The microstructure of Australian takeover announcements

Transcription

1 517247AUM / Australian Journal of ManagementBugeja et al. research-article2014 Article The microstructure of Australian takeover announcements Australian Journal of Management 201X, Vol XX(X) 1 28 The Author(s) 2014 Reprints and permissions: sagepub.co.uk/journalspermissions.nav DOI: / aum.sagepub.com Martin Bugeja Accounting Discipline Group, University of Technology Sydney, Australia Vinay Patel Finance Discipline Group, University of Technology Sydney, Australia Terry Walter Discipline of Finance, The University of Sydney, Australia; Member Services, Sirca Limited, Australia Abstract Using several microstructure variables, this study provides an intra-day examination of aggressive trading around Australian takeover announcements. We conduct this analysis for both target and bidding firms. We examine aggressive trading (i.e. by those who initiate the trade) using the abnormal behaviour of returns, trading volume, volatility and time-weighted spreads and depth. In addition, we develop a novel profit/loss measure (PLM) based on trade initiation and provide new evidence using the recently developed volume-synchronised probability of informed trading (VPIN) metric. In a univariate setting, these measures provide evidence of increased aggressive trading in Australian target firms. Further, after controlling for several microstructure variables, multivariate analysis reveals the presence of abnormally elevated time-weighted spreads prior to the announcement date for target firms. We show that VPIN is significantly elevated for target firms, especially in the four days prior to the takeover announcement. Keywords Aggressive trading, bidder, microstructure, takeover announcement, target, trade initiators, VPIN Corresponding author: Vinay Patel, Finance Discipline Group, Faculty of Business, University of Technology Sydney, P.O. Box 123, Broadway, Sydney, NSW 2007, Australia. vinay.patel@uts.edu.au Final transcript accepted 25 November 2013 by Carole Comerton-Forde (AE Finance).

2 2 Australian Journal of Management XX(X) 1. Introduction A number of US studies document increased trading, attributed to informed investors, in target firms around takeover announcements. This evidence uses abnormal returns and volume (Sanders and Zdanowicz, 1992; Eyssell and Arshadi, 1993), and several microstructure measures including bid-ask spreads, depth and other liquidity measures (Conrad and Niden, 1992; Foster and Vishwanathan, 1995). However, in Australia pre-bid trading in target firms has primarily been examined using abnormal returns and volume (Casey et al., 1987; Murray, 1994). Additionally, prior research has largely examined pre-bid trading in target firms using daily, rather than intradaily, measures. 1 Furthermore, earlier studies provide only limited analysis of whether informationally-motivated trading exists for bidding firms. 2 As a result, our study attempts to answer the following question using intra-day microstructure measures: is there evidence of increased levels of aggressive trading in both Australian target and bidding firms? This study addresses this question using three novel approaches. Firstly, we provide an intra-day analysis of abnormal returns, volume, volatility, time-weighted spreads and depth measures, allowing for more accurate inferences regarding the anticipation, arrival and dissemination of information into the marketplace (Jennings, 1994). Specifically, we examine these measures at two-hourly intervals from two weeks before until two weeks after the takeover announcement. Secondly, we develop a profit-and-loss measure (PLM) based on trade initiation to determine whether aggressive (liquidity demanding) investors profit from the takeover announcement. Lastly, we use a new measure, which captures the toxicity in order flow from aggressive traders, the volume-synchronised probability of informed trading (VPIN) developed by Easley et al. (2012). Whether increased levels of aggressive trading takes place prior to takeover announcements is important for our understanding of the price discovery process and the informational efficiency of the market. In addition, the sample of takeovers we examine have a total market value of approximately $200 billion (or approximately 14% of the total market capitalisation of ASX-listed stocks), with each deal worth on average approximately $1 billion. The economic significance of these takeovers has important implications for bidding firms to gauge whether they should continue with or revise a takeover bid, for hedge funds and other investors in assessing the likely outcome of their trading strategies and for the Australian Securities Exchange (ASX) and other regulatory bodies monitoring aggressive trading in the current marketplace. 3 We examine a sample of 195 target firms and 137 bidding firms during a 13-year period between 1997 and Our results document significant positive levels of returns, volume, volatility, bidask spreads and reduced levels of depth of target firm shares during the pre-announcement period, consistent with evidence of increased levels of aggressive trading. The abnormal behaviour observed for these metrics increases dramatically on the announcement date as investors trade on news of the takeover. During the post-announcement period the behaviour of these metrics diminishes as news of the takeover bid becomes efficiently incorporated into the target firm s share price. For bidding firms, limited evidence of increased aggressive trading is found with our results, indicating abnormal decreases in returns, increases in volume, widened spreads and reduced depth prior to the takeover announcement. On the announcement date we observe significant negative abnormal returns, positive volume, widened spreads and abnormal trade imbalances. Further, our intra-day analysis of target firms reveals higher levels of abnormal returns, volume, volatility, spreads, depth and VPIN during the morning (10 am 12 pm) and afternoon (2 4 pm) sessions of the announcement day. Examination of our profit-and-loss measure based on trade initiation suggests that investors who initiate trades in either the target or bidding firm around the takeover announcement do not seem to win. This finding is consistent with the expected returns of trading by liquidity demanders,

