of these factors may tend to reduce informational asymmetries. Such effects are formalized in, for example, Habib, Johnsen and Naik (1997), who present a model in which the trading of individual segments reduces uninformed investors uncertainty about asset values. There are, however, theoretical and institutional factors that may tend to counter this effect, making the net effect of the tracking stock structure on informational asymmetries ambiguous. The institutional factors include accounting and corporate governance issues associated with the tracking stock structure. For example, a tracking stock does not represent a legal claim on the assets of the associated business group. Instead, a tracking stock represents a claim on a fraction of the assets of the consolidated firm, where, in the event of liquidation, the claim is typically dependent on the proportion of the total market value accounted for by each class of stock. It may therefore be quite difficult for analysts to value the general division and the tracked business groups based on their liquidation value. A tracked business group is also not governed by an independent board of directors. Rather, the tracked business group is governed by the directors of the parent firm, with the shareholders typically having voting rights that float with the market value of their tracking stock relative to that of the total market capitalization of all classes of common stock for the firm. Such voting rights imply that the directors will answer to at least two groups of shareholders with potentially very different and competing interests, with the interests of the tracked group subordinate to the interests of the consolidated firm (c.f, Haas (1996)). This aspect of the tracking stock structure may introduce substantial uncertainties about how the tracked business group will be strategically managed relative to pure-plays in the same industry, and create difficulties in valuing the various business segments as going-concerns. 2

5 In addition, there are formal theoretical foundations to suggest that restructuring a conglomerate into various business segments does not reduce informational asymmetries. For example, Chang and Yu (2003) show that if markets are uncertain about the value of business segments, then informed traders are less likely to collect information on individual segments when they are structured as a conglomerate. Similarly, conglomerates may tend diversify away segment-specific informational asymmetries, as formalized by Subrahmanyam (1991), Gorton and Pennacchi (1993), Hadlock, Ryngaert and Thomas (2001). The empirical evidence on whether an equity structure based on tracking stock reduces informational asymmetry is mixed. With regard to analyst coverage, D Souza and Jacob (2000) do not find any significant increase in coverage after a firm issues a tracking stock, 1 while Zuta (2000) and Chemmanur and Paeglis (2000) find increased analyst coverage. Chemmanur and Paeglis interpret their results as indicating that decreased informational asymmetries are likely to have a positive effect on firm valuation, at least in the short run. In contrast, Billet and Vijh (2002) measure analyst earnings forecast errors, the dispersion of earnings forecasts and the market reaction to earnings announcements, each prior to the tracking stock announcement and after issuance. Their analysis suggests that there is little or no improvement in information asymmetry for the general division and some deterioration for the tracked business groups. An alternative and more direct measure of informational asymmetry, however, is based on the liquidity provided by market participants. The extant literature indicates that market makers provide less liquidity during periods of greater informational asymmetry that is, when they perceive a higher probability of trading with more informed traders. For example, Lee, Mucklow, and Ready (1993) and Krinsky and Lee (1996) find that such adverse selection costs induce market makers to widen spreads around earnings announcements. Similarly, if a corporate restructuring based on tracking stock impacts informational asymmetries, then it should possible to discern the sign and magnitude of the effect by 1 They argue that the main motivation for issuing tracking stocks is their tax-free nature compared to spin-offs, which create tax liabilities. 3

6 examining the liquidity provided by market makers during the relevant period. Similar empirical investigations have been conducted by, for example, Huson and Mackinnon (2003) and Clarke, Fee and Thomas (2003). Given the contradictory theoretical literature, the ambiguous empirical results and the availability of more direct measures of informational asymmetry, further empirical investigation seems warranted. As such, in this paper we examine the behavior of liquidity providers during periods when a conglomerate announces an intention to issue a tracking stock and when the tracking stock is actually issued. If market makers perceive that an impending restructuring based on tracking stock may allow some traders to make superior judgments about the value of the firm, then we should expect to find a decrease in liquidity around the announcement period (c.f., Kim and Verrecchia (1994)). In particular, given that tracking stock announcements are largely unanticipated, we should expect to find wider spreads on the announcement day, with a greater portion of the spread due to adverse selection. Similarly, if the restructuring effectively reduces informational asymmetries through the release of more detailed financial statements on the various business segments, then we should expect to see increased liquidity, as measured by narrower spreads, as the tracking stock structure is implemented. If, however, the other aspects of tracking stocks detailed above substantially counter this effect, or if such restructurings tend not reduce informational uncertainties for the reasons cited previously, then the impact on liquidity as the tracking stock structure is implemented may be negligible, or even negative. We conduct our analysis on the basis of 56 tracking stock announcements and 28 issuances between 1984 and 2002, utilizing data for liquidity variables from The Institute for the Study of Security Markets (ISSM) at the University of Memphis and the Trade and Quote Database (TAQ) from the New York Stock Exchange. Our results can be summarized as follows: shortly prior to a restructuring based on tracking stock, there is no significant change in liquidity. During the two-day announcement window, we find significant reductions in liquidity as market makers tend to increase spreads by about 10 basis points. Decomposing the total spread indicates a significant increase in the proportion due to adverse selection. This suggests that: (1) the announcements are largely unanticipated and (2) market makers and 4

