Stock repurchases and liquidity

Journal of Financial Economics ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Contents lists available at ScienceDirect

Journal of Financial Economics journal homepage: www.elsevier.com/locate/jfec

Stock repurchases and liquidity$ Alexander Hillert a, Ernst Maug a,n, Stefan Obernberger b a b

University of Mannheim, L9 1-2, 68131 Mannheim, Germany Erasmus University Rotterdam, Burgemeester Oudlaan 50, 3000DR Rotterdam, The Netherlands

a r t i c l e i n f o

abstract

Article history: Received 31 July 2014 Received in revised form 17 February 2015 Accepted 23 February 2015

We analyze the impact of share repurchases on liquidity based on a new comprehensive data set of realized share repurchases in the US, which covers 50,204 repurchase months between 2004 and 2010. Using instrumental variable analysis, we show that repurchases unequivocally improve liquidity and suggest that endogenous controls have confounded results in earlier studies. Liquidity also influences how firms execute repurchase programs. Repurchases provide liquidity when other investors sell the firm's stock or in times of crisis. No evidence exists that firms reduce liquidity when they trade on private information. & 2015 Published by Elsevier B.V.

JEL classification: G10 G30 G35 Keywords: Share repurchases Market microstructure Liquidity Limit order markets Informed trading

1. Introduction This paper investigates how firms affect the liquidity of the market for their own stock when they repurchase shares and which strategies firms adopt when they

☆ We are grateful to Manuel Adelino, Alon Brav, Casey Cheng, Darwin Choi, Philipp Geiler, John Graham, Jacob Oded, Ailsa Roell, and Florian Weigert for advice on this project and seminar participants at the Seventh Annual Meeting of the Swiss Finance Institute, University Carlos III, Erasmus Liquidity Conference, Frankfurt School of Finance, German Finance Association Conference, Hong Kong Polytechnic University, National University of Singapore, Singapore Management University, University of New South Wales, University of South Australia, and University of Sydney for fruitful discussions and feedback. We thank Thomas Johann and Christian Westheide for providing us with the data set on monthly fund holdings from Morningstar and Olga Lebedeva for helping us to construct some of our liquidity measures. n Corresponding author. Tel.: þ 49 621 181 1952. E-mail address: [email protected] (E. Maug).

execute buyback programs. Beginning with Barclay and Smith (1988), a large literature seeks to understand whether firms provide or demand liquidity when they repurchase shares. From the point of view of market microstructure, firms are simply another category of traders when they conduct open market repurchases. The literature has identified several dimensions in which traders could differ and that affect their demand for or supply of liquidity: traders' time horizon (patience, willingness to pay for immediacy), their informational advantage relative to other traders, and their size. Most of the prior literature on repurchases and liquidity builds on the theoretical framework of Barclay and Smith (1988), who emphasize information as the main dimension and ask whether firms act like informed investors and increase the adverseselection component of the spread or whether they enter the market as liquidity traders. By contrast, we build on more recent research on limit order markets (e.g., Foucault, Kadan, and Kandel, 2005; Kaniel and Liu, 2006) and

http://dx.doi.org/10.1016/j.jfineco.2015.08.009 0304-405X/& 2015 Published by Elsevier B.V.

Please cite this article as: Hillert, A., et al., Stock repurchases and liquidity. Journal of Financial Economics (2015), http: //dx.doi.org/10.1016/j.jfineco.2015.08.009i

2

A. Hillert et al. / Journal of Financial Economics ∎ (∎∎∎∎) ∎∎∎–∎∎∎

emphasize investors' time horizon, i.e., their willingness to pay for immediacy. The overall conclusion from our results is that firms should be regarded as large and patient investors when they buy back their own stock. Research on the question of whether share repurchases increase or reduce liquidity has not converged. Several authors have analyzed the impact of repurchases on stock liquidity and found that repurchases reduce liquidity in France and Hong Kong, while their impact is positive in Canada, Italy, Sweden, and Switzerland.1 The evidence for the US is ambiguous. The earlier literature on US data analyzes repurchase announcements as data on realized share repurchases have become available only recently and are difficult to collect.2 Apart from data availability, the diversity of results can be attributable to methodological differences across studies as well as trading environments, which vary across markets and have changed over time, with most exchanges now adopting electronic limit-order trading. In this paper, we provide a fresh look at this topic. We collect a much larger and more accurate data set than has been available in previous studies for the US and develop instruments for repurchases and for liquidity to disentangle the causal connections between these two variables. Our theoretical point of departure is the theory of modern limit order markets, because it seems the most appropriate framework for analyzing US stock markets during our sample period.3 Following Foucault, Kadan, and Kandel (2005), we conceive of limit order markets as markets for immediacy, in which traders can either demand immediacy, e.g. through placing market orders, or supply immediacy through placing limit orders (see also Grossman and Miller, 1988). The critical characteristic of traders in these markets thus is their time horizon or patience, which can be affected by liquidity needs, private information, and risk aversion. We therefore test whether firms act as patient investors by providing liquidity and investigate which firm characteristics affect their patience and their supply of liquidity. 1 See Brockman and Chung (2001) for Hong Kong, De Cesari, Espenlaub, and Khurshed (2011) for Italy, Ginglinger and Hamon (2007) for France, Chung, Isakov, and Perignon (2007) for Switzerland, McNally and Smith (2011) for Canada, and Rasbrant and Ridder (2013) for Sweden. 2 Barclay and Smith (1988) look at repurchase announcements and find a negative impact for the US, whereas Miller and McConnell (1995) find no effect. Wiggins (1994) and Franz, Rao, and Tripathy (1995) examine announcements of open market repurchases, and Nayar, Singh, and Zebedee (2008) analyze fixed price tender offers and dutch auctions. The last three studies find a positive relation between repurchases and liquidity. Cook, Krigman, and Leach (2004) provide univariate analyses of a small, hand-collected sample of realized open market share repurchases and find a positive effect. Ben-Rephael, Oded, and Wohl (2014) study recently disclosed, realized open market repurchases. The authors find ambiguous results and conclude from indirect evidence that “repurchasing firms consume liquidity rather than provide it” (p. 1301). 3 See, e.g., Jain (2005) for a discussion of market mechanisms. He classifies US markets as hybrid markets, in which traders can interact directly through the limit-order book or through dealers. The results of Comerton-Forde, Hendershott, Jones, Moulton, and Seasholes (2010) (see their Internet Appendix), and Hasbrouck and Sofianos (1993) suggest that the specialist participation rate is only about 8–15%. Hence, it seems appropriate to treat these hybrid markets as limit order markets.