3 Bugeja et al. 3 rather than individuals with knowledge about the upcoming takeover announcement. However, we observe an abnormal VPIN in approximately 33% of the hourly intervals during the preannouncement period prior to the takeover announcement for target firms, consistent with increased aggressive, and perhaps informed, trading by various investors. We also conduct multivariate analysis of abnormal time-weighted spreads during the two-week event window around a takeover announcement and find that abnormal quoted spreads remain significantly elevated during the entire pre-announcement period. This result holds for a sample of target firms in which the bidder does not increase its toehold in the period preceding the announcement. 4 This suggests that information asymmetries exist between investors prior to takeover announcements, which are in general a surprise to the market. Holistically, our tests show that in Australia there is evidence of increased levels of aggressive trading, however such evidence is primarily confined to target firms. The remainder of the paper is set out as follows. The next section provides a review of relevant US and Australian literature and this is followed by a discussion of the sample selection process. We then present descriptive statistics and results of the study. Lastly, concluding remarks are provided. 2. Prior literature There is a significant amount of prior US research on informed and aggressive trading for target firms around takeover announcements. In comparison, there are fewer studies that examine this issue for bidding firms. This literature analyses the behaviour of several metrics that are suggestive of the existence of informed trading, however whether these measures actually detect informed trading remains inconclusive. These metrics include: an increase in returns and volume, widened bid-ask spreads (as a result of increased adverse selection costs) and reduced market depth prior to a takeover announcement. In contrast to these prior studies we use intra-daily, rather than daily, measures around takeover announcements Returns and trading volume Prior literature documents abnormal returns and volume prior to takeover announcements. Jensen and Ruback s (1983) literature review documents abnormal returns for targets around the takeover announcement of 16% to 30% in 13 US studies conducted using data prior to Keown and Pinkerton (1981) also provide evidence to suggest that informed trading occurs in target firms up to 12 days prior to the public announcement of a takeover and that almost half of the market reaction to a takeover occurs before the public announcement. Other studies also document abnormal returns and volume prior to a takeover announcement and interpret this as evidence of informed trading (Meulbroek, 1992; Eyssell and Arshadi, 1993; Ascioglu et al., 2002). 5 Sanders and Zdanowicz (1992) find evidence to suggest that the target firm pre-bid run-up is due to informed trading; however, they do not observe abnormal volume until the announcement date. Similarly, within an Australian context Dodd (1976), Walter (1984) and Casey et al. (1987) also observe abnormal returns around takeover announcements. Clarkson et al. (2006) use intra-day data and find abnormal returns and volume when takeover rumours are posted on the Hotcopper internet discussion website. 6 In contrast, Jarrell and Poulsen (1989) analyse target firms listed on the NYSE and AMEX and find that increased returns and volume during the pre-announcement period can be attributed to market speculation, rather than informed trading. Similarly, Murray (1994) attributes significant returns and volume to market speculation in the Australian market. More recently, Aspris et al. (in

4 4 Australian Journal of Management XX(X) press) examine 450 Australian takeover announcements and also suggest that there is no significant pre-bid run-up in target firms after controlling for rumours, price-sensitive announcements and the acquisition of a toehold Spreads Earlier literature shows that the bid-ask spread in US dealer markets can be decomposed into three components: order processing costs, inventory holding costs and adverse selection costs. Glosten and Harris (1988) develop a model of the components of the bid-ask spread and confirm that the adverse selection component of the spread is positive. Further, they show that market makers increase the bid-ask spread when they are dealing with informed traders, consistent with the findings of Copeland and Galai (1983). Easley and O Hara (1987) find that larger trades are associated with increased adverse selection costs. Huang and Stoll (1997) also develop a spread decomposition model and find that the adverse selection cost is larger for small/medium trades. This relationship is also shown by George et al. (1991). These findings indicate that informed traders split their orders in an attempt to conceal their trades with other noise trades and that market makers must take trade size into account when determining the bid-ask spread Market microstructure target firms There have been numerous studies that focus on market microstructure metrics of target firms around takeover announcements. Foster and Vishwanathan (1995) examine the behaviour of order flow, quoted bid-ask spreads, depth and adverse selection costs for target firms around tender offers. They find that after a takeover announcement quoted spreads decrease, quoted depth increases, average transaction size increases, the number of transactions increases and that a large percentage of trades are sellerinitiated. The evidence of larger spreads and reduced depth in the pre-announcement period suggests the existence of informed trading prior to takeover announcements. In addition, Kryzanowski and Lazrak (2007) find some evidence to support Foster and Vishwanathan s (1995) results via decreasing adverse selection costs and spreads, and increased depth from the pre-announcement period to the announcement period. A number of studies find contrasting evidence to Foster and Vishwanathan (1995). Jennings (1994), 7 Conrad and Niden (1992) and Ascioglu et al. (2002) find little evidence of widening spreads or increases in adverse selection costs prior to a takeover announcement. Specifically, Conrad and Niden (1992) find that changes in order size are positively related to changes in spreads, which is consistent with Easley and O Hara s (1987) findings. Although the authors find evidence of increased spreads prior to the takeover announcement, they do not find any evidence of increased adverse selection costs. Further, Kryzanowski and Lazrak (2007) document, using the probability of informed trading (PIN) metric developed by Easley et al. (1996), increased trading intensity of uninformed traders, rather than informed, prior to the announcement period for target firms Market microstructure bidding firms In contrast to target firms, there is less literature that analyses the behaviour of returns, volume and microstructure of bidding firms around takeover announcements. Draper and Paudyal (1999) provide US evidence of significant negative abnormal returns on the announcement date. Bugeja and Da Silva Rosa (2010) find that Australian bidding firms exhibit abnormal returns of between 2.91% and 4.09% from 1 to +1 months around the announcement

5 Bugeja et al. 5 date. Cumming and Li (2011) find that bidding firms that have higher levels of information asymmetry exhibit negative run-up cumulative abnormal returns (CARs). Both Lipson and Mortal (2007) and Kryzanowski and Lazrak (2007) find some evidence to suggest that information asymmetries exist between investors prior to the takeover announcement and that there is an increase in liquidity after the announcement. Specifically, Lipson and Mortal (2007) and Kryzanowski and Lazrak (2007) find that, from the pre-announcement period to the announcement period, spreads decrease, depth increases and volume increases. In addition, Lipson and Mortal (2007) find that adverse selection costs 8 decrease and the number of trades increase from the pre- to the post-announcement period Trade imbalance More recent literature examines trade imbalance, i.e. the difference between the level of buyer- and seller-initiated trades. As documented in Chordia et al. (2002), trade imbalance provides a finer measure of trading activity than examining trading volume alone. As an extension of the PIN metric, Easley et al. (2012) use trade imbalance to develop an intra-day volumesynchronised probability of informed trading (VPIN) metric. 9 The authors use this VPIN metric to calculate the level of informed trading surrounding the 6 May 2010 flash crash for various futures contracts, and show that the VPIN measure was able to capture the increasing toxicity of the order flow in the hours and days prior to the collapse. 3. Sample and data selection We extract takeover announcements between 1997 and 2009 from the Connect 4 Takeovers database for ASX-listed firms and then apply the following criteria. First, we include only deals valued at greater than $100 million. This results in an initial sample of 460 deals. Second, we eliminate takeovers that are schemes. 10 Further, only deals in which either the target or bidding firm is listed on the ASX are included in the final sample. This requirement reduces the sample to 299 target firms 11 and 146 bidding firms. We focus on takeover announcements that are not anticipated by the market. As a result we discard takeover bids in which the acquiring firm has more than a 50% toehold in the target firm, as firms with large toeholds are likely to be making mop up bids, which are somewhat predictable. We also exclude deals in which there was a bid in the previous six months as the market may anticipate another takeover bid for the same target firm. We do not control for the success of the bid or whether there are subsequent bids as this involves hindsight bias. In relation to subsequent bids, a large majority of the market reaction will likely have taken place around the first bid made by the bidding firm. The application of these criteria results in a final sample of 195 target firms and 137 bidding firms. In addition, we also examine a subsample of targets in which the bidding firm does not increase their toehold in the target firm in the four months prior to the takeover announcement. We define this as the no-change toehold sample, which contains 141 target firms. 12 This restriction provides a finer examination of takeover bids that were more of a surprise to the market. Our analysis focuses on the official takeover announcement date disclosed to the ASX. We use this date as the basis for our analysis as it provides an unambiguous date that we can use as a valid anchor for our experiment. Furthermore, an examination of trading around the official announcement date is important to various investors and regulators. We do not trace backwards to see if there was takeover media speculation prior to an identified bid, as was done in Aspris et al. (in press), as not all takeover speculation leads to a news story or to a subsequent takeover