7 other liquidity providers perceive an increased probability of trading with more informed traders during the announcement period, and respond by reducing liquidity. The implication is that an announcement of a restructuring based on tracking stock may tend to increase informational uncertainties, at least in the short run. As the tracking stock structure is implemented, we find only a marginal and statistically insignificant increase in liquidity for the general division, relative to a large and significant increase in liquidity for our control sample that is consistent with documented market trends. Moreover, the adverse selection component of the total spread is significantly larger as firms implement the tracking stock structure. Finally, cross-sectional regressions reveal that the observed effects are not driven by a subset of firms with particular characteristics. Rather, the effects are systemic throughout our sample. Our results then reinforce the empirical findings of Billet and Vijh (2002). Markets may have interpreted announcements to issue tracking stocks as value increasing events in the short-run, but the actual implementation of tracking stock is not likely to significantly reduce the informational asymmetries impacting diversified conglomerates. More generally, our results are consistent with those of Huson and Mackinnon (2003), who find that conglomerate spin-offs do not improve liquidity, although our sample is not large enough to discern statistically significant differences between restructurings that improve focus and those that do not. The failure of restructurings based on tracking stock to mitigate informational asymmetries is likely the result of either the uncertainties introduced by this unique form of restructuring, or the more general failure of conglomerate divestitures to improve liquidity. The remainder of the paper proceeds as follows: the next section provides an overview of tracking stocks. Section 3 develops the testable hypotheses and section 4 discusses the data sources and methodology. The results are presented in section 5 and section 6 concludes. 2. Tracking Stocks Tracking stock, also known as targeted or lettered stock, is a class of common stock whose value is linked to the performance of a specific business group within a diversified firm. Since its introduction in 1984, nearly sixty firms have issued or announced plans to issue tracking stock, with a disproportionate amount in the late 1990 s. A common justification for issuing tracking stock has been 5

8 that it unlocks the hidden value of a business segment by separating it, to some degree, from the parent. Since 1998, the tracked business group has often been, or intended to be, an internet pure-play, such as those proposed and/or issued by Donaldson, Lufkin & Jenrette (DLJ Direct), Staples (Staples.com), New York Times (Times Company Digital), Korn Ferry (Futurestep.com) and others. Despite its increasing prevalence, however, tracking stock is not particularly well understood. The creation of tracking stock for a business group within a diversified firm is, nominally, similar to spinning-off or carving out the division, in that each of the restructurings creates a new security whose value is linked to the associated business group. 2 The differences between the three forms of restructuring, however, are considerable. Under both spin-offs and carve-outs a new corporate entity is created with shareholders possessing the conventional rights: the right to elect a board of directors to oversee management, the right to vote on matters of great importance, and a claim against the new entity s net assets A tracking stock, however, does not represent a new corporate entity. A tracking stock structure is formed by creating a new class of common stock whose value is linked to the performance of a specific business group through special provisions introduced into the firm s articles of incorporation. This link is usually strongest through a limited claim on the earnings generated by the division. Typically, the dividends paid to the owners of tracking stock are dependent on the earnings generated by the tracked group, expressed as function of shareholder s equity or net income, although many firms issuing tracking track have also indicated that earnings for the tracked group are not likely to be positive in the near future. A substantial complication of the tracking stock structure is that the tracked group may share with the parent firm the cost of fixed inputs, such as corporate offices and payroll services, that it otherwise would not as separate corporate entity. Another complication introduced by the tracking stock structure relates to the allocation of the firm s physical assets to the various business groups. In particular, a tracking stock does not represent a 2 Shares in the tracking stock may be distributed either as a public offering, as dividends to existing shareholders, or as currency for an acquisition. 6