To test our hypotheses, we collect monthly data on all repurchase programs and stock repurchases from all US companies from 2004 to 2010 from Securities and Exchange Commission (SEC) forms 10-Q and 10-K and compute three different liquidity measures. Our data set covers 6,537 repurchase programs with an average (median) size of 6.59% (5.27%) of the firm's market capitalization. We collect data on 6,150 firms, of which 2,930 firms conduct at least one repurchase during our sample period. Our data set is significantly larger and also more accurate than the ones used in previous research.4 In addition, we collect information on program characteristics, which allows us to condition on them and develop new instruments. Our methodology departs from previous research in three important ways. First, we avoid contemporaneous control variables. Second, we use firm fixed effects and time fixed effects to control for cross-sectional characteristics and macroeconomic factors. Hence, no part of our identification comes from cross-sectional differences between firms. While simple, these two steps together already account for most qualitative differences between our results and those in the literature, as well as for differences between previous contributions themselves. Third, we recognize that liquidity and repurchases are simultaneously determined and, therefore, introduce instruments for both directions in this relation. Our analysis focuses on repurchases under previously announced repurchase programs, and our data allow us to use two characteristics of these programs as instruments for realized repurchases, namely, the size and the time that has elapsed since the inception of the program.5 The time since program initiation increases each month by one month and the size of the program is fixed at the beginning, when the program is announced. Thereby, we ensure that predicted repurchases are not related to the dynamic development of liquidity during the execution of the program. We use three instruments for liquidity. The first instrument is the median monthly trading volume of all firms that never undertake a repurchase. This instrument measures a factor of liquidity that is common to all firms and cannot be influenced by the execution strategy of any particular firm's stock repurchase program. Alternatively, we use lagged trading volume in some specifications. The third instrument is the absolute difference between the 4 Previous work on realized repurchases (Dittmar, 2000; Stephens and Weisbach, 1998) is mainly based on a measure constructed from Compustat purchases of common stock. Banyi, Dyl, and Kahle (2008) show that this measure, which is available only on a quarterly basis, “deviates from the actual number of shares repurchased by more than 30% in about 16% of the cases” (p. 460). The only exception is Cook, Krigman, and Leach (2004), who analyze the daily open market repurchases of 64 firms that voluntarily disclosed their repurchase programs. 5 Given our setup, we need instruments for actual repurchases under previously announced programs. The only other paper that uses instruments for repurchases is Bonaimé, Hankins, and Harford (2014), who use state-by-state transitions in regulation, which removed a preference for dividends for some institutional investors. These transitions took place before our sample period and are not suitable to instrument for actual repurchases.

Please cite this article as: Hillert, A., et al., Stock repurchases and liquidity. Journal of Financial Economics (2015), http: //dx.doi.org/10.1016/j.jfineco.2015.08.009i

A. Hillert et al. / Journal of Financial Economics ∎ (∎∎∎∎) ∎∎∎–∎∎∎

stock price and $30, which has been used by Choi, Getmansky, and Tookes (2009), and is motivated by the notion of an optimal trading range centered around $30, so that stocks within this range are more liquid. All instruments are motivated in more detail below and validated using several standard specification tests. We first establish the overall impact of repurchases on liquidity and find that it is unequivocally positive: All liquidity measures indicate an improvement in liquidity in response to repurchases. We identify two sources of bias that account for differences between our results and those in the previous literature. First, our instrumental variables (IV) analysis shows that repurchases affect contemporaneous trading volume net of the repurchase volume itself. Controlling for contemporaneous trading volume (or stock returns or volatility) therefore controls for the effect we wish to measure. Once we remove contemporaneous controls we consistently obtain a positive impact of repurchases on liquidity using OLS, even though this approach does not adequately address endogeneity concerns. Second, comparing ordinary least squares (OLS) results with instrumental variable regressions indicates a significant positive bias, which suggests that unobserved variables move repurchases and the illiquidity measures in the same direction. Further univariate analyses suggest that these omitted variables are associated with higher short interest, lower returns, and higher volatility, indicating that liquidity and repurchases are both influenced by the same investor perceptions. We follow up on these findings by searching for subsets of repurchases that we believe are most likely to reduce liquidity. We hypothesize that firms have a higher demand for immediacy when they reach the expiration date of repurchase programs with a fixed duration or when they need to disburse large amounts of cash. However, we cannot identify any liquidity-reducing repurchases. In the second step of the analysis we investigate firms' trading strategies in more detail. We follow Brockman and Chung (2001) and hypothesize that firms attempt to reduce their transaction costs and repurchase more shares when the market is more liquid. In addition, if firms are risk-averse and try to minimize price impact, they should spread their repurchases over time but front-load the execution of the program to the earlier months in the program. We find strong support for both hypotheses. Next, we ask if firms sometimes try to actively influence the liquidity of the market. Hong, Wang, and Yu (2008) model firms as “buyers of last resort” who provide price support when the stock price falls and when other investors exert downward price pressure. We identify such situations by using order imbalances, trades of mutual funds, and short interest as alternative measures of the trading behavior of other investors and find that firms do act as contrarian traders who buy when other investors sell. In addition, we follow Nagel (2012) and view firms as potential providers of liquidity in times of crisis, i.e., when the volatility of the stock market is high and when stock prices fall. Again, the evidence confirms our prediction. Hence, we conclude that firms provide liquidity whenever downward pressure exists on their stock price, but absent such pressure they are careful not to move the price up.

3

Accordingly, they spread their trades over time and adapt their trading strategies to the liquidity of the market. Finally, we investigate if trading on private information induces firms to behave as more impatient traders who try to execute repurchases fast and thereby consume liquidity. We test whether informed repurchases are associated with higher spreads. We adapt the methodology used in the insider trading literature and measure the information content of repurchases by using abnormal stock returns. In our sample, repurchases provide even more liquidity on average when they are followed by higher abnormal stock returns around the filing date. Analyzing subsequent abnormal returns leads to a similar conclusion. This finding is inconsistent with the notion that information-based repurchases reduce market liquidity. However, our results resonate well with recent findings in the microstructure literature, which suggests that informed traders supply liquidity by placing limit orders, at least when they exploit long-lived information. Our measurement of abnormal returns extends over six months, which should be regarded as long term. Bloomfield, O'Hara, and Saar (2005) and Kaniel and Liu (2006) both show in different contexts that informed traders use limit orders and supply liquidity when their information is sufficiently long-lived and their informational advantage is not too large. (See the more detailed discussion in Section 2.) We contribute to the literature in several ways. First, we combine several methodological improvements (a combination of fixed effects, exogenous control variables, and instruments for repurchases) that help us to avoid a number of biases that have confounded previous results. We also use a broader set of liquidity measures than previous papers. Second, we base the analysis on the fullest and most accurate data set that has been used in any study of repurchases in the US so far. Previous papers either rely on repurchase announcements or manually collect data on much smaller samples. Third, we derive and test a number of additional hypotheses that are new to the literature by reframing the issue in terms of the theory of limit order markets. The paper proceeds as follows. The next section (Section 2) outlines our theoretical framework in more detail. Section 3 describes our data and motivates the instrumental variables 134. Section 4 analyzes how repurchases affect liquidity, whereas Section 5 analyzes how liquidity affects repurchases. Section 6 uses event-study methodology to show how the information content of share repurchases affects how repurchases influence liquidity. Section 7 concludes.