6 6 Australian Journal of Management XX(X) Table 1. Descriptive statistics for sample target and bidding firms between 1997 and Characteristic Mean Median Minimum 25% 75% Maximum Deal value $1, $ $ $ $ $17, Target NA $ $ $5.057 $ $ $6, Target NPAT $ $ ($ ) $1.067 $ $ Target market $ $ $ $ $ $24, capitalisation Bidder NA $1, $ $0.211 $ $1, $29, Bidder NPAT $ $ ($ ) $6.779 $ $3, Bidder market $1, $ $9.554 $ $2, $34, capitalisation Premium 29.62% 25.48% (19.79%) 12.68% 38.25% % Toehold 10.70% 9.83% 0.00% 0.00% 19.78% 48.76% Table 1 presents descriptive statistics for 195 target firms and 137 bidding firms. All values are reported in $A million, except for the characteristics Premium and Toehold. Deal value is the effective offer price of the takeover bid multiplied by shares outstanding in the target firm on the announcement date of the takeover. Target (bidder) NA is the difference between the total assets and total liabilities of the target (bidder) firm as reported in the annual report prior to the takeover bid. Target (bidder) NPAT is the net profit after tax as reported by the target (bidder) firm in the annual report prior to the takeover bid. Target (bidder) market capitalisation is the number of shares issued in the target (bidder) firm in the annual report prior to the takeover bid multiplied by the target (bidder) firm share price on the announcement date. Premium is the percentage difference between the effective offer price and target firm share price 20 trading days prior to the takeover announcement. Toehold is the number of shares held by the bidding firm divided by shares outstanding in the target firm. bid. 13 In addition, to study actual bids that were made (as in Aspris et al., in press) and then to trace backwards to see if these bids had takeover media speculation involves a hindsight bias. In any event, Aspris et al. (in press) report that, even after controlling for takeover media speculation and price-sensitive announcements, there still exists evidence of positive and significant CARs from 30 to 1 days suggesting that there is a level of information asymmetry between investors in the target firm prior to the official announcement date. To avoid a research design hindsight bias, we do not use a media-adjusted announcement date. Our results for the 141 nochange toehold target firms provide evidence of increased levels of aggressive trading prior to takeover announcements. All intra-day data needed for our metrics are provided by Sirca Limited (SIRCA, New South Wales, Australia). 4. Descriptive statistics Descriptive statistics for target and bidding firms are detailed in Table 1. The average deal value is $1068 million and aggregate deal value is approximately $200 billion. This fact and the substantial costs incurred by bidding firms during the takeover process highlight the importance of this study in determining whether aggressive trading is taking place. The average market capitalisation of bidding firms is approximately 2.11 times the size of target firms, highlighting the relative difficulty of observing the effects of a takeover for bidding firms. The average bid premium of 29.62% is similar to prior Australian research (Duong and Izan, 2012). 14 The average toehold of 10.7% is similar to prior US studies (Bradley et al., 1988; Betton and Eckbo, 2000).

7 Bugeja et al Methodology To determine the existence of increased levels of aggressive trading in the Australian market we examine the intra-day behaviour of several metrics that measure returns, volumes, volatility, liquidity, profit/loss of trade initiators and the VPIN. This study uses intra-day timing as implemented by Chiyachantana et al. (2004) in their study of information asymmetry around earnings announcements made before and after the introduction of Regulation Fair Disclosure Periods of interest: Event period and control period Chiyachantana et al. (2004) use three periods in their earnings study: the benchmark period, preannouncement period and event period. Each period comprises 26 half-hour intervals. The benchmark period is a two-day period 14 days prior to the earnings announcement, the pre-announcement period occurs two days before the earnings announcement and the event period continues for two days from the earnings announcement. This study adopts a similar approach and considers an event period that consists of a pre-announcement period, announcement date and post-announcement period and compares these to a control period for each takeover announcement. The event period for each takeover announcement is a 21-day period. Within the event period the pre-announcement period lasts for 10 days prior to the takeover announcement and the post-announcement period lasts for 10 days after the takeover announcement. This study implements intra-day timing by segmenting each trading day into three two-hour intervals. In the event period there are 63 two-hour intervals in total. The pre-announcement period comprises 30 two-hour intervals ( 30,, 1), the announcement period comprises three two-hour intervals (0, +1, +2) and the post-announcement period comprises 30 two-hour intervals (+3,, +32). The calculation of the metrics used within this study exclude trades that occurred during the opening and closing auction of each trading day because the identity of the trade initiator cannot be determined in these auctions. Chiyachantana et al. (2004) do not control for day-of-the-week patterns in their study, potentially introducing a day-of-the-week bias. To overcome this potential bias we ensure that the same days of the week are represented within the pre-announcement period, post-announcement period and control period. Thus, in the pre- and post-announcement periods each day-of-the-week is represented twice allowing us to examine patterns without the known day-of-the-week effects influencing the results. The control period used is a 10-week period from 13 to 3 weeks prior to the event period. For each two-hour interval (63 in total) during the event period there is a control interval, which is the average of 10 corresponding twohour intervals during the 10-week control period. Thus, we control for time-of-day effects and use each stock as its own control in isolating abnormal behaviour in the studied metrics. 15 Figure 1 below illustrates our methodology. For example, a takeover announcement made on Wednesday 21 November 2012 means that the pre-announcement period is a two-week period occurring from Wednesday 7 November to Tuesday 20 November 2012 and the post-announcement period is a two-week period occurring from Thursday 22 November to Wednesday 5 December In this example, where the bid was announced overnight prior to the start of trading on Wednesday 21 November, the control period is a 10-week period beginning on 22 August and ending on 31 October Thus, the control period value used to calculate volume-weighted average price (VWAP) returns, for example, is the average of the 10 VWAP return values calculated for each of the 10 am 12 pm intervals on Wednesday from week 13 to week 3 prior to the takeover announcement. This control period VWAP return is then compared to the 10 am 12 pm Wednesday event intervals occurring on 7, 14, 21 and 28 November and 5 December If VWAP returns