9 legal claim on the particular assets of the associated business group. Instead, tracking shareholders typically have a claim on a fraction of the assets of the consolidated firm, where that fraction fluctuates with the proportion of the total market value accounted for by each class of stock. Probably one of the most controversial aspects of a tracking stock is that the tracked group is governed by the directors of the parent firm, rather than their own board. This suggests that the interests of the tracked group will be dominated by the interests of the consolidated firm, potentially introducing serious conflicts of interest (Hass 1996). It also creates the opportunity for considerable crosssubsidization across business groups, either through exposure to the liabilities of the consolidated firm or through purposeful redirection of resources. These features of tracking stock significantly curtail the extent to which a tracked business group can be considered a pure play. Some firms have even indicated the potential for such conflicts in regulatory filings, such as Sprint has with its two tracking stocks, FON Group and PCS Group. These unconventional features may account for the range of opinions expressed in the financial press. Headlines have ranged from numerous claims, including by many practitioners, that tracking stocks unlock value 3 to On the Wrong Track: Complex Financial Innovations Like Tracking Stocks... Bring Few Benefits to Shareholders 4. More recently, critical press seems to dominate, with headlines such as Sprint Shows Pitfalls of Investing in Tracking Stocks Tracking Stocks and Liquidity: Testable Hypotheses Kyle (1985) and Glosten and Milgrom (1985) depict that, in the presence of information asymmetry, market makers earn the bid-ask spread from the uninformed noise traders (who trade for liquidity reasons) and lose the difference between full information value of the stock and trade price given 3 Boston Globe, Genzyme Tracking Stocks Are Off Track on Returns, June 24, 1999; Dow Jones Newswires Firms Turn to Tracking Stocks to Unlock Value of Web Units July 12, 1999; and New York Times Shares that Track Assets Add Value at a Cost, July 18, Financial Times, Complex Financial Innovations Like Tracking Stocks Allow Managers to Retain Control, but Bring Few Benefits for Shareholders, May Wall Street Journal, Sprint Shows Pitfalls of Investing in Tracking Stocks, March 7,

10 to the informed traders (who trade on the basis of private information). The magnitude of the spreads depends on the proportion of liquidity traders and informed traders, which affects the probability of adverse selection -- i.e. trading with an informed trader. Several studies have analyzed bid-ask spreads to empirically investigate the adverse selection environment in a market. Lee, Mucklow, and Ready (1993) and Krinsky and Lee (1996) examine earnings announcement effects for the existence of asymmetric information about expected earnings. They find significant and increasing adverse-selection costs around earnings announcements, with market makers widening spreads and decreasing quoted depth immediately before and after earnings announcements. Kim and Verrecchia (1994) argue that spreads widen because earnings announcements provide new information that allows certain traders to make judgments about a firm s valuation that are superior to the judgments of other traders. In this respect, restructurings based on tracking stock are similar in that they result in the production and release of new financial data about each tracked business segment. This data is typically released after the announcement and prior to implementation. Since some traders are perceived to better gauge the impact on firm value at the time of the announcement, as evidenced by significant announcement abnormal returns (e.g., Elder and Westra), we examine liquidity around the time of announcement, and hypothesize that: H1: Bid-ask spreads increase in the short-term when a firm announces the creation of a tracking stock. The short-term referred to in H1 represents the time period starting with the announcement and immediately following it. It excludes the period before the announcement. Unlike periodic earnings announcements which are anticipated by market makers, tracking stock announcements aren t released according to a pre-determined schedule. This leads to our second hypothesis: H2: Bid-ask spreads do not change before a firm announces the creation of a tracking stock, except in case of information leakage. The long-term implications of tracking stock issuance on liquidity is more interesting than these short-term changes. As discussed earlier, if the additional financial disclosures on the tracked business groups effectively mitigate the information asymmetry between informed and uninformed investors, and, 8