2. Theoretical framework In this section, we provide a brief overview of the theoretical framework that lies at the basis of the hypotheses we develop in subsequent sections. Theoretical models of limit order markets emphasize traders' time horizon or patience as the critical characteristic for their decision to supply or demand liquidity. (See Foucault, Kadan, and Kandel, 2005; Parlour and Seppi, 2008, for a survey of recent research on limit order markets.) By

Please cite this article as: Hillert, A., et al., Stock repurchases and liquidity. Journal of Financial Economics (2015), http: //dx.doi.org/10.1016/j.jfineco.2015.08.009i

4

A. Hillert et al. / Journal of Financial Economics ∎ (∎∎∎∎) ∎∎∎–∎∎∎

comparison, the trading motive, i.e., whether traders want to make use of private information or whether they trade for liquidity reasons, seems relevant only to the extent that it influences their time horizon. The determinants of investors' patience can be grouped into three broad categories: liquidity needs, private information, and risk aversion. Another relevant characteristic could be the speed with which traders can respond, which creates market power to benefit fast traders relative to slow traders and affects limit-order strategies (Hoffmann, 2014). Analyzing this aspect is infeasible without access to higher-frequency data on share repurchases and therefore is not discussed here in more detail. Liquidity needs have always been emphasized in the microstructure literature. Because firms usually buy back significant proportions of their own stock, they are probably more akin to large institutional investors, whose liquidity needs have been associated with shocks to their balance sheets. These shocks can take the form of investor flows in case of mutual funds (Coval and Stafford, 2007) or of shocks to refinancing costs in case of hedge funds (Aragon and Strahan, 2012; Franzoni and Plazzi, 2013). Based on the prior literature, we expect that the major characteristics of firms' balance sheets that are relevant for their trading horizon when buying back their stock are their leverage, their cash position, and their investment opportunities. Private information was the focus of the earlier literature on repurchases and liquidity, which follows Barclay and Smith (1988) and equates informed trading with a demand for liquidity. By contrast, more recent research in limit-order settings suggests that informed traders provide liquidity if they trade on sufficiently long-lived information. Kaniel and Liu (2006) develop a model in which traders prefer market orders only if they trade on shortlived information. They prefer limit orders if they trade on long-lived information, which allows informed investors to earn the spread on their transactions. Limit orders should then reveal more information than market orders if information is long-lived.6 Kaniel and Liu (2006) also provide empirical evidence to support these claims (see also Anand, Chakravarty, and Martell, 2005). Bloomfield, O'Hara, and Saar (2005) show in an experimental asset market that traders with long-lived information could use limit orders and that uninformed traders seem to avoid limit orders, which expose them to the risk of trading against informed traders. The setting of Barclay and Smith (1988) abstracts from these dynamic considerations and equates informed trading with liquidity demand, because in their static framework informed traders (i.e., firms) do not have the opportunity to trade patiently and delay trades or spread them over a longer period. Risk aversion is emphasized by the literature on the trading strategies of block traders (e.g., Almgren and Chriss, 2001; Vayanos, 2001). In these models, more riskaverse traders are less patient and more willing to pay for 6 Chakravarty and Holden (1995) and Harris (1998) provide earlier theoretical analyses of informed trading in limit order markets. Goettler, Parlour, and Rajan (2009) develop a dynamic model of a limit order market with heterogeneous investors and show that investors with the highest inclination to become informed (speculators) use limit orders.

immediacy, because they want to reduce their exposure to stock price risk. In the following sections, we develop more specific hypotheses based on this theoretical framework by identifying empirical proxies for firms' liquidity needs (Section 4) and private information (Section 6) and by making indirect inferences from their trading behavior on their risk aversion (Subsection 5.1).

3. Data and methodology This section discusses our data collection and sample construction (Subsection 3.1), how we instrument for the endogenous variables (Subsection 3.2), and finally provides descriptive statistics for the main variables (Subsection 3.3). 3.1. Sample construction New disclosure requirements in the US mandate the publication of monthly share repurchases under the new Item 2(e) of Form 10-Q and under the new Item 5(c) of Form 10-K. The requirement applies to all periods ending on or after March 15, 2004. Under these rules, firms have to report the total number of shares purchased, the average price paid per share, the number of shares purchased under repurchase programs, and either the maximum dollar amount or the maximum number of shares that could still be purchased under these programs. We are interested in the shares repurchased under a program, which often differs from the total number of shares repurchased (see Appendix A.1 for further details). To collect data on realized repurchases, we use Center for Research in Security Prices (CRSP) to identify all ordinary shares (share codes 10 and 11) that are traded on the NYSE, Amex, and Nasdaq (exchange codes 1, 2, and 3), which gives us 6,504 firms over the period from January 2004 to December 2010. We are left with 6,315 firms after matching these data with Compustat. For all firms we use a computer script to download and extract the repurchase data from all 10-Q and 10-K filings between January 1, 2004 and March 31, 2011. Because many firms do not adhere to the proposed disclosure format, we manually check and correct all observations. We identify 9,100 repurchase programs and apply several screens to assure completeness and integrity of our data (see Appendix A.1 for further details). After these screening procedures, we are left with 6,537 repurchase programs, half of which have no fixed expiry date, i.e., they remain active until they have been completed. This aspect can indicate some desire on the part of firms to retain more flexibility in their future execution strategy and poses a potential concern for one of our instruments. However, we provide robustness checks in which we restrict the sample to those observations with a definite program length and find no differences. We therefore ignore this aspect of the data. The average completion rates in our sample are 45.53%, 53.17%, and 59.31%, respectively, one, two, and three years after the beginning of the program. These completion rates are below those

Please cite this article as: Hillert, A., et al., Stock repurchases and liquidity. Journal of Financial Economics (2015), http: //dx.doi.org/10.1016/j.jfineco.2015.08.009i

A. Hillert et al. / Journal of Financial Economics ∎ (∎∎∎∎) ∎∎∎–∎∎∎

found in the literature for earlier samples.7 We attribute this difference to the decline in repurchase activity during the financial crisis from September 2008 to December 2009. Precrisis completion rates are four to eight percentage points higher than the sample average. Usually, repurchase programs specify a maximal number of shares or a maximal dollar amount that can be purchased. If a program is specified in shares (dollars) we compute Program Size as the maximal number of shares (maximal dollar amount) that could be purchased divided by the number of shares outstanding (market capitalization) at the time of the announcement. (Table 1 contains variable definitions.) In the last step, we merge the data with Institutional Brokers' Estimate System (I/B/E/S) and Daily Trade and Quotes (TAQ). The final sample contains 6,150 firms. Of these, 2,930 firms have repurchase programs. We have 106,898 firmmonths with an active program. Firms conduct share repurchases in 50,204 of these firm-months. Table A1 of the Internet Appendix shows the number of observations, (repurchasing) firms, and repurchase months. The large number of observations allows us to overcome the limitations of monthly data. The monthly frequency of the data makes our measurements more noisy if there are relations within months. This problem is no different from measurements at the daily level if relations are also intraday, which is most likely the case with liquidity variables and repurchases. The lower the data frequency, the harder it is to detect such relations, which is why having a large sample is useful here. Brockman and Chung (2001) use daily data, but even within a single day liquidity can vary and we would need instruments as we could not be sure whether repurchases happened before or after our measurements of liquidity. Table A2 of the Internet Appendix shows the volumes of dividends and repurchases for each year and for the entire sample period for the baseline sample. Repurchases exceed dividends in all sample years except 2009 and amount to $2,170 billion in total over our sample period.

