Liquidity externalities and buyout delisting activity

Transcription

1 Liquidity externalities and buyout delisting activity José-Miguel Gaspar * Laurence Lescourret ** Abstract We study the impact on industry liquidity of delistings resulting from leveraged buyout activity. Using data on U.S. LBOs during the period, we uncover evidence of negative liquidity externalities. We find that the liquidity of firms in the same industry as the LBO target drops during the month of the LBO delisting. This temporary decrease is especially strong for industry firms highly correlated with the LBO firm. It is also stronger in industries characterized by higher information asymmetry and higher information heterogeneity, and in cases in which the information disclosed by the LBO firm before it goes private is more precise. Overall, our findings are consistent with the hypothesis that financial market participants use cross-asset information arising from correlated order flow and prices in their trading decisions. Delistings induced by LBO activity bring about temporary greater information asymmetry, which results in a negative transitory impact on liquidity at the industry level. JEL classification: G14, G34 Keywords: leveraged buyouts; liquidity externalities; delisting; asymmetric information * Finance department, ESSEC Business School, Avenue Bernard Hirsch, Cergy-Pontoise, France. Tel: +(33)(0) gaspar@essec.fr. ** Finance department, ESSEC Business School, Avenue Bernard Hirsch, Cergy-Pontoise, France. Tel: +(33)(0) lescourret@essec.fr. We thank Andras Fulop, Armin Schwienbacher, and seminar participants at University of Toulouse, the first Leuven-Louvain Finance Research Workshop, and Universidade Católica Portuguesa for their helpful comments. Laurence and José-Miguel gracefully acknowledge financial support from the Europlace Institute of Finance. José-Miguel gracefully acknowledges financial support from the ESSEC Private Equity Chair. Any remaining errors are our own. 1

2 1 Introduction A recent focus of attention in Finance research has been the study of liquidity externalities. One key idea from this literature is that agents use cross-asset information (learned via observed prices and order flows) in their valuation and trading decisions. A second key idea is that liquidity attracts liquidity, over time and over trading venues, for reasons linked to changing asymmetric information and changing competition between liquidity providers. 1 Our paper attempts to contribute to this literature by formulating and testing hypotheses related to liquidity externalities in a precise setting: stock delistings arising from leverage buyout (LBO) activity. During the recent credit bubble of , LBOs activity increased significantly following particularly favorable credit market conditions. 2 In the U.S. in 2007, about 170 USD Billion of equity was removed from the public markets as a result of buyout activity, representing about 1% of total CRSP market value. In fact, in the annual value of U.S. domestic firms targeted in LBO transactions was greater than the equivalent annual value of U.S. domestic firms entering the market via IPOs. 3 We believe that LBOs are a promising setting to study questions related to liquidity. First, LBO-related delistings are relatively exogenous as far as the liquidity of remaining firms is concerned. Second, when the LBO deal is completed, trading in the LBO firm ceases completely, and any information on stock price of the LBO firm is no longer observable. This is unlike delistings related to going dark firms (that is, firms 1 Theoretical papers include, among others, Admati and Pfleiderer (1988), Pagano (1989), Caballé and Krishnan (1994), and Bernhardt and Taub (2008). For empirical papers investigating liquidity externalities, see e.g. Amihud, Mendelson and Lauterbach (1997), Domowitz, Glen and Madhavan (2002), Barclay and Hendershott (2004), Hendershott and Jones (2005), and Bessembinder, Maxwell, and Venkataraman (2006). A related area of the literature, initiated by Chordia, Roll and Subrahmanyam (2000) and Hasbrouck and Seppi (2001), documents the existence of common factors in liquidity. 2 Ljungqvist et al. (2007) and Axelson et al. (2008) document the importance of credit market conditions on Private Equity investment and LBO pricing. According to Thomson Financial data, at the peak about 274 Billion dollars of capital were raised by U.S. and European Private Equity funds in 2006, and more than 70 % of this capital was directed towards funds specializing in buyouts. 3 Source: Authors calculations using Thomson Financial/SDC and CRSP data. Equity value includes the acquisition premium and is for completed LBO deals only. CRSP data includes U.S. domestic firms listed on NYSE/AMEX and NASDAQ. IPO figures are for completed deals only. 2