11 more generally, if the focus-increasing events reduce informational asymmetries, then we should expect that: H3: Bid-ask spreads, in particular the adverse selection component of spreads, decrease in the long-term after a firm announces and implements a tracking stock structure. In contrast, rejection of this third hypothesis would be evidence that the unique aspects of the tracking stock structure mitigate any effect of increased financial disclosure, or, more generally, that conglomerate divestitures do not to reduce informational asymmetries. As discussed previously, both the theoretical and empirical evidence on this hypothesis is ambiguous, although recent empirical evidence, such as that provided by Huson and Mackinnon, suggests it may be rejected. Finally, the magnitude of change in spreads may depend on the motives for the restructuring and the parent firm s characteristics. For example, Harper and Madura (2002) find that some firm-specific characteristics related to corporate governance help explain cumulative abnormal returns around a firms announcement of a tracking stock. Such variables may also be associated with a changes in liquidity around the announcement and implementation dates. We examine this possibility in a cross-sectional regression, which is described in detail below. 4. Data Sources and Empirical Methodology 4-1. Tracking Stock Announcement Dates Our initial sample consists of 57 tracking stock announcements occurring between 1984 and 2002, for which ISSM and TAQ data is available. These announcements were identified by searching the SEC documents, news archives of the Wall Street Journal, the Dow Jones Newswire, the New York Times, and other major news sources, as well as by checking against published lists of trading stocks. Table 1 catalogs these announcements in chronological order. The first column gives the announcements date, second column mentions the parent company s name, and the third column describes the purpose and/or the outcome of the tracking stock proposal. Of the 57 announcements, 27 were subsequently cancelled, and 1 other is postponed. For those tracking stocks that were issued, we give the ticker symbol 9

12 of the tracked group in the fifth column and the ticker symbol of the control firm in the final column. The control firms are selected based on a matching procedure, as described in sub-section 4-3 below. To be classified as an announcement, management must have indicated an intention to formally propose the tracking stock to shareholders. In many cases, this occurred prior to a formal filing with the Securities and Exchange Commission. We do not include news reports that describe management as considering a tracking stock, or other reports that do not indicate a direct decision by management to formally propose a tracking stock. Thus, many reports, especially during the year 2000, where management indicated only that they were considering a restructuring based on tracking stock are not included in our sample. For example, one prominent non-announcement was by Microsoft, who in July 1999 simultaneously floated several restructuring options, including the prospect of issuing tracking stock for its various divisions, but never committed to formally proposing the plan. Inclusion of such vague references would introduce considerable identification issues, since it is not clear how they are interpreted in the marketplace. Due to the accounting irregularities engulfing WorldCom, we also exclude that announcement and implementation from our analysis, although inclusion has no significant impact on our results. Thus, our estimates are based on a sample of 56 announcements. Note that for about half the announcements (28), the intention to issue the tracking stock was subsequently canceled or indefinitely postponed. Cancellations and postponements usually occur because the proposal was rejected by shareholder vote, or because management withdrew the proposal prior to a vote by shareholders due to negative feedback. In some cases, a formal registration statement was never filed with the SEC, while for others, such as Bellsouth, a registration statement was filed and later withdrawn. Since it was not known at the time of the announcement which of these would be subsequently canceled, and since deletion would introduce a survivorship bias, we do not exclude these announcements from our announcement sample. Our sample of implemented tracking stocks therefore consists of the remaining 28 firms (again, excluding Worldcom) Liquidity Variables 10

13 Data for liquidity variables is obtained from The Institute for the Study of Security Markets (ISSM) at the University of Memphis and the Trade and Quote Database (TAQ) from the New York Stock Exchange. We extract bid-ask quotes, transaction prices and volumes for these firms for every transaction in our sample windows, which are summarized below: Period Trading days relative to publication of announcement Benchmark period (-100,-93) Alternative benchmark period (-14,-13) H1: Announcement period (-1,0) H2: Pre-announcement (-3,-2) H3: Post-announcement: Short-term (+1,+2) H3: Post-announcement: Medium- term (+13,+14) H3: Implementation (Implement,Implement+1) The recorded announcement date is the date when the tracking stock news appeared in a published news source. Since some announcements may have been carried on previous day by newswire or the internet, the announcement window includes t= 1 and t=0. To ensure the robustness of our results, we report the liquidity statistics over two alternative benchmark windows. Both benchmark windows yield similar results, so our reported inferences are relative to the benchmark closest to the event date. Each observation in the data file includes the quote date, time-stamp, ticker symbol, bid price, ask price, bid-depth, ask-depth, and exchange code where the quote originated. Our initial sample has 3,672,239 quote observations. We apply the following data filters, which are standard in the microstructure literature (for example, Huang and Stoll 1996), to clean the data of errors and outliers: i. Delete quotes if either the bid price or the ask price is negative; ii. Delete quotes if either bid size or the ask size is negative; iii. Delete quotes if the bid-ask spread is greater than $4 or negative; iv. Delete trades and quotes if they are out of time sequence, or involve an error; v. Delete before-the-open and after-the-close trades and quotes; vi. Delete trades if the price or volume is negative; vii. Delete trades and quotes if they changed by more than 10% compared to last tick; viii. Delete unlisted firms and other firms that are missing in TAQ on any event date; ix. Delete firms that are missing in CRSP or Compustat. This filtering process reduced the number of useable observations by 2.4%, resulting in a final sample of 3,584,105 observations pertaining to 56 tracking stock announcements from 48 unique firms. 11