3.1.1. Methodology and variables Our generic specifications either regress a liquidity measure on Repurchase Intensity or Repurchase Intensity on a measure of stock market liquidity. In each case we include a range of controls: l= K

LIQ i, t = αL + δLLIQ i, t − 1 + βL RIi, t +

(1)

and 7

j=N

RIi, t = αR + δRRIi, t − 1 + βR LIQ i, t +

∑ γR, j Controli, j, t + μRi j=1

+ ηRt + uR, i, t ,

(2)

where LIQ is a liquidity measure, and RI is Repurchase Intensity, defined as the number of shares repurchased under a program during the month, divided by the number of shares outstanding at the beginning of the month. Sometimes we replace Repurchase Intensity with Repurchase Dummy, which assumes a value of one in a month in which the respective firm repurchased shares. Control refers to the control variables, μLi and μRi are time-invariant firm fixed effects, and ηLt and ηRt are month fixed effects. The month fixed effects account for changes in the macroeconomic environment not otherwise captured by the control variables. We sometimes omit monthly fixed effects in the repurchase regression, because one of our instruments for liquidity has no cross-sectional variation and would then be absorbed by the month fixed effects. We instead use year fixed effects and control for macroeconomic conditions. We use GMM-IV (Generalized Method of MomentsInstrumental Variables) to account for reverse causality and for unobserved common effects, which could create a correlation between the error terms uL, i, t and uR, i, t . The fixed effects methodology significantly raises the hurdle for finding significant relations between repurchases and liquidity. Our approach only relies on the within-firm variation in liquidity and repurchases that is firm-specific and is not driven by common macroeconomic effects. All timeinvariant cross-sectional differences among firms and all time-varying factors that are common to all firms are absorbed into the fixed effects. 3.1.2. Liquidity measures We use three different measures of stock market liquidity. The precise details of how these measures are calculated are provided in Appendix A.2 and we only summarize them here. For all measures, we first calculate a daily average and then the mean over all trading days within a particular month. Spread is based on all relative spreads for a given stock, weighted by the time the quote is valid. We calculate Price Impact as the absolute value of the change in quotes over a five-minute interval and the Amihud measure as the absolute daily return, divided by the daily trading volume in dollar. 3.2. Instrumental variables

∑ γL, l Controli, l, t + μLi l= 1

+ ηLt + uL, i, t

5

Stephens and Weisbach (1998) report average program completion rates of, respectively, 54.10%, 68.70%, and 73.80% one, two, and three years after the program announcement. Oded (2009) reports unconditional completion rates of 56.1% and 92.3%, respectively, four and eight quarters after the announcement. Bonaimé (2012) finds an average completion rate of 72.57% eight quarters after the quarter of the program announcement.

In this subsection, we develop the instruments for Repurchase Intensity (Subsection 3.2.1) and for the liquidity variables (Subsection 3.2.2). 3.2.1. Instruments for repurchases The instruments for Repurchase Intensity are Program Size and Program Month. We calculate Program Size as the maximal number of shares that can be purchased under a particular program, divided by the number of shares outstanding at the time of the announcement if the program volume is reported in shares. If there are multiple active

Please cite this article as: Hillert, A., et al., Stock repurchases and liquidity. Journal of Financial Economics (2015), http: //dx.doi.org/10.1016/j.jfineco.2015.08.009i

A. Hillert et al. / Journal of Financial Economics ∎ (∎∎∎∎) ∎∎∎–∎∎∎

6

Table 1 Description of variables. The table describes all control variables and repurchase variables. For each variable, the table reports the definition, the data source, and the unit of measurement. Variables denoted with (ln) are expressed as natural logarithms. Data sources include the Center for Research in Security Prices (CRSP), Comupustat, Daily Trade and Quote (TAQ), Institutional Brokers' Estimate System (I/B/E/S), Securities and Exchange Commission (SEC), Security Data Company (SDC), Thomson Reuters Insider Data (TR Insider Data), and Wharton Research Data Services (WRDS). Name Acquiror

Amihud Analysts Accelerated Share Repurchase Average Spread Book to Market Book Value Equity CAR (Six months) Cash to Assets Deviation from $30 Dividends to Assets EBITDA Exogenous Order Imbalance Exogenous Trading Volume Filing CAR Fund Trading Leverage Market Return

Market Cap Options Exercised Order Imbalance Price Price Impact Private Repurchase Program Month Program Size (scaled) Repurchase Volume Repurchase Dummy Repurchase Intensity Repurchase Intensity (TV) Return S&P 500 Shares Outstanding Short Interest ΔShort Interest Spread

Definition One if firm is currently (time between announcement and end of the offer) bidding for another company Monthly average of daily Amihud illiquidity ratio Number of analysts (ln) Repurchase via accelerated share repurchase Average of Spread from t  6 to t  1 Book Value Equity / Market Cap winsorized at 1% Common equity (Compustat item: ceqq) Cumulative abnormal return over six months following repurchase month Cash (item: cheq) scaled by assets (item: atq) Absolute difference between Price and $30 (ln) Dividends (item: dvc) scaled by assets (item: at) Operating income before depreciation (item: oibdpq) Median Order Imbalance in current month of all firms with no repurchase activity between 2004 and 2010 Median Trading Volume in current month of all firms with no repurchase activity between 2004 and 2010 (ln) Cumulative abnormal return over three days around the filing day of the quarterly report Sum of all mutual fund trades (buys and sells) calculated for each stock in each month (Total Assets  Book Value Equity) / (Total Assets  Book Value Equity þ Market Cap) One if value-weighted market return including dividends is below its median over all sample months Monthly average of daily market capitalization (ln) Number of shares obtained by option exercises of corporate insiders in the respective month divided by shares outstanding (Trade value buy  trade value sales)/ (trade value buy þ trade value sales) Monthly average of daily closing price (ln) Monthly average of intraday price impact, transaction based (ln) Repurchase via private transaction Difference between current month and start month of the repurchase program plus one (ln) Size of the repurchase program scaled by shares outstanding as of the beginning of the program Number of shares repurchased during the month One if repurchase transaction takes place Number of shares repurchased during the month divided by the number of shares outstanding at the last trading day of the previous month Number of shares repurchased during the month divided by the number of shares traded over the current month Monthly stock return One if firm is in the Standard & Poor's (S&P) 500 Number of shares outstanding at last trading day of month Shares sold short as of the 15th business day scaled by end of previous month shares outstanding First difference of Short Interest Monthly average of intraday relative spread,

Source

Unit

SDC

Binary

CRSP I/B/E/S SEC

Ratio Unit Binary

TAQ Compustat

Unit Ratio

Compustat CRSP

Million Unit

Compustat CRSP

Ratio Unit

Compustat

Ratio

Compustat

Million

CRSP

Unit

CRSP

Unit

CRSP

Unit

Morningstar

Unit

Compustat, CRSP CRSP

Ratio Binary

CRSP TR Insider Data TAQ

Million Ratio

CRSP TAQ

Unit Ratio

SEC SEC

Binary Unit

SEC

Ratio

SEC SEC SEC, CRSP

Million Binary Ratio

SEC, CRSP

Ratio

CRSP Compustat CRSP

Unit Binary Million

Compustat

Ratio

Compustat TAQ

Ratio Ratio

Ratio

Please cite this article as: Hillert, A., et al., Stock repurchases and liquidity. Journal of Financial Economics (2015), http: //dx.doi.org/10.1016/j.jfineco.2015.08.009i