3 that simply deregister with the SEC) for which trading may continue in over-thecounter markets (OTCBB or Pink Sheet). Finally, during the bubble, several market participants, including regulators, conjectured that buyout activity could be detrimental to stock market liquidity. 4 Our analysis allows us to investigate the veracity of these conjectures. We use a set of existing theoretical efforts to motivate two competing views and take these models to the data against the null hypothesis that LBO activity has no impact on the market liquidity. The liquidity deterioration hypothesis predicts a negative relationship between buyout activity and liquidity, for several reasons. First, recent papers have shown that market-makers exploit information in cross-asset order flows to learn about the fundamental price of one asset (Caballé and Krishman, 1994; Heidle, 2004; Bernhardt and Taub, 2008; Pasquariello and Vega, 2008). When a company switches to private ownership, its market information (prices and order flow) that used to convey valuable information to market-makers is no longer observed. 5 As a consequence, market-makers face greater adverse selection risk. In addition, a large public firm that goes private may adversely impact the incorporation of common information into prices of small firms. It has been shown that small-cap market-makers use large-cap quotes to update their own quotes since common information is first incorporated into large firms due to lower transaction costs (Chordia, Sarkar, and Subrahmanyam, 2008). Finally, the disappearance of one asset affects the trading strategies of speculators who observe prices and use that information to trade in other assets. If the intensity of competition among speculators weakens, this would also lead to an increase in adverse selection risk (Bernhardt and Taub, 2008). The liquidity enhancement hypothesis predicts a positive relationship between buyout activity and liquidity, for the following reasons. First, we could observe order flow migration from the LBO firm s stock to the remaining public correlated securities, increasing competition among liquidity suppliers (Hendershott and Jones, 2005). 4 The Financial Services Authority, the U.K. s stock market regulator, mentions reduction of capital market efficiency and market opacity as key risks deriving from LBO activity (FSA, 2006). See also The paradox behind the invasion of the privateers, Financial Times, February 13, In contrast, Strömberg (2007) argues that private equity activity is a net contributor to the stock market. 5 Moreover, privately-held companies cease disclosure obligations with the SEC, substantially decreasing the amount of information available to investors. 3

4 Second, dealers behavior depends on inventory position in the assigned stock, but also on inventory positions in substitutable stocks. Removing one correlated security might decrease the market-maker s total inventory risk, lowering the costs to provide liquidity, and enhancing it (Ho and Stoll, 1983; Andrade, Chang and Seasholes, 2008). Furthermore, in a situation where funding capital is scarce, an exit via a buyout increases the risk-bearing capacity of market-makers in the remaining listed stocks. Market-making capital is now allocated across fewer stocks, increasing liquidity (Brunnermeier and Pedersen, 2007). To distinguish between these two hypotheses and investigate the existence, direction, and magnitude of the impact of buyout activity on stock market liquidity, we collect data on buyouts for the U.S. market during the period. We use univariate tests and regression analysis to test for significant changes in liquidity following the delisting of the LBO firm (moment in which its price movements stops being observed and disclosure obligations cease). Two important remarks are in order. First, throughout the paper we define the group of firms affected by the LBO delisting as firms belonging to the same industry as the LBO firm. 6 Firms in the same industry are more likely to be affected by the same underlying flow of information and of news than firms in unrelated businesses. Information production agents like analysts in research departments are typically organized by industry sector and many asset managers follow sector strategies. Furthermore, market-making tends to be organized around industry groups (Anand, Chakravarty and Chuwonganant, 2007), and the NYSE tends to allocate new stocks to individual specialists that already trade related securities (Corwin, 2004). All these reasons lead us to believe that the industry is the most natural affected group by the buyout and subsequent delisting. Second, our main liquidity measure is the illiquidity ratio proposed by Amihud (2002), mostly because it correlates highly with intra-daily liquidity measures and it would be impractical to calculate the latter for the CRSP universe across 23 years. In our implementation, we detrend the Amihud measure to adjust for the significant secular time-series trend in liquidity (Jones, 2002). We also adjust for contemporary entry and exit of firms in the market via IPOs, mergers, and delistings in our regression specifications. 6 We use Eugene Fama and Kenneth French s industry definitions available at 4

5 We find that the liquidity of the affected industry is statistically significantly lower during the month of the delisting of the LBO firm. The implied increase is equivalent to a jump of about 0.26 standard deviations, a small but non-negligeable change. The decrease in liquidity is limited to the delisting month and does not carry on to subsequent months. This result indicates that LBO activity has a negative transitory impact on liquidity at the industry level. We uncover evidence that these results are due to a transitory increase in adverse selection. First, our findings are stronger for non-lbo firms that have a high return correlation with the LBO firm prior to the deal. This indicates that common information was prevalent among these firms, affecting market agents relatively more once the LBO firms delist. Second, the decrease in liquidity is stronger for LBOs for which the information disclosed by the LBO firm prior to going private was larger and more precise (in the sense that the LBO firm was itself large relative to its industry, and the mean forecast error of analyst earnings forecasts was low). This is also consistent with the theoretical predictions made by the liquidity deterioration hypothesis. Finally, we find that the decrease in liquidity is stronger for industries characterized by a higher information asymmetry and a higher uncertainty (higher dispersion of analyst forecasts). These are the industries for which precisely the loss in information production from the LBO firm is higher. We also investigate alternative explanations from our results. We report similar findings using an alternative low-frequency liquidity measure, the frequency of zero daily returns in the month (Lesmond, Ogden and Trzcinka, 1999). We check whether our results are an artifact of a short time span between the announcement of the LBO and its subsequent delisting. We find that around announcement there is no significant change in industry liquidity, and that our results go through once we control for announcement as well as delisting dates. We examine whether our results are related to temporary price-pressure due to demand from investors rebalancing their portfolios around the delisting. To investigate this issue, we use the setting in which finding such pressure would be most natural: firms belonging to the S&P500. Our results are robust to the exclusion of S&P500 firms from the sample, whether they are or not LBO targets. Finally, we replicate our analysis for delistings due to non-respect of the stock exchange listing requirements, voluntary delistings, and bankruptcies. We find no impact on industry-wide liquidity, reflecting the fact that firms involved in these 5