8 8 Australian Journal of Management XX(X) Control period Event period Time: 10am 12 pm 10 am 12 pm 10 am 12 pm 10 am 12 pm 10 am 12 pm 10 am 12 pm 10 am 12 pm 10 am 12 pm Day: Wed Wed Wed Wed Wed Wed Wed Wed Date: 22Aug Aug Oct 12 7Nov Nov Nov Nov 12 Week: Dec 12 Figure 1. Timeline. This figure shows the 10-week control period and two-week event period for a takeover announcement made on 21 November do not exist for a particular interval in the event period then this metric is not calculated for these specific intervals. A bootstrapped control and event period distribution is generated in order to test the abnormal behaviour of the metrics examined. The distributions are estimated using 1000 observations. Each of these distributions is created using random sampling with replacement. A t-statistic 16 is used to test for the differences between the mean of the event and control period bootstrapped distributions. As a robustness measure an additional t-statistic test 17 is conducted on the difference between the average event period and average control period values (averaged over n sample firms) to examine the abnormal behaviour of each metric during all 63 intervals in the event period Returns metrics Two price metrics are used to calculate target and bidding firm returns around takeover announcements. 1. Volume-weighted average price (VWAP) n VWAPk = ( Trade Pricek Trade Volumek)/ n Trade Volume k = 1 k = 1 2. Time-weighted midpoint price (TWMP) n k TWMP = Midpoint Price Time / Time k n k k k = 1 k = 1 n k The return for each interval in the event period is calculated as the natural logarithm of the price in one interval (i.e. either the VWAP or the TWMP) divided by the price in the previous interval. Abnormal returns are calculated as the event period return less the return in the corresponding control period Volume metrics We calculate total volume, total value of trading, the number of trades and the average trade size for target and bidding firms in each of the 63 event periods.

9 Bugeja et al Volatility metrics The following volatility metrics are calculated for target and bidding firms around takeover announcements: 1. VWAP volatility VWAP Volatility = ( MAXPRICE MINPRICE)/ VWAP 2. TWMP volatility TWMP Volatility = ( MAXPRICE MINPRICE)/ TWMP 3. Standard deviation of trade to trade returns within an interval, where the returns are defined as the natural logarithm of the last trade price relative to the previous trade price Liquidity metrics To analyse the microstructure effects of aggressive trading occurring around takeover announcements we examine the time-weighted quoted spread and time-weighted depth at the best prices and aggregated over five price levels (and compared to the control period values): 1. Time-weighted quoted spread TimeWeighted Spread = ( Ask Bid) Time / Time k n k k k = 1 k = 1 n k 2. Time-weighted depth n TimeWeighted Depthk = ( AskVolume + BidVolume) ktimek / Time k k = 1 k = 1 n 5.6. Profit/loss measure based on trade initiation (PLM) We develop a specific measure to further examine whether the trading strategies of trade initiators around takeover announcements are profitable. We define trade initiators as liquidity demanders or aggressive traders. This measure calculates whether a trade resulted in a profit or a loss by comparing buyer- and seller-initiated trades to a reference price (RP), where BP is the value of buyerinitiated trades that made a profit and BL is the value of buyer-initiated trades that made a loss. Similarly, SP and SL are defined using seller-initiated trades. The profit/loss measure based on trade initiation is calculated as follows

10 10 Australian Journal of Management XX(X) PLM ( BP + SP) = ( BP + SP + BL + SL) where BP = number of shares bought (reference price buying price) when RP > BP, BL = number of shares bought (buying price reference price) when BP > RP, SP = number of shares sold (selling price reference price) when SP > RP, and SL = number of shares sold (reference price selling price) when RP > SP. A buyer (seller) initiated trade is identified as a profit trade if it is at a price that is below (above) the reference price and a loss if its price is higher (lower) than the reference price. Evidence of investors making abnormal profits, by taking short and long positions in bidding firms and target firms respectively, prior to the announcement of a takeover is indicative of informed trading. This profit/loss measure based on trade initiation has a range between 0 and 1, hence we test to see if this measure is significantly different from 0.5. Values significantly above (below) 0.5 are consistent with investors making profits (losses) around takeover announcements. The reference price for target firms is the closing price on the 20th trading day after the takeover announcement (reference price 1). Two prices are used for bidders. The first is the closing price on the 20th trading day after an announcement (reference price 2) and the second is the closing price on the 250th trading day after an announcement (reference price 3). Our motivation for using reference prices 1 and 2 arises from the short-run trading strategy literature that examines return predictability over a 20-day horizon following large price changes and the dissemination of public information (e.g. Savor, 2012). We use reference price 3 for bidders to capture any long-run downwards drift in the bidder price. It also ensures that the effect of the takeover announcement and subsequent negotiations and revisions are fully revealed in the reference price (e.g. Loughran and Vijh, 1997). In order to control for changes in the market index occurring over the period between the date of the trade and the reference price date, the reference price is scaled by the following factor Index t 1 / Index where Index (t 1) = index value at the close of trading the day before the interval of the date of trade and Index ref = index value at the close of trading on the reference price date Volume-synchronised probability of informed trading (VPIN) metric To investigate the probability of informed trading around the takeover announcement we employ the Easley et al. (2012) VPIN metric. For this measure we use a 500-day control period and 21-day event period in which each trading day is split into six one-hour intervals. The VPIN metric is calculated in the following manner ref OrderImbalance VPIN = ( nv ) where Order Imbalance 18 = seller-initiated trades buyer-initiated trades, n = number of buckets and V = daily volume bucket size. In order to calculate the VPIN metric we conduct a number of steps. Firstly, for each stock we calculate the average daily trading volume within the 500-day control period. Following this, we