A. Hillert et al. / Journal of Financial Economics ∎ (∎∎∎∎) ∎∎∎–∎∎∎

7

Table 1 (continued ) Name

Target

Tender Offer Total Assets Trading Volume Trading Volume (scaled) Volatility VIX

Definition time-weighted (ln) One if firm is currently (time between announcement and end of the offer) a target of another company Repurchase via tender offer or Dutch auction Total assets (Compustat item: atq) (ln) Monthly total dollar trading volume (ln) (Number of shares traded  number of shares repurchased)/number of shares outstanding Standard deviation of daily returns over one month (ln) One if Chicago Board Options Exchange (CBOE) S&P 500 Volatility Index (VIX) is above its median over all sample months

programs, Program Size and Program Month are both calculated from the most recently announced program. If the program volume is reported in dollars, we use the maximal dollar volume that can be repurchased under the program divided by the firm's market capitalization at the time of the announcement. The size of the program is fixed before the execution begins. It predicts future repurchases and is therefore exogenous with respect to future variations in liquidity. We can use neither the realized size of the program nor the unutilized portion of the program as instruments, because both depend on firms' actual repurchase behavior and could be related to the within-firm variation in liquidity. All our regressions include firm fixed effects, so unobserved factors that drive the design of repurchase programs and also influence subsequent expected liquidity would be included in these fixed effects. For example, a firm with a more liquid stock can (and typically does) have larger repurchase programs and also a more liquid market during the execution of the program. Such a relation is also reflected in the firm fixed effects and does not bias our results. For our instruments, it is only important that unobserved factors do not influence the within-firm variation in liquidity after the start of the program. Naturally, we expect Program Size to have a positive impact on repurchases. Average (median) program size in our sample is 6.59% (5.27%) of the shares outstanding. Program Month is the number of calendar months since the announcement of the repurchase program. The motivation is that the period for which the program has been active is also an ex ante feature of the program that is not influenced by the subsequent within-firm variation of liquidity. However, we do not include observations of the program beyond one year if the program has been active for more than one year, because a disappointing development in a firm's liquidity could lead managers to extend the length of the program beyond one year, which would render Program Month endogenous. Hence, Program Month is simply a number between one and 12. We expect firms to front-load the execution of their programs. As a result, Program Month should have a negative impact on realized repurchases. The correlation of Repurchase Intensity and Program Month is  0.0929. Effectively, Program Month predicts repurchases based on the stage in the execution of

Source

Unit

SDC

Binary

SEC Compustat CRSP CRSP, SEC CRSP

Binary Million Million Ratio

CBOE WRDS

Binary

Unit

the repurchase program. If actual repurchases deviate from predicted repurchases, then these deviations are not used for identifying our effects and do not bias our estimates, even if the firm does not repurchase any shares in a certain month. Both Program Size and Program Month could reflect liquidity conditions at the time of the program announcement. However, to the extent that these conditions are firm-specific and predict liquidity during the later stages of the program, they would be absorbed into firm fixed effects. 3.2.2. Instruments for liquidity We use two instruments for liquidity in our baseline analysis. Exogenous Trading Volume is calculated as the median trading volume of all firms that never conduct any repurchase during the period from 2004 to 2010. The motivation for this instrument is that liquidity depends on numerous unobservable factors that are common to many firms. We require that the instrument incorporates these factors but is not influenced by the specific factors that would influence the liquidity of a particular firm, because these factors could then also influence the firm's repurchase behavior directly. It seems highly implausible that the specific circumstances of the repurchase programs of some individual firm affect the stock market liquidity of sufficiently many non-repurchasing firms to affect Exogenous Trading Volume. Exogenous Trading Volume has no cross-sectional variation. We can, therefore, not use it and also estimate month fixed effects and we estimate only year fixed effects instead in those regressions that use this instrument. To make sure that Exogenous Trading Volume does not capture within-year macroeconomic fluctuations that could also have a direct impact on repurchases, we use four controls for macroeconomic factors suggested by Baker and Wurgler (2006) to remove within-year macroeconomic fluctuations: growth in the industrial production index, growth in consumer durables, nondurables, and services. Alternatively, we use Lagged Trading Volume, which is the previous month's trading volume of the same firm and therefore has cross-sectional variation and allows us to also include month fixed effects. Here the identifying assumption is that last month's trading volume has no

Please cite this article as: Hillert, A., et al., Stock repurchases and liquidity. Journal of Financial Economics (2015), http: //dx.doi.org/10.1016/j.jfineco.2015.08.009i

A. Hillert et al. / Journal of Financial Economics ∎ (∎∎∎∎) ∎∎∎–∎∎∎

8

impact on repurchases other than through its ability to predict current liquidity. We are not aware of theoretical arguments identifying a variable that simultaneously affects liquidity and also affects repurchases through channels other than the liquidity of the market itself. The second instrument for liquidity is Deviation from $30, which is defined as the absolute value of the difference between the stock price and $30. This instrument has been used successfully by Choi, Getmansky, and Tookes (2009) and is calculated as the log of the absolute price deviation from $30, where Price is the monthly average of the daily closing prices from CRSP. The motivation of this instrument, as in Choi, Getmansky, and Tookes (2009), is the notion of an optimal trading range around $30. The further a stock trades away from this assumed center of the trading range, in either direction, the less it is traded. Some stocks trade far away from $30 and yet are very liquid. Because we always include firm fixed effects, such cross-sectional variations can be ignored. The instrument is related only to the within-firm variation of liquidity and relies on the assumption that liquidity improves for stocks as firms move closer to the optimal trading range. We always conduct several tests to validate our instruments. We report the Hansen J statistic on the overidentifying restrictions. We test for underidentification by

using the statistic proposed by Kleibergen and Paap (2006). Their test is for the rank of a matrix. In our case, it checks the rank of the matrix of reduced-form coefficients and tests whether the instruments are sufficient to identify the endogenous variables. Finally, we test for weak instruments by using the weak-identification test of Stock and Yogo (2005), which tests the hypothesis that the maximal bias of the IV estimator does not exceed a certain threshold, expressed as a percentage of the corresponding OLS estimator. In all of our specifications, the test statistic is above the critical value, and the test rejects the hypothesis that the maximal bias according to this test is above 5% of the OLS estimator. Given the uniformity of the results, we do not report this test in the tables. However, we do report the t-statistics from the first stage regression to provide further information on the strengths of the instruments. 3.3. Descriptive statistics Table 2 describes the sample restricted to active programs, which we use for testing our main hypotheses in Sections 4 and 5. This sample covers 106,753 firm-months except for the repurchase volume and Repurchase Intensity, which are reported only for the months in which repurchases are positive (50,204 firm-months). The Spread is

Table 2 Descriptive statistics: active programs. This table provides descriptive statistics for the firm months with active repurchase programs over the sample period from January 2004 to December 2010. The sample contains all ordinary shares (share codes 10 and 11) that are listed on the NYSE, Amex, and Nasdaq (exchange codes 1, 2, and 3). Appendix A.2 provides definitions of the liquidity measures. The repurchase variables and the control variables are defined in Table 1. We report the arithmetic mean, the median, the standard deviation, the 1st percentile, and the 99th percentile of the distribution for each variable. None of the variables is expressed in natural logarithms unless otherwise stated.