6 delistings tend to be small, illiquid, and probably convey less information to market participants. Our paper makes two main contributions. First, we contribute to the liquidity externalities literature by using LBO delistings as a natural setting to test hypotheses on the impact of removing one asset from the market. Our evidence is consistent with the story that an important part of these externalities are related to flows of information across assets. Second, to our knowledge this is the first paper to examine the impact of buyout activity on market liquidity, thereby answering a question relevant to practitioners and regulators alike. We find that buyouts have a detrimental impact on liquidity, although the effect is transitory. Hence although LBO activity impairs liquidity, with negative consequences for the cost of capital of corporations, the impact of this activity seems moderate. There is an increasing body of literature which examines the impact of corporate finance events and firm characteristics on stock liquidity. Harris, Panchapagesan, and Werner (2007) investigate go dark delistings and find a drop in market quality around the delisting date. Mortal and Lipson (2007) document the permanent liquidity changes associated with mergers and acquisitions. However, both of these papers focus on liquidity changes of the affected firm and do not analyze any potential spillover effect on the liquidity of other firms. Bortolotti et al. (2007) show that privatization IPOs positively affect stock market liquidity. Levine and Schmulkler (2006) analyze the impact of internationalization on domestic liquidity and uncover evidence of order flow migration and liquidity spillovers between markets. Our paper contributes to this literature by documenting liquidity externalities arising from delisting decisions resulting from LBO deals. The remainder of the paper is articulated as follows. Section 2 lays out our testable hypotheses. Section 3 describes the sample construction and the variables we use. Section 4 analyzes the impact of LBO delisting on the liquidity of non-lbo firms. Section 5 provides accessory evidence to support our main results, by looking at which types of LBO firms and which industries the effect is more prevalent. Section 6 investigates whether our results are specific to the liquidity measure we use, if they are dependent on announcement effects, if they are the result of temporary price-pressures, and whether they extend to non-lbo delistings. A brief conclusion follows. 6

7 2 Main hypotheses To our knowledge, there exists no theoretical paper dealing directly with the spillover effects of removing one asset from the market on the price formation of other assets. However, we can use existing theories developed by the microstructure literature to build hypotheses on how the delisting of LBO firms from the market may affect the liquidity of other public firms. 2.1 The liquidity deterioration hypothesis Inter-asset competition between market-makers Hagerty (1991) analyzes the equilibrium bid-ask spreads in markets with multiple assets, where in each market, there is one monopolist specialist. When assets are more correlated, specialists in charge of trading related securities emerge as close competitors. Then, as the number of highly correlated assets decreases, the market power of the specialist increases and bid-ask spreads enlarge Cross-asset information flows When a firm goes private in an LBO transaction, its stock price is no longer observable by market participants. Information dissemination through prices ceases, which may affect the liquidity of related stocks in two ways. First, the delisting increases adverse-selection costs faced by market-makers who used to extract information from LBO firm s prices and order flows to price other assets. In recent multi-asset extensions of Kyle (1985), Caballé and Krishnan (1994), 7 Note that Hagerty s (1991) prediction is dependent on its relatively simple competition structure. First, in practice market-makers do not make market in just one security: for instance, a NYSE individual specialist may simultaneously handle several stocks in the same industry (Anand, Chakravarty and Chuwonganant, 2007). Second, Gehrig and Jackson (1998) suggest, in a setting where each specialist can trade multiple assets, that the impact of removing one asset on specialist competition depends on factors such as the return correlation structure, the initial distribution of portfolio holdings, and whether specialist portfolios are diversified across industries. 7

8 Bernhardt and Taub (2008), and Pasquariello and Vega (2008) show that marketmakers exploit the information in cross-asset order flows to learn about the fundamental price of one asset. This intuition also applies to the case of a sequential trade model: Heidle (2004) shows that the market-maker will not only update his quoted prices after trades in the assigned security, but also after trades in the other stocks. A public firm that goes private thus leads to the disappearance of a public order flow conveying valuable information to liquidity providers, who thus face more adverse selection. 8 Furthermore, Chordia, Sarkar, and Subrahmanyam (2008) show that small-cap market-makers use quotes of the large-cap stocks to update their own quotes since common information is first incorporated into large firms (due to lower transaction costs). 9 A large public firm that goes private thus may adversely impact the incorporation of common information into prices of small firms. Second, the delisting affects the behavior of investors engaged in speculation using cross-asset information. Bernhardt and Taub (2008) show that speculators use their private information, the correlation structure of asset fundamentals, and the observability of stock prices, to trade strategically across assets. The disappearance of one asset affects the trading strategies of speculators in other related assets, causing them to trade less aggressively. As a consequence, speculators trades reveal less information to market makers. This latter effect leads to an increase in the overall adverse selection risk, and thus to a deterioration in the liquidity of the market in related stocks. In summary, these models suggest that LBO delistings should exacerbate asymmetric information costs, reduce competition among speculators, and decrease the liquidity of related assets, and in particular, the liquidity of firms within the same industry. 8 An empirical observation supporting this logic is that the NYSE tends to allocate stocks to individual specialists that already trade other related securities (Corwin, 2004). The underlying rationale is that specialists have larger information precision that comes from handling a portfolio of highly correlated stocks. 9 In a related paper, Chan (1993) shows that market-makers condition quotes on lagged prices of other stocks, causing stock returns to be positively cross-autocorrelated. 8