11 Bugeja et al. 11 calculate V by dividing the average daily trading volume by 30, which is the same value of n used in the Easley et al. (2012) study. Then, from the beginning of the 500-day control period for each stock we count the trades (buyer- or seller-initiated) in chronological order until the first bucket is filled. A bucket is filled when the total traded volume in the time sequence equals or exceeds the bucket size V. Next, we calculate the order imbalance for the first bucket and thus the VPIN from the formula above. Following this, we continue to calculate the VPIN for the second, third,, n buckets. 19 Further, we record the total VPIN at the end of each trading hour during the 500-day control period and the 21-day event period. The VPIN is the sum of all of the VPIN figures calculated for each bucket filled within a particular trading hour. Note that part of any trade that overfills the last bucket on a particular trading day (or hour) will be included in the first bucket in the next trading day (or hour). In order to examine abnormal changes in VPIN we calculate the average hourly VPIN during the 500-day control period for each of the am, 11 am 12 pm,, 3 4 pm intervals for each stock. We then subtract the respective hourly average control figures from the event VPIN figures for each stock during the am, 11 am 12 pm,, 3 4 pm intervals to calculate abnormal VPIN. Thus, we calculate an abnormal VPIN figure for each stock for each hourly interval within the 21-day event period. 6. Empirical results This section examines the abnormal behaviour of returns, volume, volatility, liquidity, a profit/loss measure based on trade initiation and VPIN for 195 target firms and 137 bidding firms around takeover announcements that were not anticipated by the market. Where necessary we also include analysis of the 141 no-change toehold and 90 no-toehold target firm subsamples. The analysis will focus on the results obtained using a t-statistic test 20 that tests for abnormal differences between the control and event period distributions generated using a bootstrap process Abnormal returns, volume and volatility Targets. Figures 2 6 present the abnormal and cumulative abnormal VWAP returns, 22 abnormal total volume, 23 abnormal average trade size and abnormal VWAP volatility, 24 respectively, for 195 target firms over 63 intervals around the takeover announcement. For our sample of takeover bids there is evidence of a pre-bid run-up in returns, volume and volatility. Further, we observe highly significant abnormal returns on the announcement date accompanied by significant levels of volume and volatility. Following the announcement the abnormal behaviour in volume shows a jagged downtrend (with clear evidence of intra-day patterns) and lower levels of volatility as news becomes incorporated into prices. Over the pre-announcement period significant positive abnormal returns of between 0.3% and 0.65% are observed during separate intervals (Figure 2), reaching a maximum of 1.07% during interval 6. Figure 3 shows a steady increase in the cumulative abnormal VWAP returns 26 during the entire pre-announcement window. These findings support the notion that some investors may have knowledge of the upcoming takeover announcement and are benefitting from the welldocumented pre-bid run-up of the target firm s stock price. These patterns might of course be associated with the bidder building a toehold stake in the target. We test this possibility by investigating a subsample of 141 no-change targets where the bidder does not increase their toehold in the target preceding the announcement. The results are similar and remain statistically significant; specifically for the full sample the CAR for the interval 30 to 3 is 4.39%, while for the 141

12 12 Australian Journal of Management XX(X) Figure 2. Target firm abnormal VWAP return. This figure shows abnormal VWAP return for 195 target firms during the 21-day event period. VWAP is the sum of the total dollars traded for each stock within each two-hour trading interval divided by the volume traded for the same stock within each two-hour interval. The 10-day pre-announcement period is between intervals 30 and 1, the announcement date is between intervals 0 and +2 and the 10-day post-announcement period is between intervals +3 to +32. The 10-week control period occurs between 13 to 3 weeks prior to the takeover where each stock is used as its own control. For each two-hourly interval and controlling for time-of-day and day-of-week factors, abnormal VWAP is calculated as the difference between the means of the event and control bootstrapped distributions that contain 1000 observations (using random sampling with replacement). Due to the large sample size a standard t-statistic test is used under the assumption that the bootstrapped distributions are normally distributed. Figure 3. Target firm cumulative abnormal VWAP return. This figure shows cumulative abnormal VWAP return for 195 target firms during the 21-day event period. VWAP is the sum of the total dollars traded for each stock within each two-hour trading interval divided by the volume traded for the same stock within each two-hour trading interval. The 10-day pre-announcement period is between intervals 30 and 1, the announcement date is between intervals 0 and +2 and the 10-day post-announcement period is between intervals +3 to +32. The 10-week control period occurs between 13 to 3 weeks prior to the takeover where each stock is used as its own control. For each two-hourly interval and controlling for time-of-day and day-of-week factors, abnormal VWAP is calculated as the difference between the means of the event and control bootstrapped distributions that contain 1000 observations (using random sampling with replacement.) Cumulative abnormal VWAP is calculated as the sum of abnormal VWAP for each two-hour interval over the event period. The reported t-statistic is the summation of t-statistics for each interval divided by n. This is based on the assumption of independence between each two-hour interval, which is reasonable because in calendar time there is no considerable clustering of the deals included within the sample.