Liquidity measures Spread Price Impact Amihud Repurchase measures Repurchase Volume (millions) Repurchase Intensity Repurchase Intensity (TV) Control variables and instruments Analysts Accelerated Share Repurchase Book to Market Deviation from $30 Exogenous Trading Volume Leverage Market Cap (millions) Options Exercised Price Private Repurchase S&P 500 Short Interest ΔShort Interest Tender Offer Total Assets (millions) Trading Volume (millions) Trading Volume (scaled) Volatility Program descriptives Program Size (scaled) Program Month

Mean

Median

Standard Deviation

1st Percentile

99th Percentile

N

0.71% 1.88% 3.60

0.14% 0.73% 0.00

1.74% 3.50% 96.00

0.02% 0.18% 0.00

9.14% 17.03% 40.80

106,753 106,753 106,753

49.4 0.66% 6.74%

4.6 0.35% 3.29%

180.8 0.98% 9.91%

0 0.00% 0.00%

768.5 4.51% 52.13%

50,204 50,204 50,204

7.65 0.01 0.68 18.26 16.61 0.45 6,074 0.00 27.57 0.00 0.21 4.96% 0.01% 0.00 14,463 1,094 0.19 0.03

6.00 0.00 0.54 15.58 16.55 0.39 807 0.00 21.41 0.00 0.00 3.21% 0.00% 0.00 1,171 145 0.14 0.02

7.09 0.25 0.59 34.77 6.27 0.29 21,158 0.00 39.20 0.07 0.41 5.42% 1.45% 0.12 96,163 3,284 0.23 0.02

0.00 0.00 -0.10 0.33 3.73 0.03 12 0.00 0.86 0.00 0.00 0.00%  4.06% 0.00 21 0 0.00 0.01

29.00 0.00 3.50 74.73 33.78 0.98 106,925 0.01 104.73 0.00 1.00 26.77% 4.01% 0.00 220,005 15,047 0.98 0.11

106,753 106,753 106,753 106,753 106,753 106,753 106,753 106,753 106,753 106,753 106,753 106,753 106,753 106,753 106,753 106,753 106,753 106,753

6.59% 16

5.27% 12

4.86% 14

0.47% 1

25.11% 67

6,537 6,537

Please cite this article as: Hillert, A., et al., Stock repurchases and liquidity. Journal of Financial Economics (2015), http: //dx.doi.org/10.1016/j.jfineco.2015.08.009i

A. Hillert et al. / Journal of Financial Economics ∎ (∎∎∎∎) ∎∎∎–∎∎∎

on average (median) 0.71% (0.14%). The average Repurchase Volume over 50,204 repurchase months is $49.4 million, which is equivalent to buying back 0.66% of shares outstanding or 6.74% of monthly trading volume. Ben-Rephael, Oded, and Wohl (2014) study a sample of CRSP firms randomly drawn from all NYSE size deciles, and their descriptive statistics are similar to ours (see their Table 1). The median repurchase volume is 0.35% of shares outstanding or 3.29% of trading volume. Table A3 of the Internet Appendix provides descriptive statistics for all variables of the entire sample, which covers 346,978 firm-months and is used for the eventstudy analysis in Section 6.8 To be protected by the Safe Harbor Rule 10b-18, firms' daily repurchases must not exceed 25% of the average daily trading volume (ADTV) of the preceding four calendar weeks. However, once a week firms can purchase shares in a block trade that is not subject to the 25% restriction. Because repurchases are reported only monthly, we are not able to assess in detail whether repurchasing firms comply with the volume condition of the Safe Harbor Rule. Instead, we approximate ADTV from last month's trading volume. In 3,019 of the 50,204 repurchase-months firms purchase more than 25% of last month's trading volume. Because we cannot observe block repurchases, which do not count toward the 25% limit, the true number of months in which firms exceed the safe harbor limit is lower than 3,019.

4. The influence of repurchases on liquidity We begin the analysis by testing for the overall impact of repurchases on liquidity. We expect firms to have a long time horizon regarding repurchases for two reasons. First, most repurchase programs in our sample have a time horizon of 12 months or longer, and many of them do not have a fixed expiry date at all, which indicates that firms take a long-term view with respect to repurchases. Second, repurchase programs do not commit firms, which often let repurchase programs expire without completing them (see Subsection 3.1 for details). Hence, firms should be patient traders, predominantly supplying liquidity to the market, and we hypothesize that repurchases improve liquidity. This hypothesis is similar to the competing market-maker hypothesis of Barclay and Smith (1988) but relies on a somewhat different framework. It does not assume that firms are uninformed (see Section 2). We restrict our baseline analysis to the sample of firmmonths with active repurchase programs. Our discussion of repurchase initiations in the Internet Appendix reveals only a negligible impact of repurchase announcements on liquidity. Therefore, any significant impact of repurchases on liquidity has to come from the execution of repurchase programs. Table 3 reports the results. The dependent variables in Regressions 1 to 3 are the three main liquidity measures: the Spread, Price Impact, and Amihud (illiquidity measure). In 8

The number of observations for cumulative abnormal returns is lower because it requires more data. The number of observations on Filing CAR depends on the number of filings.

9

addition, we include trading volume (Turnover) as a dependent variable in Regressions 4 and 5. Because repurchases can mechanically increase trading volume, we subtract the repurchase volume from total trading volume in Regression (5). We include firm fixed effects and therefore require instruments to vary within firms. We use Program Month and Program Size as instruments, which both vary over time. All regressions are correctly specified. The Hansen J statistic for the test of overidentifying restrictions cannot reject the null that the instruments are valid for any specification in Table 3. The Kleibergen-Paap test for underidentification always rejects the null of underidentification at all conventional significance levels. First stage t-statistics indicate that none of our instruments is weak. The signs of the instruments are as predicted: Repurchase Intensity increases with Program Size and declines with Program Month. The Stock-Yogo test on the weak-instrument bias always rejects the hypothesis that the bias exceeds 5% of the bias from OLS (not tabulated). We find negative coefficients on repurchases in all regressions and they are statistically significant at the 1% level for all liquidity measures except Price Impact, which is significant at the 5% level. A coefficient of 7.487 on Repurchase Intensity in the regression for Spread implies that a change in Repurchase Intensity by 0.61% (its within-firm standard deviation) changes Spread by 4.57% (¼7.487  0.61%; we report all time series standard deviations in Table A9 in the Internet Appendix). Hence, there is a clear and unequivocally positive impact of repurchases on stock market liquidity. Columns 4 and 5 show that repurchases also increase trading volume. Column 5 defines trading volume net of repurchases, which reduces the coefficient on Repurchase Intensity by about 30%. Hence, the impact of repurchases on volume is not just a mechanical effect. Most likely, increased repurchases reduces spreads and therefore transaction costs, which in turn induces more trading in the stock among investors other than the firm and thereby increases volume. This finding is critical for our argument that trading volume is endogenous and can therefore not serve as an exogenous control variable. In Table A4 in the Internet Appendix, we impose a stricter requirement and limit the analysis to those repurchase programs with a definite expiry date to address the potential concern that firms might not specify expiry dates when they want to have flexibility over program duration. In Table A5 in the Internet Appendix, we exclude all programs that are completed within less than 12 months. Thereby, we exclude programs that complete faster, because the market for the stock could have been more liquid in the past. The sample restrictions in Tables A4 and A5 have no consequences for our results, which supports our baseline specification, presumably because our instruments already address these issues. Repurchase Intensity predicted by Program Month and Program Size cannot be correlated with subsequent shocks to liquidity. Overall, we conclude that repurchases improve liquidity. 4.1. Controls Our regressions control for size using the logarithm of the book value of assets, which implies that larger firms are more liquid, confirmimg earlier findings of Stoll (2000) and Chung, Elder, and Kim (2010). Lagging total assets does not change