9 2.2 The liquidity enhancement hypothesis Order flow migration A counterpoint to the afore-mentioned predictions comes first from positive liquidity externalities arising from order flow migration. Hendershott and Jones (2005) find that order flow is diverted away from the Island electronic communications network, and into other trading venues, when the former no longer disclosed its limit order book on particularly active ETFs ( go dark ). 10 Applied to our setting, this logic would imply that we could observe order flow migration from the stock going private to the remaining public correlated securities. Liquidity in the remaining securities could increase because of gains in order flow and because of increased competition among liquidity providers due to the relocation of investors who used to trade in the LBO stock Inventory risk Standard microstructure models relate liquidity and market-marker s risk aversion to inventory position in risky assets. Dealers quoting behavior depends on the inventory position in the assigned stock, corrected by any reinforcing effect arising from inventory positions in substitutable stocks (Ho and Stoll, 1983; Andrade, Chang and Seasholes, 2008). Thus correlated risks arise from correlated order flow across assets and from inventory positions in all other securities of the total portfolio handled by the specialist. In some cases, removing one correlated security decreases the total inventory risk, lowering the costs to provide liquidity for the specialist, and enhancing market liquidity. More recent literature has also investigated the link between the market liquidity and market-makers funding constraints. Brunnermeier and Pedersen (2007) show that liquidity provision depends crucially on capital risk faced by market-makers. In particular, when market-makers are close to their capital constraints, they reduce the amount of liquidity provision in risky and illiquid securities, which use more capital, 10 In an international setting, Domowitz, Glen and Madhavan (1998) and Levine and Schmukler (2006) find evidence of order flow migration in domestic markets arising from cross-listings abroad. 9

10 leading to liquidity dry-ups. Comerton-Forde et al. (2008) find, on the NYSE, that the liquidity of risky stocks is more sensitive to shocks in the specialist s inventory, consistent with limited risk-bearing capacity. These papers would therefore suggest that when a firm goes private, market-makers are not any more concerned about financing their inventory position in the delisted stock. This should increase the riskbearing capacity of market-makers in the remaining listed stocks. We should therefore observe a positive impact of the LBO activity on the overall liquidity. 3 Data 3.1 Sample construction Our source for buyout data is the Thomson Financial/SDC Platinum database, from which we select completed LBO deals for publicly listed U.S. companies announced during the period 1985 to SDC contains information on deal characteristics (e.g., deal size, industry, exchange where the LBO target is trading, premium paid) and on LBO announcement dates and effective dates (the date the deal is considered complete by SDC). The initial sample contains about 850 deals. In parallel, we collect information on prices, returns, market capitalization, and volume from the entire daily CRSP database over the same period. We use CRSP to obtain the variables mentioned for our LBO sample. We also use it to create the benchmark group of non-lbo affected firms necessary to our analysis. We restrict the LBO sample to target firms that are present in the CRSP database. The CRSP delisting codes allow us to pinpoint the exact date of delisting of the LBO firm. The delisting codes also allow us to separate LBO deals from other delistings not associated with financial sponsors (e.g. going dark ). We drop deals with (i) a missing CRSP delisting date, (ii) deals whose delisting date is prior to the announcement date reported in SDC, and (iii) deals with a delisting date more than one year after the effective date of the deal reported in SDC. We also drop some firms that continue to trade after the deal under a new Permno identifier. Finally, we retain only LBO targets whose common stock has a CRSP share code equal to 10 or 11. The final sample contains 517 LBO deals. 10