14 We compare spreads, volume, number of trades, and an adverse-selection component of spreads for the periods indicated above. Spreads are defined as follows: Quoted spread = (Ask Price Bid Price) (1) Effective spread = Transaction price Quote midpoint * 2 (2) Percentage (or relative) spread = Quoted spread / Quote midpoint (3) Percentage (or relative effective spread = Effective spread / Quote midpoint (4) Quoted spreads represent the ex-ante expected costs of trading. Effective spreads reflect the price improvement received in a trade and represent the actual ex-post cost of liquidity. Although we present results for both quoted and effective spreads, these two measures should be viewed as alternative expressions of the same concept. Spreads are inverse measures of liquidity and higher spreads indicate poor liquidity. When an adverse selection problem is severe, market makers widen their spreads to recover the increased costs of trading with informed traders. To better gauge changes in adverse selection, we use Glosten and Harris s (1988) model to decompose the spread. In their model, the adverse selection, inventory holding, and order processing components are expressed as a linear function of transaction volume. The model can be described in the following equation: P t = c 0 Q t + c 1 Q t V t + z 0 Q t + z 1 Q t V t + ε t (5) where Q t is a trade indicator that is +1 if the transaction is buyer-initiated and 1 if seller-initiated; P t is the transaction price at time t; V t is the volume traded at time t; and ε t captures innovations in public information and specification error. In the model, the adverse selection component is Z 0 =2(z 0 + z 1 V t ). The inventory holding and order processing components are given by C 0 =2(c 0 + c 1 V t ). Employing Lee and Ready s (1991) procedure for trade classification 6, an estimate of the adverse selection component is 2( z0, i + z1, ivi) Zi = (6) 2( c0, i + c1, ivi) + 2( z0, i + z1, ivi) 6 Trades are defined as buys (sells) if the trade price is greater (less) than the bid-ask midpoint. We define the quotes as the most recent quotes that were time stamped at least five seconds before the trade. Changing this interval from 5 seconds to 0 seconds in a robustness test had no impact on our conclusions. 12

15 where Vi is the weekly average transaction volume for stock i. This measure yields the proportion of total spread that is due to adverse selection. To obtain the adverse selection component of the spread in dollar terms, we multiply these proportions by the dollar spreads. Changes in spreads are calculated as the difference between mean value of the liquidity variable over the window of interest relative to the mean value over the benchmark. For example, the abnormal spread (AS) for each firm over the announcement window is computed as: AS announcement = Announcement Spread Benchmark Spread (7) We then perform t-tests to examine if these differences are statistically different from zero Tests for Robustness This sub-section discusses our tests for robustness. First, we expand our analysis to include alternative benchmark spreads. This rules out the possibility that our results are driven by nonrepresentative benchmark windows. Thus, we compute abnormal spreads using an alternate benchmark window (-100,-93), as well as (-14,-13), with no significant impact on the empirical results. Second, we add a matched control sample to rule out the possibility that our results are driven by a trend in spreads over time. For each announcement we find a matching firm by employing the methodology of Huang and Stoll (1996) and Weston (2000). The matching criteria include (a) share price (b) market capitalization, and (c) volume to minimize the following expression: where Track X i 3 i= 1 Track X i X Track X i + X 2 Control i Control i 2 denotes the value of ith matching variable for the tracking stock, and Control X i denotes the value of the ith matching variable for the control stock. We find relatively good matches for each announcement observation, with composite matching scores of 0.10 or less. Studies adopting this type of matching procedure typically impose a maximum value of Table 2 reports the mean, standard deviation and percentile statistics of share price, market capitalization, and trading volume for all 13