Please cite this article as: Hillert, A., et al., Stock repurchases and liquidity. Journal of Financial Economics (2015), http: //dx.doi.org/10.1016/j.jfineco.2015.08.009i

10

A. Hillert et al. / Journal of Financial Economics ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Table 3 The influence of repurchases on liquidity: generalized method of moments (GMM). The table presents GMM regressions of liquidity on Repurchase Intensity and control variables. We use Program Size and Program Month as instruments for Repurchase Intensity. Repurchase Intensityt is an instrumented variable. The sample is restricted to the first 12 months of a repurchase program. Appendix A.2 provides definitions of the liquidity measures. The repurchase variables and the control variables are defined in Table 1. Standard errors are clustered at the firm level. t-Statistics are provided in parentheses. n, nn, and nnn indicate significance at the 10%, 5%, and 1% level, respectively. The Hansen J statistic tests for the validity of the overidentifying restrictions. The Kleibergen-Paap test is for underidentification and tests for the full rank of the reduced-form coefficient matrix following Kleibergen and Paap (2006). The table reports the test statistics and the p-values for both tests. The t-tests for the instruments are from the first stage regressions. The test suggested by Stock and Yogo (2005) rejects the hypothesis that the bias exceeds the ordinary least squares bias by more than 5% in all cases.

Spread

Price Impact

(1)

(2) nnn

 7.4870 (  7.85) Deviation from $30t 0.0160nnn (8.31) Returnt  1  0.1155nnn (  9.44) Volatilityt  1 0.0187nnn (4.12) Trading Volume excluding  0.0800nnn Repurchaset  1 (scaled) (  4.26) Total Assetst  0.0900nnn (  8.13) S&P 500t  0.0749nnn (  3.87) Pricet  0.1691nnn (  18.13) Analystst  0.0254nnn (  4.18) Book to Markett  3 0.0496nnn (5.65) Leveraget  3  0.0476 (  1.32) Accelerated Share Repurchaset  0.0091nnn (  2.99) Private Repurchaset  0.0052 (  0.70) Tender Offert  0.0068 (  0.99) Dependent variablet  1 0.6520nnn (95.05) Number of observations 59,558 Firm fixed effects Yes Time fixed effects Month Hansen J (test) 0.88 Hansen J (p-value) 34.83% Kleibergen-Paap (test) 405.0 Kleibergen-Paap (p-value) 0.00% First stage t-Statistics of included instruments Program Month (ln)  23.01 Program Size 7.15 Repurchase Intensityt

Dependent variable Amihud

(3) nn

Trading Volume including repurchases (4)

Trading Volume excluding repurchases (5)

 3.4352 (  2.42) 0.0068nn (2.18)  0.0965nnn (  5.79)  0.0009 (  0.13)  0.1125nnn (  4.44)  0.0327nn (  2.03)  0.0730nn (  2.03)  0.1130nnn (  8.40)  0.0309nnn (  3.19) 0.0605nnn (4.91)  0.0347 (  0.65)  0.0079n (  1.76)  0.0041 (  0.33)  0.0162n (  1.90) 0.5900nnn (52.18) 59,255 Yes Month 2.14 14.31% 407.2 0.00%

 17.1465nnn (  7.32) 0.0204nnn (4.89)  0.3371nnn (  9.76)  0.0413nn (  2.47)  0.4403nnn (  4.95)  0.4403nnn (  11.87)  0.1391nnn (  3.17)  0.6889nnn (  20.55)  0.0860nnn (  4.62) 0.2074nnn (7.00) 0.2817nn (2.49)  0.0224nnn (  4.82)  0.0201 (  0.97)  0.0589nnn (  5.30) 0.3809nnn (23.74) 59,611 Yes Month 0.25 61.71% 404.3 0.00%

2.3461nnn (4.78)  0.0031nn (  2.36)  0.0272nnn (  3.47)  0.0021 (  0.34)

1.6727nnn (3.48)  0.0031nn (  2.37)  0.0253nnn (  3.18)  0.0029 (  0.46)

0.0216nnn (3.16) 0.0130 (1.16) 0.0100 (1.41) 0.0100nnn (3.23)  0.0167nnn (  2.86) 0.0605nn (2.54) 0.0035nnn (2.98) 0.0081 (1.30) 0.0063nn (2.51) 0.4266nnn (6.38) 59,756 Yes Month 0.06 81.10% 410.7 0.00%

0.0214nnn (3.11) 0.0128 (1.15) 0.0095 (1.34) 0.0100nnn (3.26)  0.0167nnn (  2.87) 0.0598nn (2.52) 0.0036nnn (3.10) 0.0082 (1.31) 0.0063nn (2.47) 0.4304nnn (6.36) 59,574 Yes Month 0.04 83.50% 410.6 0.00%

 23.10 7.21

 23.00 7.24

 23.20 7.25

 23.21 7.26

our estimation results. We do not control for market capitalization, because our fixed effects model already removes average size and we control for total assets, stock price, and lagged returns. Adding market capitalization as a control variable does not change our results. Firms in the Standard & Poor's (S&P) 500 and with higher levels of the stock price also tend to be more liquid. In line with Chung, Elder, and Kim (2010), we control for the number of analysts. To avoid a high number of missing observations, we assume that a firm has no analyst coverage if there is no information about the firm on I/B/E/S. We find that firms are more liquid if more analysts

follow the stock. Chung, Elder, and Kim (2010) conjecture the same relation, but find the opposite result. Firms are consistently less liquid when their book-tomarket ratio is high. Leverage is not significantly related to spread and price impact. Firms with high leverage are less liquid based on the Amihud illiquidity measure but have higher trading volume. Leverage typically does not change significantly over a seven-year period, so the impact of leverage has probably no meaningful interpretation. We also control for transaction characteristics. We include dummies for accelerated share repurchases (ASRs),

Please cite this article as: Hillert, A., et al., Stock repurchases and liquidity. Journal of Financial Economics (2015), http: //dx.doi.org/10.1016/j.jfineco.2015.08.009i

A. Hillert et al. / Journal of Financial Economics ∎ (∎∎∎∎) ∎∎∎–∎∎∎

11

Table 4 Liquidity-consuming repurchases. The table presents generalized method of moments (GMM) regressions of liquidity on Repurchase Intensity and control variables. High Cash is a dummy variable equal to one if Cash to Assets at the beginning of the program is above the median over all programs. End of Program is a dummy variable equal to one if the number of remaining program months is three or fewer. We use Program Size and Program Month as instruments for Repurchase Intensity. Our instruments are also interacted with High Cash and (1  High Cash) or End of Program and (1  End of Program). Thus, every analysis makes use of four instruments. Instrumented variables are Repurchase Intensityt  (1  High Casht), Repurchase Intensityt  High Casht, Repurchase Intensityt  (1  End of Programt), and Repurchase Intensityt  End of Programt. The sample is restricted to the first 12 months of a repurchase program. We include the same control variables as in Table 3 but omit them from the table for readability. Standard errors are clustered at the firm level. t-Statistics are provided in parentheses. n, nn, and nnn indicate significance at the 10%, 5%, and 1% level, respectively. The Hansen J statistic tests for the validity of the overidentifying restrictions. The Kleibergen-Paap test is for underidentification and tests for the full rank of the reduced-form coefficient matrix following Kleibergen and Paap (2006). The table reports the test statistics and the p-values for both tests. The t-tests for the instruments are from the first stage regressions. The test suggested by Stock and Yogo (2005) rejects the hypothesis that the bias exceeds the ordinary least squares bias by more than 5% in all cases.