11 The non-lbo sample consists of all remaining firms in CRSP that are not a target of an LBO transaction. 11 The following filters are applied to the non-lbo sample (e.g. Amihud, 2002). We require a firm to have: (i) at least 200 days of return and volume in a given year; (ii) a non-missing market capitalization at the beginning of the previous year; and (iii) a stock price above $5 and below $999. We drop all stocks with a CRSP share code different from 10 or 11. Finally, we winsorize the sample at 1% and 99% percentiles when calculating our monthly liquidity measures to reduce the impact of outliers (see below). There are roughly 1.9 million monthly observations in the (firm-level) non-lbo sample. 3.2 LBO activity The use of LBO delisting in a setting to test our hypotheses depends on the assumption that the magnitude of LBO activity is sizeable enough to affect stock market liquidity. Figure 1 presents the total pre-event market value of sample firms targeted in LBO transactions, showing clearly two periods of hot markets over past decades with two pronounced peaks, one in 1988 and another in At peak times, LBO activity represents a sizeable portion of M&A activity, e.g. 23% in 2006 and 18% in late 1980s. 12 Another important characteristic of LBO activity is that it is industry clustered (e.g. Strömberg, 2007). Table 1 reports the industries that are most active by number of deals and by deal value for different sub-periods of our sample. Among the most active ones during the period are retail, business services, and healthcare businesses. Telecom and utilities have seen more deals since the mid-90s, while others like consumer goods and machinery have seen their incidence of deals decrease since the 1980s. How does LBO activity size up to other corporate events? Table 2 answers to this question by presenting data on the relative magnitude of different corporate events for 11 Firms targeted in LBO transactions are kept in the sample but only during a time horizon up to 4 months before the announcement date of the corresponding LBO transaction. 12 We measure pre-event market value as the average market capitalization of LBO firms during a period of 12 months ending 3 months before the announcement date. This approach (of using pre-event market value as a measure of activity) allows us to (i) compare directly the magnitude of LBO activity with the aggregate stock market size and (ii) compare LBO and M&A activity without introducing biases due to different acquisition premiums and different leverage structures across the two types of deals. 11

12 the sample period: IPOs, M&A, repurchases and non-m&a related delistings. 13 Panel A shows that in our sample an average of 20 public companies per year undergo an LBO, representing an average (median) pre-deal market value of 15.1 (4.8) Bn dollars. LBO firms represent a fairly small proportion of CRSP market value (0.20%) and of trading activity (0.25%) before announcement. 14 Using this metric, the size of LBO activity is roughly on par with that of delistings (0.13% of CRSP market value), but below that of IPO s (1.32%), repurchases (2.14%) or M&A (4.6%). LBO activity constitutes 6% of M&A deals in number and 4% in value. These numbers ignore the fact that LBO activity is clustered and hence affects some industries much more than others. Panel B and C of Table 2 repeat the analysis for the top 10 industries in LBO activity by value and number of deals. The economic importance of LBO activity is now much more visible. LBOs represent 1.55% of CRSP market value, which is an order of magnitude smaller than that of M&A activity (5.81%) but larger than that of repurchases (1.11%) and delistings (0.17%), and roughly on par with that of IPO s (1.74%). In fact, in these industries LBOs size up to 40.6% of all M&A deals in number and 77.4% in value. We conclude that LBO activity, at least in industries where it is more prevalent, has an economic significance on a par with most other corporate actions studied in the finance literature. 3.3 Illiquidity measure To test our hypotheses concerning the impact of buyout activity on stock market liquidity, we employ the monthly time series of Amihud s (2002) illiquidity ratio, the monthly average of the daily ratio between a stock s absolute return and its dollar volume, averaged over all days in the month with non-zero volume: ILLIQ i,m = 1 Days i,m Daysi,m d = 1 R i,m,d DV ol i,m, d where Days i,m is the number of valid observation days in month m, and R i,m,d and DVol i,m,d are, respectively, the daily return and dollar volume of stock i on day d of 13 See caption of Table 2 for details on the data sources and sample construction. 14 As mentioned before, market capitalization and trading volume relating to LBOs are averages measured over a period of one year ending 3 months before the announcement date of the LBO deal. 12

13 month m. 15 In our empirical implementation, we rescale the ratio by a factor of We use a low-frequency measure of liquidity because it is impractical to use intra-daily data to calculate liquidity for the entire CRSP universe of non-lbo firms across 23 years. The illiquidity ratio ILLIQ reflects the impact of order flow on price, that is, the price response associated with one dollar of trading volume. The order flow of a less liquid stock will have a larger price impact. The magnitude of the price impact should be a positive function of the perceived amount of informed trading on a stock (Kyle, 1985), although illiquidity will also undoubtedly reflect the inventory costs associated with supplying immediacy to a given order size (Garman, 1976; Amihud and Mendelson, 1980). The advantage of Amihud s measure is that it is highly correlated with existing microstructure measures of liquidity costs (Hasbrouck, 2008; Korajczyk and Sadka, 2008; Holden, 2009), but can be directly calculated from daily stock market data. The Amihud measure performs particularly well when compared to some high-frequency measures of price-impact. 16 The raw illiquidity data needs to be adjusted for time trends and seasonal regularities. It has been well documented that liquidity has remarkably improved across the U.S. market during the past decades (e.g. Jones, 2002). Figure 2 illustrates this phenomenon by showing that the illiquidity ratio has decreased substantially over our sample period, as a consequence of electronization, stock market reforms (e.g. reduction in tick size), and expanded investor participation. We use two detrending methods. The first method is a standard 12-month moving-average model with a monthly seasonal component (e.g. Murteira et al., 1993). The second method is a regression model based on Hameed, Kang and Viswanathan (2007), in which detrended illiquidity is the residual of a firm-level regression using different time trends across the different tick-size periods, as well as time dummy variables for significant events affecting liquidity. We use both methods to make sure 15 To account for the overstatement of the NASDAQ volume due to interdealer trading, we divide NASDAQ volume by 2 (see, e.g., Atkins and Dyl, 1997). 16 Goyenko, Holden, and Trzcinka (2008) run horse-races between CRSP-based liquidity to Kyle s priceimpact benchmarks calculated from the NYSE s Trade and Quote (TAQ) from 1993 to The Amihud illiquidity measure has the highest correlation with TAQ-price-impact benchmarks and its performance is statistically significantly higher than any other measures. 13