16 56 tracking stock announcements; the 28 tracking stocks actually implemented; the matched-control sample of 56 firms; and the matched-control sample of 28 firms. Table 2 reveals that firms announcing tracking stocks tend to be large in terms of market capitalization (approximately $18 billion) and relatively liquid, with high trading volumes. This is not surprising given that tracking stocks are a restructuring tool for diversified conglomerates. The firms that actually implement the tracking stocks tend to be somewhat smaller. Third, we recalculate our abnormal spreads after identifying and excluding nine announcements that are contaminated by other concurrent announcements. These announcements are indicated in the notes column of table Cross-sectional Regressions Finally, we examine whether the changes in liquidity are associated with cross-sectional differences in firm characteristics. In particular, we regress abnormal spreads during the announcement and implementation windows on a dummy variable for whether the tracking stock was issued to finance an acquisition, the dollar value of sales, dollar value of total debt, the ratio of price to earnings, the ratio of market to book value, and the dollar value of assets. The regression for the announcement abnormal spread also includes a dummy variable indicating whether the tracking stock was ultimately implemented. The regression for the implementation abnormal spread also includes a variable for the size of the tracked group relative to the parent and a dummy variable for whether the tracking stock was focus improving. We follow Huson and Mackinnon and define a restructuring to be focus improving if the tracked division has a different two-digit SIC code than the general division. These data are obtained from CRSP, Compustat, and Lexis-Nexis. 5. Empirical Results Our empirical results are reported in Tables 3 through 5 and Figure 1. Table 3 presents the liquidity and volume statistics for the samples consisting of all 56 announcements, 28 implementations, and the controls for each sample. The first column indicates the statistic reported, and the second column indicates the sample. The remaining columns indicate the respective windows over which the liquidity 14

17 and volume statistics are calculated. For example, the third column reports the statistics over the first benchmark period, which begins 100 trading days prior to the announcement and terminates 93 trading days prior to the announcement; and the fourth column reports the statistics over the second benchmark period, which begins 14 days prior to the announcement and terminates 13 days prior. We report two benchmark windows because the earlier dated window is least likely to be affected by information leakage, while the later benchmark window reduces the possibility of confounding events relative to the actual announcement. Liquidity and volume measures, and our qualitative results, are not substantially different over the two benchmark windows, however, so we use the second benchmark window ( 14, 13) as the basis for our hypothesis tests. For the full sample of 56 announcements, the mean quoted spread over the second benchmark window is cents or 56 basis points; the mean effective spread is cents. Mean daily volume is about $60M in less than 1,000 trades. Similar liquidity measures are obtained for the alternate benchmark window, although the volume measures are somewhat greater, but not significantly so. This gives us confidence that our benchmark statistics are representative. The fifth column reports mean liquidity and volume during the pre-announcement window. Consistent with our second hypothesis, there are no significant changes in liquidity shortly before tracking stock announcements, suggesting that these events are, to a large degree, unanticipated. The direction of change, however, suggests that there may be some information leakage, although the impact is not statistically significant. For example, the quoted spreads increase to cents or 60 basis points. This finding is at least consistent with Elder and Westra (2000), who report positive, but insignificant, pre-announcement drift in share prices. The next column reports means over the announcement window. Our results reveal that liquidityproviders increase spreads significantly during the announcement window, suggesting increased information asymmetry and a greater dispersion in opinion about the firm s valuation. Quoted spreads increase by more one cent, to 16.81, or 9 basis points, to 65 basis points, over the benchmark window. Note that the pattern is similar over the announcement window for the subset of firms that subsequently implement the tracking stock structure mean spreads increase significantly, by more than one cent, and 15

18 the percent spread increases by about 10 basis points. The effect is also highly significant for the quoted and effective spreads, and slightly less so for the relative spread measure. Note that there are no such changes observed in the control sample, indicating that the observed changes are likely caused by the tracking stock announcements, rather than some market-wide effect. Note also that the larger p-value for the declining relative spread is not surprising, given that market makers tend to quote absolute spreads that, for example, cluster around 1/8 and 1/16 with small variations relative to the underlying share price. This tends to make the sample variance of the relative spread much greater. Thus, statistical tests applied to measures of the absolute spread are most appropriate for detecting changes in spreads. These patterns are consistent with our first hypothesis that market makers and other liquidity providers tend to widen spreads during the announcement window. They do not reveal, however, whether the wider spreads are due to greater costs of holding inventory and processing orders, or due to adverse selection. To discern the effect of adverse selection, we refer to our estimate for the proportion of the spread due to adverse selection, as reported in the fourth section of rows of Table 3. These estimates reveal that, in addition to a widening of the total spreads, the proportion representing adverse selection also increases, from 18% in the benchmark window to 24% in the announcement window. This increase is statistically significant at very low p-values, and is strong evidence that market makers widen spreads during the announcement window to safeguard against trading with those whose judgments about these firms valuations are superior. Figure 1 reinforces this point. Spreads for firms announcing tracking stocks are plotted with diamonds, and spreads for the control sample are plotted with circles. Before the announcement window, the two lines are intermingled, which is consistent with our first hypothesis. During the announcement window, there is a substantial, but transitory, widening in the spread for the tracking stock sample but not the control sample. Subsequent to the announcement, average spreads return to approximately the preannouncement range. The volume measures reported in Table 3 also rise significantly during the announcement window, in terms of both share volume and dollar volume. Both measures are more than double their 16