13 Bugeja et al. 13 Figure 4. Target firm abnormal total volume. This figure shows abnormal total volume for 195 target firms during the 21-day event period. Total volume is the sum of the total shares traded within each two-hour interval. The 10-day pre-announcement period is between intervals 30 and 1, the announcement date is between intervals 0 and +2 and the 10-day post-announcement period is between intervals +3 to +32. The 10-week control period occurs between 13 to 3 weeks prior to the takeover where each stock is used as its own control. For each two-hourly interval and controlling for time-of-day and day-of-week factors, abnormal total volume is calculated as the difference between the means of the event and control bootstrapped distributions that contain 1000 observations (using random sampling with replacement). Due to the large sample size a standard t-statistic test is used under the assumption that the bootstrapped distributions are normally distributed. Figure 5. Target firm abnormal average trade size. This figure shows abnormal average trade size for 195 target firms during the 21-day event period. Average trade size is calculated for each two-hour interval. The 10-day pre-announcement period is between intervals 30 and 1, the announcement date is between intervals 0 and +2 and the 10-day post-announcement period is between intervals +3 to +32. The 10-week control period occurs between 13 to 3 weeks prior to the takeover where each stock is used as its own control. For each two-hourly interval and controlling for time-of-day and day-of-week factors, abnormal average trade size is calculated as the difference between the means of the event and control bootstrapped distributions that contain 1000 observations (using random sampling with replacement). Due to the large sample size a standard t-statistic test is used under the assumption that the bootstrapped distributions are normally distributed. no-change toehold and 90 no-toehold samples the CAR is 3.35% and 2.33%, respectively. Accordingly, while some of the pre-announcement abnormal returns can be associated with bidder toehold acquisition, there is still evidence of anticipation of takeover bids by aggressive investors (potentially informed traders).

14 14 Australian Journal of Management XX(X) Figure 6. Target firm abnormal VWAP volatility. This figure shows abnormal VWAP volatility for 195 target firms during the 21-day event period. VWAP volatility is the difference between the maximum and minimum price weighted by VWAP for each two-hour interval. The 10-day pre-announcement period is between intervals 30 and 1, the announcement date is between intervals 0 and +2 and the 10-day post-announcement period is between intervals +3 to +32. The 10-week control period occurs between 13 to 3 weeks prior to the takeover where each stock is used as its own control. For each two-hourly interval and controlling for time-of-day and day-of-week factors, abnormal VWAP volatility is calculated as the difference between the means of the event and control bootstrapped distributions that contain 1000 observations (using random sampling with replacement). Due to the large sample size a standard t-statistic test is used under the assumption that the bootstrapped distributions are normally distributed. Figure 4 reveals significant abnormal increases in trade volume for 18 of the 30 intervals in the pre-announcement period for targets. More specifically this increase in volume is observed during nine of the 10 intervals directly preceding the announcement date. We also observe intra-day patterns in volume (increased volume during the 10 am 12 noon and 2 4 pm intervals) during the four days prior to the announcement. Similar results are obtained for the sample of 141 no-change toehold and 90 no-toehold targets, which suggest that the observed increases in pre-bid volume are not caused by toehold acquisition by the bidder. In addition to the results observed for returns, these findings provide evidence to support the notion of some information leakage taking place prior to the takeover announcement. During the pre-announcement period abnormal VWAP volatility (Figure 6) fluctuates between 0.5% and 0.65%, with 16 intervals of significant positive volatility interspersed with four intervals of significant negative volatility. When returns, volume and volatility are considered together, our results reveal positive increases in all of these metrics during 11 intervals; however this relationship is only significant during intervals 21, 12, 10, 4 and 2. On the announcement date, significant abnormal VWAP returns of 12.72% (13.10% for the 141 no-change toehold targets) are observed during the 10 am 12 noon period and these continue until interval +6 (10 am 12 noon interval), which suggests that news of the takeover announcement, as well as the subsequent expectations of the bid being revised, becomes fully embedded into the target firms share price within two days of the announcement. In support of these findings, volume and volatility also significantly increase during all three two-hour intervals on the announcement date, also consistent with the bidding firm standing ready to purchase target firm shares at the bid price and hedge funds building a long position in the target firm in expectation that the bid price will be revised upward. During the remainder of the event period abnormal VWAP returns generally fluctuate above and below zero. VWAP volatility is significantly negative between intervals 4 to 10, consistent with

15 Bugeja et al. 15 the resolution of uncertainty regarding the bid s prospects. Ceteris paribus, the takeover announcement has a significant impact upon volume, which is elevated during 31 of 32 intervals during the post-announcement period. In further analysis of the four volume metrics, one key difference is that whilst total volume, value and number of trades are significantly positive in at least six of the 10 intervals prior to the announcement date, average trade size (Figure 5) is significantly negative or insignificant in seven of these 10 intervals. 27 Such behaviour might possibly be explained by aggressive traders concealing their trades from the market through order splitting. Another key difference is that the average trade size remains significantly positive (and at a reasonably constant level) for the remainder of the post-announcement period, whilst the other volume metrics trend downwards as news of the announcement becomes embedded in the target s share price Bidders. Before and after the announcement date most bidder abnormal returns fluctuate around zero. 28 More importantly, there is a significant decrease in returns in the two days prior to the announcement, which suggests that aggressive investors who have prior knowledge of the bid attempt to profit by selling or taking a short position in the bidding firm. In addition, total volume, value and number of trades are significantly positive during intervals 30 to 19, while average trade size is significantly negative during most of the same intervals. This provides additional evidence that aggressive traders may be attempting to hide their trades from the market. However, volume is largely insignificant in the intervals prior to the takeover announcement. It needs to be noted, of course, that the opportunity to benefit from advanced knowledge of a takeover is more likely to be evident in the trade volume of the target (and perhaps in derivative trading in targets and bidders). 29 On the announcement date all four volume metrics and VWAP volatility are significantly positive and are accompanied by significant negative returns in the bidding firm. This result is consistent with the previous literature, which explains that bidding firms are penalised by the market as a result of offering a large premium for the target firm, or for undertaking bad acquisitions of target firms. Note however, that none of the CARs for bidders within the entire investigation window show strong statistical deviation from zero. 30 Further, consistent with the trends in returns and volume, volatility fluctuates between 1% and 1% during the post-announcement period Abnormal time-weighted quoted spread and depth In the following sections several microstructure metrics that are proxies for the level of information asymmetry are examined. A widening of the bid-ask spread and reduced depth prior to a takeover announcement is indicative of traders and pseudo market makers protecting themselves from trading with aggressive traders, or increased levels of asymmetric information among market participants (Lee et al., 1993). While this study focuses on firms listed on the ASX, which is an orderdriven market in which spreads and prices are determined by the flow of buy and sell orders, both Glosten (1994) and Brockman and Chung (1999) argue that spreads in order-driven markets are similar to spreads set in quote-driven markets Targets. Figures 7 and 8 display the behaviour of the abnormal time-weighted quoted bid-ask spreads for 195 target firms and time-weighted depth at the best prices for 141 target firms, respectively. The results show that quoted spreads significantly widen up to four days before the announcement and that this is accompanied by significantly reduced depth. On the announcement date, quoted spreads remain significantly elevated and depth significantly increases. During the post-announcement period both positive and negative 31 abnormal time-weighted quoted spreads