14 that our results are not sensitive to any particular detrending method employed. The Appendix contains details concerning the two detrending models. 3.4 Defining the affected non-lbo group Throughout the paper, the group of affected firms will be defined as the set of firms that belong to LBO firm s industry group. We employ the Fama and French (1997) industry definitions to create 48 industry groups. If removing a firm affects information production (in particular, that relevant to the entire industry), firms within the same industry should be the most affected by the LBO firm s delisting from the public market. Furthermore, specialists and/or market-makers are often organized on a sector basis. Firms belonging to the same industry thus emerge as the natural group of firms affected by the LBO delisting. For each industry group J k corresponding to each LBO deal k, we compute aggregate monthly illiquidity measures by industry using detrended illiquidity, or ILLIQd. In particular, we calculate monthly industry illiquidity in two ways: valueweighted (using each firm s market value as weight), and equally-weighted. Each aggregated industry illiquidity measure is built without including the illiquidity of the LBO firm. Given that size is a proxy for liquidity, we look at both value- and equal-weighting groups to make sure that findings are not particular to one set of firms. In summary, we use 4 measures of aggregated illiquidity in our analysis, that depend on whether we use the moving-average or the regression method to detrend (ILLIQd MA, ILLIQd REG), and whether we use value-weighted or equally-weighted industry illiquidity. 3.5 Summary statistics Panel A of Table 3 presents the summary statistics for the 517 LBO deals in the sample. The average (median) LBO has a market capitalization of 733 (121) million dollars, representing 0.65% (0.1%) of the industry s market value. The industry groups corresponding to each LBO are composed of 124 (84) firms on average (in the median). 14

15 To better understand the relative characteristics of LBO firms, the table also presents the percentile of LBOs in terms of size, liquidity and trading activity within industry. Compared with their peers, LBO firms are below the median in terms of size (the median LBO is in percentile 36%), are above median in terms of illiquidity (percentile 69%) and are among the less traded firms (percentile 36%). Firms listed on the NYSE and AMEX constitute 46% of the sample. The average time between the deal announcement and the delisting of the LBO is 4.84 months. The structure of our regression specification (see next section) requires that we use a window going from 6 months before to 6 months after the delisting date of the LBO (hence the number of observations matches roughly 13 months times 517 deals). Panel B shows statistics for the regressions sample having the pair deal-month as unit of observation. Each industry group considered has an average market value of about 152 million dollars. Average equal-weighted industry illiquidity is higher that valueweighted illiquidity by a factor of 10, reflecting the fact that large firms are considerably more liquid than small firms. The table also shows statistics for the detrended industry illiquidity. Although the reported average values of detrended illiquidity have a negative sign, t-tests (not shown) report that detrended industry illiquidity is not statistically different from zero. Along with statistics for industry volatility and volume, the table also shows the importance of other corporate events in the months surrounding the LBO delisting. IPO volume, for example, represents on average 0.15% of the industry s market cap during that period, while M&A activity represents 0.45%. In our regressions we control for this contemporaneous activity. 4 Results 4.1 Univariate results This subsection presents preliminary evidence concerning our hypotheses by looking at changes in industry liquidity around the delisting date of the LBO. Table 4 presents the level of detrended industry illiquidity for the month immediately before the delisting and for the month during which the delisting occurs. The results show that both average and median industry illiquidity increase in the delisting month. The average value-weighted (equal-weighted) detrended illiquidity increases 0.4% to 0.6% (2.0% to 2.5%), depending on the detrending method. The 15

16 median value-weighted (equal-weighted) illiquidity increases 0.1% to 0.2% across the 2 different detrended measures (1.3% to 2.7%). Statistically speaking, the results are stronger for ILLIQd MA, the moving-average adjusted illiquidity, for which the p- values of the difference are below or close to the 1% confidence level. For ILLIQd REG, the regression-adjusted illiquidity, the p-values vary between 3% and 13%, with 3 out of 4 coefficients being statistically significant at the 10% confidence level or better. These changes are economically significant. Taking value-weighted ILLIQd MA as an example, the implied mean change of 0.6% corresponds to an increase of a magnitude of 26% of illiquidity s monthly standard deviation. 17 As a robustness check, we directly look at the distribution of changes in illiquidity, that is, the distribution of the differences in illiquidity from the month before delisting to the month of the delisting. Table 5 confirms the direction and magnitude of earlier results. The coefficients of the average change in value-weighted (equal-weighted) detrended illiquidity are 0.4% to 0.6% (2.0% to 2.5%), depending on the detrending method. The median change in value-weighted illiquidity is 0.1%, and 0.5% to 1.5% for equal-weighted illiquidity. All coefficients are statistically significant at 5% or lower. We conclude that there is evidence of a decrease in liquidity for industry firms around the month of the LBO delisting. 4.2 Multivariate results To confirm the robustness of our findings to other explanatory factors and assess the persistence of any changes in industry illiquidity, we run the following regression model: (1) where the dependent variable is one of our 4 measures of detrended industry illiquidity and D is a set of indicator variables that take the value 1 in each of the months τ=-6,- 5,..., -2, 0, 1,..., 6 relative to m DLST, the month during which the LBO firm k is delisted. In other words, we use a window starting 6 months before the LBO delisting up to 6 months after to gauge the persistence of any changes in liquidity. Note that the 17 Table 3 shows a value-weighted ILLIQd MA standard deviation of Hence the change represents 0.006/0.023=26% of the standard deviation. Similar results apply to equal-weighted ILLIQd MA, for which we find a change of 0.027/0.155=18% of standard deviation. 16