19 level during the benchmark period. Note it is possible for the increased volume to impact spreads, although the direction of the effect is theoretically ambiguous. For example, in the model of Easley and O Hara (1992), market makers use trading volume as a signal that an information event has occurred, and they respond by widening spreads. In contrast, Harris and Raviv (1993) suggest that shocks to volume may reflect a lack of consensus, in which case higher volume may correspond with additional limit orders on both sides of the spread, which would tend to tighten, rather than widen spreads. Lee, Mucklow and Ready (1993) find empirical support for the former relationship, and such a relationship is consistent with our finding that the adverse selection component of the total spread increases. Therefore, the volume dynamics lend further support to our interpretation of market makers reducing liquidity during the announcement window in response to a perceived increase in probability of trading with an informed trader. Returning to Table 3, the last three columns report statistics over three post-announcement windows. Immediately after the announcement, liquidity improves, with spreads declining to levels closer to that of the benchmark window, and volume declines to levels near that of the benchmark, although the effect on volume is slightly more persistent. All of the previous results are robust to the exclusion of the nine announcements with confounding events. Table 4 reports the results for the remaining 48 announcements that were not contaminated by confounding announcements. Again, spreads, volume and the adverse selection component of spreads increased significantly during this announcement window. The final column reports the spread and volume measures as the tracking stock is implemented and trades. At this time, the parent company has submitted all the regulatory filings and made the necessary accounting disclosures, breaking out the results for the general division and the tracked business group. If the net effect of these additional disclosures, relative to the other aspects of the tracking stock structure discussed previously, is to reduce informational asymmetries, then we should observe significantly lower spreads at this time. The data indicate, however, that spreads for the sample of firms that implemented tracking stock are not statistically different from those during the benchmark 17

20 period. While there is some decline in the spread, the decline is not statistically significant for either the quoted spread, the relative spread or the effective spread. In contrast, the quoted and effective spreads for the control group decline significantly. For example, the effective spreads for the control group declines from during the benchmark period to during the implementation window, and this drop is statistically significant at the 0.05 level. Such a decline for the control group is not surprising, given the documented effects (c.f., Jones 2002) of alternative trading mechanisms and reductions in tick-size that were implemented over our sample. Note that a direct comparison of the magnitude of the decline in our control sample with published sources such as Jones (2002) is not possible, because our sample is in event time rather than calendar time. However, the magnitude of the decline in our control sample is not inconsistent with Jones (2002), suggesting that our control sample is, approximately, representative of market trends. The sharp decline in spreads from the control sample relative to the tracking stock sample reinforces the conclusion that implementing tracking stock has not reduced, and may tend to increase, information asymmetry. Interestingly, the above result is not sensitive to whether the tracking stock was focus improving, in contrast to the analysis of spin-offs by Huson and Mackinnon, although our statistical tests on this item will tend to suffer from low power due to relatively few observations (only five focus improving restructurings). The most striking result is that, while the spread measures drift insignificantly lower at implementation, the proportion of the spread due to adverse selection actually increases. Moreover, the indicated increase in the adverse selection component is large in magnitude and statistically significant. During the benchmark window, the adverse selection component of the spread is 20% for the sample of firms that subsequently implemented the tracking stock. At implementation, the adverse component of the spread is 27%, which is even greater than the level during the announcement window. For the control sample, the adverse selection component is unchanged at 22% from the benchmark to the implementation windows. We conclude from the above analysis that the information asymmetries introduced by a restructuring based on tracking stocks are likely to outweigh any benefits obtained from additional 18