16 16 Australian Journal of Management XX(X) T-Statistic Abnormal Quoted Spread Figure 7. Target firm abnormal quoted time-weighted spread. This figure shows abnormal quoted time-weighted spread (in AUD) for 195 target firms during the 21-day event period. Time-weighted quoted spread is the sum of the bid-ask spread multiplied by the length of time that each specific quote was outstanding divided by the sum of the length of time each quote was outstanding for each two-hour interval. The 10-day pre-announcement period is between intervals 30 and 1, the announcement date is between intervals 0 and +2 and the 10-day post-announcement period is between intervals +3 to +32. The 10-week control period occurs between 13 to 3 weeks prior to the takeover where each stock is used as its own control. For each two-hourly interval and controlling for time-of-day and day-of-week factors, abnormal quoted time-weighted spread is calculated as the difference between the means of the event and control bootstrapped distributions that contain 1000 observations (using random sampling with replacement). Due to the large sample size a standard t-statistic test is used under the assumption that the bootstrapped distributions are normally distributed. 12,000,000 10,000,000 8,000,000 6,000,000 4,000,000 2,000, ,000, T-Statistic Abnormal Total Depth Figure 8. Target firm abnormal time-weighted depth. This figure shows abnormal time-weighted depth for 141 target firms during the 21-day event period. Time-weighted depth is the sum of the bid volume and ask volume at the best prices sampled at five-minute intervals weighted within each two-hour interval. The 10-day pre-announcement period is between intervals 30 and 1, the announcement date is between intervals 0 and +2 and the 10-day post-announcement period is between intervals +3 to +32. The 10-week control period occurs between 13 to 3 weeks prior to the takeover where each stock is used as its own control. For each two-hourly interval and controlling for time-of-day and day-of-week factors, abnormal time-weighted depth is calculated as the difference between the means of the event and control bootstrapped distributions that contain 1000 observations (using random sampling with replacement). Due to the large sample size a standard t-statistic test is used under the assumption that the bootstrapped distributions are normally distributed.

17 Bugeja et al. 17 are observed for the 195 targets. Further, we observe significantly increased depth following the takeover announcement. The abnormal time-weighted quoted spread is significantly negative between intervals 21 to 16 in the pre-announcement period reaching a low of $ Relative to the control period this behaviour in the spread suggests that there is no additional information asymmetry between investors. However, during intervals 12, 11, 8 and 7 the observed abnormal quoted spread is significantly positive, increasing to a maximum value of $0.013 (or $0.034 for the 90 no-toehold sample). We note that during the six intervals prior to the announcement date the bid-ask spread remains insignificantly different from its expected value. In contrast, for the 141 no-change toehold and 90 no-toehold samples, the spread significantly increases during two of these same six intervals, as these takeover deals are a surprise to the market. In addition, we observe abnormally lower levels of depth during 12 intervals prior to the announcement date, reaching a maximum of 23,343 shares. In unreported findings, such reduced levels of depth are driven by lower depth on the bid side of the order book. In contrast, for the sample of 195 target firms we observe significant increases in depth prior to the takeover announcement. An increased spread and decreased depth during these intervals could be explained by reduced liquidity from liquidity suppliers due to increased information asymmetry between investors. 32 On the announcement date the abnormal quoted time-weighted spread peaks at $0.026 (or $0.047 for the 90 no-toehold targets) during the 10 am 12 noon morning session and depth increases to approximately 10 million shares. This increase in the spreads and depth is due to abnormal depth on the bid side of the order book as the announcement creates divergence in investor opinion and investors trade on news of the takeover announcement. On the day following the announcement the abnormal time-weighted quoted spread becomes insignificant. Further, between intervals +7 to +12 and +20 to +26 the abnormal time-weighted spread is significantly negative. 33 Such behaviour can be expected after a period of uncertainty (takeover announcement) and also as a result of abnormal levels of volume and liquidity during these intervals. 34 In addition, during the entire post-announcement period, depth is significant, exhibiting intra-day patterns and reaching a maximum of approximately 10 million shares. Again, these depth findings are primarily driven by abnormal levels of depth on the bid side of the order book (abnormal bid depth is approximately 14 times larger than abnormal ask depth). 35 In unreported results, we observe similar patterns in time-weighted bid, ask and total depth measures aggregated over the first three and five price levels Bidders. Throughout the pre-announcement period the quoted bid-ask spread remains significantly elevated and ranges between $0.01 and $0.02. During intervals 7 and 6 the quoted spread is insignificant but during intervals 5 to 1 an upward trend in spreads is evident. On the announcement date the time-weighted quoted spread increases to approximately $0.09. Again, this behaviour could be the result of an increase in the diversity of investor beliefs. In addition, we observe reduced levels of depth during five intervals prior to and during all three two-hour intervals on the announcement date. Thus, the behaviour of both quoted spreads and depth provide some evidence that suggests the existence of asymmetric information for bidders prior to takeover announcements. In unreported results, we confirm that ask depth drives these findings, which could be explained by the presence of aggressive traders who are prepared to sell or short bidding firm shares, to potentially profit from the (typically observed) underperformance of the bidding firm s stock price around takeover bids. The announcement effect on the quoted bid-ask spread lasts until interval +8, two days after the announcement date. During the post-announcement period significantly elevated time-weighted spreads and depth are observed, possibly due to information asymmetry between investors as to