17 dummy related to the month immediately preceding the delisting month (m DLST-1 ) is omitted and serves as reference point to interpret the regression coefficients of the month indicator variables. 18 We use two sets of control variables in our regressions. The matrix X controls for characteristics of the LBO firm k, namely: pre-event market value; the LBO firm s pre-event within-industry rank in terms of market value, trading volume, and illiquidity; the exchange in which the LBO firm was traded; and the time elapsed from the announcement date to the delisting date. All pre-event variables are measured as averages over a period of 12 months ending 3 months before the LBO announcement date to avoid contamination by deal rumours. The matrix W contains two groups of variables. The first group respects to industry characteristics, such as the industry s market value, trading volume, and volatility. 19,20 The second group represents the contemporaneous entry and exit (in % of market value) of firms in the industry occurring via IPO s, mergers, and delistings. Note that all these variables are time-varying over the window of the deal. They are meant to control for possible contemporaneous events that might affect liquidity in the industry group during the LBO delisting period. Finally, we use year dummies and industry dummies to control for unobserved heterogeneity across time periods and across industries. 21 Table 6 presents estimates of the regression model. Each column shows the results obtained for each of the dependent variables, equivalent to the 4 different measures of industry illiquidity. We report two specifications: the first uses a partial set of control variables (equivalent to matrix X) and the second uses the complete set of control variables (equivalent to matrices X and W) described by equation (1). 18 This amounts to a difference-in-differences approach. 19 The aggregation method used for these variables (i.e., value- or equal-weighting) matches that of the left-hand side variable employed when estimating equation (1). For example, when using value-weighted ILLIQd, the industry variables in the right-hand side are value-weighted as well. 20 It is well-known that return volatility affect price impact by raising adverse selection risks and inventory risks. Having the aggregate return volatility as a control in our regressions ensures that our results are not due to changes in industry risk. As an unreported robustness check, we run our regression model using volatility as a dependent variable. We find no evidence of a significant change in industry volatility around the LBO delisting date. 21 Please see the caption of Table 6 for more details on variable construction. 17

18 The main result from Table 6 confirms the negative and statistically significant effect of LBOs delisting on industry liquidity during the month of the delisting of the LBO firm. Across all 4 measures of industry illiquidity and all specifications, the industry illiquidity observed during the delisting month (Month 0) is significantly higher compared to that of the previous month (the t-statistics vary between 1.96 and 2.47). The parameter estimates of the illiquidity changes during the delisting month are in accordance with those reported in Tables 4 and 5. The table shows that this effect is not persistent: none of the other month dummies of the observation window are consistently statistically significant across all measures and/or specifications. It is worth noticing that for equal-weighted illiquidity measures, the trading activity of the industry (industry NVOL) is negatively related to the industry illiquidity, which is consistent with previous studies (e.g., Chordia, Roll, and Subrahmanyam, 2001). However, other control variables do not seem to have any significant explanatory power. Most are not statistically significant, and the regression R-squared (admittedly low in the case of ILLIQd MA) does not change much whether we use the partial or the total set of control variables. In summary, our results are consistent with the liquidity deterioration hypothesis that predicts an increase in illiquidity for firms affected by the LBO delisting. 5 Information asymmetry and impact on liquidity This section investigates additional testable implications arising from the liquidity deterioration hypothesis, using the cross-sectional heterogeneity in firms, LBO firm characteristics and industry. 5.1 Results conditioning on asset return correlation Hagerty (1991) predicts that, as substitutability between assets increases, the higher the increase in market power of the remaining specialists when one asset is removed from the market. Multi-asset asymmetric information models (e.g., Bernhardt and Taub, 2008) support the logic that, as correlation between asset values increases, the greater is the value of cross-asset private information for liquidity providers, and the larger is the increase in adverse selection risk when one correlated signal disappears. 18