21 financial disclosures. Our results then reinforce the discussion in Elder and Westra (2000) and the empirical results of Billet and Vijh (2002). Markets may have interpreted announcements to issue tracking stocks as value increasing events in the short-run, but the actual implementation of tracking stock is not likely to reduce the informational asymmetries impacting diversified conglomerates and it may tend to increase the informational asymmetries. Finally, in Table 5 we report the results of the cross-sectional regressions, where the dependent variables are the changes in spreads during announcement and implementation windows, respectively, and a number of firm and event specific characteristics are used as independent variables. In both regressions, only two of the independent variables are statistically different from zero, and only marginally so. In the first regression that ratio of market to book value is positive and significant at the 10% level. In the second regression, the dollar value of sales is also significant at the 10% level. In addition the R-squared measures for both regressions are about 14% and 27% respectively. The failure of these cross-sectional to explain the variation in spreads suggests that the observed effects are not driven by a subset of firms with particular characteristics. Rather, the effects are systemic throughout the sample. 6. Conclusion Several studies have investigated the impact of tracking stocks on informational asymmetries. The usual premise is that, since the SEC requires the disclosure of additional financial statements detailing the performance of the general division as well as the tracked business group, analysts can better focus on the performance of each segment. This would tend to increase both the number of analysts following the firm, and, since analysts tend to specialize in particular industries, the accuracy of their forecasts. Both of these effects should tend to reduce informational asymmetries. There are, however, theoretical and institutional factors that may tend to counter this effect, making the net effect of the tracking stock structure on informational asymmetries ambiguous. This paper explores these issues and re-examines the impact of tracking stock on informational uncertainties by utilizing a relatively new data set and methodology. Rather than examining the behavior of equity 19

22 analysts, which has produced some conflicting results, we examine the behavior of market makers, who provide liquidity to the market by posting bid and ask prices. If tracking stocks reduce informational asymmetries, than market makers should respond by providing additional liquidity to the market. Our results, however, indicate that market makers significantly reduce liquidity as a firm announces an intention to issue a tracking stock, and a decomposition indicates a significant increase in the proportion of the total spread due to adverse selection. In the long-run, as the tracking stock structure is implemented, we find only a marginal and statistically insignificant increase in liquidity for the general division, relative to a large and significant increase in liquidity for our control sample a trend which is at least consistent with documented market-wide effects. Finally, the adverse selection component of the total spread significantly increases as a firm implements the tracking stock structure. We conclude that the actual implementation of tracking stock is not likely to significantly reduce the informational asymmetries impacting diversified conglomerates. Rather, the uncertainties induced by the tracking stock structure substantially mitigate any potential benefits associated with more detailed financial disclosure, and may even tend to increase informational asymmetries. 20

23 References Billet, Matthew T., and David C. Mauer, 2000, Diversification and the value of internal capital markets: The case of tracking stocks, Journal of Banking and Finance 24, Billet, Matthew T., and Anand M. Vijh, 2002, Long-term returns from tracking stocks, Working paper, University of Iowa. Chemmanur, Thomas J., and Imants Paeglis, 2000, Why issue tracking stock? Insights from a comparison with spin-offs and carve-outs, Working paper, Boston College. Clarke, Jonathan E. and C. Edward Fee and Shawn Thomas, 2003, Corporate Diversification and Asymmetric Information: Evidence from Stock Market Trading Characteristics, Journal of Corporate Finance, forthcoming Clayton Matthew, J., and Yiming Qian, 2002, Wealth gains from tracking stocks: Long-run performance and ex-date returns, Working paper, University of Iowa. D Souza, Julia, and John Jacob, 2000, Why firms issue target stock, Journal of Financial Economics 56, Easley, David and Maureen O Hara, 1992, Time and the process of Security Price Adjustment, Journal of Finance, 47, Elder, John, and Peter Westra, 2000, The reaction of security prices to tracking stock announcements, Journal of Economics and Finance 24, Gilson, Stuart C., Paul M. Healy, Christopher F. Noe, and Krishna G. Palepu, 2001, Analysts specialization and conglomerate stock breakups, Journal of Accounting Research 39, Glosten, Lawrence R and Lawrence E. Harris, 1988, Estimating the components of the bid-ask spread, Journal of Financial Economics 21, Glosten, Lawrence, and Paul R. Milgrom, 1985, Bid, ask, and transaction prices in a specialist market with heterogeneously informed agents, Journal of Financial Economics 14, Gorton, G. and G. Pennacchi, 1993, Security baskets and index-linked securities, Journal of Business 26, Haas, Jeffrey (1996), Directional fiduciary duties in a tracking stock equity structure: The Need for a Duty of Fairness, Michigan Law Review (June) 94: Habib, M., D. Johnsen, and N. Naik, 1997, Spinoffs and information, Journal of Financial Intermediation 6, Hadlock, C., M. Ryngaert, and S. Thomas, 2001, Corporate structure and equity offerings: are there benefits to diversification?, Journal of Business 74,