18 18 Australian Journal of Management XX(X) Figure 9. Target firm profit/loss measure based on trade initiation. This figure shows a profit/loss measure based on trade initiation (PLM) and a cumulative abnormal profit/loss measure based on trade initiation (CPLM) for 195 target firms during the 21-day event period. The 10-day pre-announcement period is between intervals 30 and 1, the announcement date is between intervals 0 and +2 and the 10-day postannouncement period is between intervals +3 to +32. PLM calculates whether a trade resulted in a profit or loss by comparing buyer- and seller-initiated trades to a reference price for each two-hour interval. Abnormal PLM occurs when the event period PLM is significantly different from 0.5 (determined by a standard t-statistic test). CPLM is calculated as the sum of the abnormal PLM for each interval over the event period. A reference price occurring 20 trading days after the takeover announcement is used. T-statistics are reported for the PLM metric only. Both the t-statistics and CPLM share the same scale on the right vertical axis. whether the takeover bid will succeed, whether the bid will be revised and whether a competing bidder will enter the market Abnormal profit/loss measure based on trade initiation (PLM) The results so far display abnormal increases (decreases) in returns, volume, volatility and spreads (depths) during several intervals around the takeover announcement for target firms. A unique profit/loss measure based on trade initiation is examined in this section in order to infer whether these changes can be attributed to trade initiators making profits around takeover announcements. If the PLM is significantly greater than 0.5, then this indicates that trade-initiating investors are achieving abnormal profits from their trading. This inference assumes that investors hold their position in the stock and close out their strategy on the specified reference date Targets. Figure 9 displays the results for this metric for target firms. The PLM fluctuates insignificantly from its expected value throughout the event period; however a significant decrease is evident during and just after the announcement date. Analysis of target firms shows that during the pre-announcement period the PLM is significantly below 0.5 on only one occasion, i.e. during interval 19. Closer inspection of each component of the PLM reveals that a large increase in the value of seller-initiated losses is the cause of this behaviour. Further, Figure 9 shows that during the pre-announcement period the cumulative abnormal profit/loss measure based on trade initiation (CPLM) is negative and reaches a minimum of approximately These results show that investors who implement trading strategies prior to the takeover announcement do not seem to win, which is consistent with trading conducted by liquidity demanders. Thus, on average our findings suggest that informed traders do not use market orders to buy target firms around takeover announcements. We acknowledge that it is possible that

19 Bugeja et al. 19 informed traders may prefer to use limit orders around information events (see Beber and Caglio, 2005; Kaniel and Liu, 2006). The PLM shows significant decreases during intervals +1 and +2 on the announcement date. This decrease on the announcement date can be attributed to both buyer- and seller-initiated losses, which suggests that investors who have purchased and sold target firm shares in order to profit from the announcement of a takeover do so at unfavourable prices relative to the (market-adjusted) reference price. Similar findings are found during intervals +3 to +7 and +11. In sum, significant losses are made by trade-initiating investors immediately after the bid announcement. Further, although abnormal returns are made by the target firm during the preannouncement period, the evidence (on average) from the PLM indicates that liquidity demanders make losses from trades initiated around the takeover announcement Bidders. In this study two separate reference prices are used to determine whether investors are making abnormal profits by trading bidding firm shares. Evidence presented in Figure 10 (using a reference price 20 days following the announcement date) suggests that trade initiators who trade in the bidding firms do not seem to win. In addition, the results obtained using a reference price 250 days following the announcement date also indicates that trade initiators do not make significant profits or losses. 37 During the pre-announcement period (Figure 10) the PLM fluctuates around 0.5 though it is significantly lower during intervals 10 and 7. These losses are mainly attributed to buyerinitiated losses. Further, during the 12 2 pm session on the day before the announcement the PLM increases to This occurs due to an increase in seller-initiated profits as trade initiators sell or go short in the bidding firm prior to the announcement of the takeover. However, this finding provides limited evidence to suggest that trade initiators are informed traders or have knowledge of the upcoming takeover bid as the observed value is not statistically different to 0.5. After the announcement the PLM is significantly less than 0.5 in intervals +5 and +20, due to buyer- and seller-initiated losses, respectively. Throughout the entire event period investors only make significant profits during interval +18, in which investors profit by buying bidding firm shares at prices below the reference price. Using a reference price 20 days following the announcement date, the evidence reveals six negative and only one positive significant PLM value during the entire event period. Consistent with findings in the target firm, trade initiators in bidder-firm shares do not seem to win, which is consistent with the trading behaviour of liquidity demanders Abnormal volume-synchronised probability of informed trading (VPIN) In addition, we also examine takeover announcements using the novel volume-synchronised probability of informed trading (VPIN) metric (Easley et al., 2012) Targets. Analysis of Figure 11 reveals evidence of unusual trading prior to the takeover announcement as measured by the VPIN metric. Further, abnormal VPIN is observed on the announcement date and during the post-announcement period as information about the takeover becomes incorporated into prices. During the pre-announcement period the VPIN metric is significant at a 5% level in of the 60 intervals, particularly in the four trading days prior to the takeover announcement (intervals 24 to 1). On average, these abnormal increases in the VPIN measure suggest that the probability of informed trading is much higher when compared to control period values. As a result, regulators might be concerned about information leakage and asymmetric information between investors. We

20 20 Australian Journal of Management XX(X) T-statistic CPLM RF1 PLM RF1 Figure 10. Bidding firm profit/loss measure based on trade initiation. This figure shows a profit/loss measure based on trade initiation (PLM) and a cumulative profit/loss measure based on trade initiation (CPLM) for 137 bidding firms during the 21-day event period. The 10-day pre-announcement period is between intervals 30 and 1, the announcement date is between intervals 0 and +2 and the 10-day post-announcement period is between intervals +3 to +32. PLM calculates whether a trade resulted in a profit or loss by comparing buyerand seller-initiated trades to a reference price for each two-hour interval. Abnormal PLM occurs when the event period PLM is significantly different from 0.5 (determined by a standard t-statistic test). CPLM is calculated as the sum of the abnormal PLM for each interval over the event period. A reference price occurring 20 trading days after the takeover announcement is used. T-statistics are reported for the PLM metric only. Both the t-statistics and CPLM share the same scale on the right vertical axis. Figure 11. Target firm abnormal VPIN metric. This figure shows abnormal VPIN for 191 target firms during the 21-day event period. VPIN is the absolute value of order imbalance weighted by the number of buckets multiplied by the daily volume bucket size for each hourly interval. The 10-day pre-announcement period is between intervals 60 and 1, the announcement date is between intervals 0 and +6 and the 10-day post-announcement period is between intervals +7 to +66. The 500-day control period occurs prior to the 21-day event period where each stock is used as its own control. For each hourly interval and controlling for time-of-day and day-of-week factors, abnormal VPIN is the difference between VPIN during the event window and the average control VPIN calculated over a 500-day control window. A standard t-statistic test is conducted on the difference between average event period and average control period values (averaged over n sample firms) during all 126 intervals within the event period. acknowledge that this measure does not distinguish between whether volume is driven by buyer- or seller-initiated trades.