19 Therefore, we expect that the delisting of LBO firm has a stronger illiquidity effect in stocks that are relatively more correlated with the LBO firm. To test this prediction we construct measures of return correlation between the LBO firm and every firm in the same industry for a period of one year ending 3 months before the announcement date of the LBO deal. We then divide them into two polar sub-samples: a High correlation group, consisting of industry firms that are in the top third in terms of correlation with the LBO, and a Low correlation group consisting of firms in the bottom third. Table 7 reports the results, using the same regression specification and all controls as those described in Table 6. For clarity we report only the coefficient of interest corresponding to the indicator variable related to the delisting month (Month 0). The month before the delisting (Month -1) is the month of reference. 22 Results show that the delisting of an LBO has a stronger negative impact on the group of more correlated firms (the t-statistics vary from 2.57 to 1.84 across all 4 illiquidity measures). In contrast, the effect on the group of less correlated firms is much weaker since none of the coefficients is statistically significant. 5.2 Results conditioning on the characteristics of LBO firms This subsection is meant to characterize more precisely LBO firms whose impact on industry illiquidity at delisting time is expected to be stronger. First, multi-asset asymmetric information models (e.g. Pasquariello and Vega, 2008) predict that a firm that discloses more precise information should have a greater impact on industry illiquidity when going private, because its stock s price conveys a relatively more informative signal. Second, Chordia, Sarkar and Subrahmanyam (2008) suggest that the delisting of large LBO should affect more strongly the industry illiquidity since market-makers on small-cap stocks lose valuable information that used to be impounded first into prices of large LBO firms. To proxy for the quality of the signal disclosed by a firm, we use errors in analysts 'earnings forecasts available through the Institutional Brokers Estimate System (IBES). Panel A of Table 8 reports results based on two sub-samples according to whether the mean forecast error of the LBO firm is above (versus below) the median 22 All other coefficients are not significantly different from zero. 19

20 mean forecast error within the industry prior to the LBO announcement. 23 The results show that LBOs disclosing a more precise information (Low mean forecast error) have a stronger impact on illiquidity: the coefficients are positive and statistically significant at 10% or better across all illiquidity measures. In contrast, the sub-sample of LBOs characterized by a High mean forecast error has positive but non statistically significant coefficients, except for the specification using the value-weighted ILLIQd MA as dependent variable. We can thus conclude that the more precise the information produced by the LBO firm is, the stronger the impact on the industry illiquidity is when it goes private. Panel B of Table 8 tests the prediction of Chordia, Sarkar and Subrahmanyam (2008), by partitioning the LBO sample into two sub-samples: the large versus the small LBO firms according to whether the market cap of the LBO firm is above (versus below) the median market cap of the LBO sample. Results confirm that the delisting of large LBOs has a more severe effect on the industry illiquidity. Once again, the coefficients based on the sub-sample of large LBOs are positive and statistically significant across all illiquidity measures, as opposed to those based on small LBOs. These evidence therefore suggest that the relative accuracy and importance of the information disclosed by the LBO firm matter for industry liquidity at delisting time, supporting the hypothesis of the existence of cross-information linkages between assets. 5.3 Results conditioning on the industry characteristics Bernhardt and Taub (2008) show that the division of uncertainty across assets matters for speculators trading strategies, causing them to compete more or less aggressively, thereby affecting liquidity. We thus conjecture that the impact of LBO delistings could depend on the type of informational environment that characterizes a firm industry. We investigate two different aspects of the informational environment. The first one is the level of asymmetric information between industry firms and investors, which we proxy by the mean error in analysts 'earnings forecasts available through IBES. The second feature is the heterogeneity in the way agents obtain information, or heterogeneity on how they process the same type of information. We proxy the latter 23 See caption of Table 8 for more details on the mean forecast error variable. 20

21 by the standard deviation of analyst forecasts (see, e.g., Krishnaswami and Subramaniam, 1999). 24 Panel A of Table 9 reports the results based on two sub-samples of non-lbo firms formed according to whether the industry mean forecast errors is above or below the median. Results show that the delisting has a stronger negative impact on the liquidity of industries characterized by a high degree of information asymmetry. Once again, only coefficients of industry with High mean forecast error are positive and statistically significant across all illiquidity measures (the t-statistics vary from 1.68 to 2.45), unlike those resulting from dependent variables constructed on industries with a Low mean forecast error. In Panel B of Table 9 we divide industries into two groups respectively above and below the median industries analysts dispersion of opinion. Results show that industry illiquidity of industries is more severely impacted by the delisting of the LBO firm in cases of industry characterized by a greater information heterogeneity. Indeed the coefficients related to the sub-sample characterized by a High dispersion of opinion (above median) are positive and statistically significant at 5% or better across all illiquidity measures. In contrast, no coefficients resulting from the regression using the other sub-sample (Low dispersion of opinion) are statistically significant. Therefore the negative impact of LBO activity on industry illiquidity during the delisting month is stronger for industries with larger levels of information asymmetry (greater prediction errors and greater disagreement among analysts). This is consistent with the notion that it is in these industries that removing an asset will be particularly damaging in terms of information production. 6 Robustness checks This section presents a set of checks meant to gauge the robustness of our results. We look at four possibilities. First, we investigate whether our results are particular to the measure of liquidity that we use. Second, we check whether the observed increase in illiquidity is due to deals characterized by a short interval between the announcement of the LBO and the delisting of the target firm (as this could mean that we would be picking up the effect of the announcement news). Third, we examine the 24 More details on the construction of the information uncertainty variables are given in Table 9. 21