Spread

Price Impact (2)

(1) Repurchase Intensityt  (1  High Casht) Repurchase Intensityt  High Casht High Casht

nnn

 10.5858 (  6.11)  4.8773nnn (  4.66)  0.0114 (  1.11)

nnn

 7.3323 (  2.98)  0.6699 (  0.42)  0.0351nn (  2.19)

Dependent variable Amihud Spread (3)

59,558 0.536 76.50% 313.281 0.00% 9.12 0.25% Yes Month

59,255 1.212 54.56% 322.486 0.00% 5.96 1.46% Yes Month

repurchases conducted as tender offers, and privately targeted share repurchases.9 For all three liquidity measures, we obtain a negative and significant coefficient of the ASR dummy, indicating that accelerated share repurchases improve liquidity. Tender offers are also associated with improvements in liquidity. However, the coefficient is significant only for Price Impact and Amihud. ASRs and tender offers are both associated with higher trading volume. The stock return over the previous month has a highly significant and positive impact on stock liquidity. We also control for three lagged variables. The autoregressive coefficient is highly significant and large. We include the lags of stock volatility and trading volume, which have been used as controls for the spread in the previous literature (see Stoll, 2000; Brockman and Chung, 2001; Ginglinger and Hamon, 2007). Trading Volume is scaled by the number of shares outstanding and defined excluding share repurchases to avoid contaminating the repurchase variable itself with the influence from trading volume. The coefficients on Volatility change signs across 9 Accelerated share repurchases (ASRs) involve a contract with an intermediary who borrows the shares, delivers them to the firm, and subsequently covers its short position through repurchases in the open market. See Bargeron, Kulchania, and Thomas (2011) for a more detailed discussion of ASRs.

 7.8394nnn (  3.46)  7.6630 (  1.15)  0.0006 (  0.05) 20,369 0.559 75.63% 97.864 0.00% 0.09 76.42% Yes Month

 4.0482 (  1.17)  22.1324nn (  2.49) 0.0461nnn (2.82) 20,258 0.659 71.91% 98.574 0.00% 4.34 3.72% Yes Month

 18.1738nnn (  3.45) 3.9506 (0.24)  0.0524n (  1.83) 20,383 0.146 92.96% 98.528 0.00% 1.18 27.74% Yes Month

(6)

nnn

 24.5558 (  5.69)  11.9217nnn (  4.72)  0.0026 (  0.10)

Repurchase Intensityt  (1  End of Programt) Repurchase Intensityt  End of Programt End of Programt Number of observations Hansen J (test) Hansen J (p-value) Kleibergen-Paap (test) Kleibergen-Paap (p-value) Wald test (high–low) Wald test (p-value) Firm fixed effects Time fixed effects

Amihud

(4)

Price Impact (5)

59,611 2.537 28.13% 312.977 0.00% 7.12 0.76% Yes Month

liquidity measures and are difficult to interpret. We attribute these patterns to the fact that we use lagged volatility. We obtain the usual positive coefficient if we control for contemporaneous volatility. The directions of causality between volatility, liquidity, and trading volume are not clear. These variables are probably determined simultaneously and are omitted from the standard specifications. 4.2. Tests for liquidity-reducing repurchases The result from Table 3 that repurchases improve liquidity leaves open the possibility that some repurchases reduce liquidity. In the context of our framework, we would expect that firms demand liquidity whenever they are under time pressure to execute transactions and when completing a repurchase program is important to them. If repurchases primarily disburse cash, then firms with large cash-to-asset ratios should find a greater need to pay out the excess cash and complete repurchase programs. Hence, we expect repurchases of firms with higher cash-to-asset ratios to reduce liquidity. We test this in Table 4 by interacting Repurchase Intensity with the dummy variable High Cash, which assumes a value of one if the repurchase program is associated with an above-median cash-to-asset ratio at the beginning of the program. We use a setup in which we can distinguish between the effect of repurchases for high-cash

Please cite this article as: Hillert, A., et al., Stock repurchases and liquidity. Journal of Financial Economics (2015), http: //dx.doi.org/10.1016/j.jfineco.2015.08.009i

12

A. Hillert et al. / Journal of Financial Economics ∎ (∎∎∎∎) ∎∎∎–∎∎∎

firms (Repurchase Intensity  High Cash) and the effect of repurchases for low-cash firms [(Repurchase Intensity  (1 High Cash)]. We choose a specification where we interact Repurchase Intensity separately with High Cash and 1  HighCash to be able to instrument for these variables. If highcash firms are under pressure to complete the repurchase program, then the coefficient on the interaction of Repurchase Intensity and High Cash should be positive, indicating that these repurchases reduce liquidity. Columns 1 to 3 of Table 4 report the results. The control variables are the same as in Table 3, but they are omitted from the table. The models in Table 4 pass all specification tests. We do not find positive coefficients on the interaction of Repurchase Intensity and High Cash. The coefficient remains negative in all cases and statistically significant for Spread and Amihud. However, we do observe that the interactive term with High Cash is significantly larger, i.e., smaller in absolute value, than the coefficient on the interaction of Repurchase Intensity and (1  High Cash) alone. At the bottom of Table 4, we report Wald tests for the significance of this difference and find that it is always significant with p-values of 1.5% or less. Hence, repurchases of cash-rich firms supply less liquidity to the stock market compared with repurchases of other firms. In the Internet Appendix, in Table A6, we report additional results, in which the dummy variable High Cash is defined with respect to the 75th, 90th, and 95th percentile. It could be that only a small fraction of firms with very high cashto-asset ratios consumes liquidity and the results in Table 4 could hide the liquidity-reducing repurchases if we split the sample at the median. However, we find that the qualitative patterns are similar for higher cutoffs and the statistical significance is lost, probably because the number of observations above the cutoff becomes too small. Overall, we find no evidence for repurchases that reduce liquidity based on cash-to-asset ratios. Firms can also become less patient if they approach the end of the repurchase program. We define a dummy variable End of Program to indicate those program months for which the repurchase program is within three months of the expiry of the program and perform interactive regressions in the same way as for cash. In this case, we would expect a positive coefficient on the interaction of Repurchase Intensity and End of Program if firms demand liquidity toward the end of the repurchase program. Columns 4 to 6 of Table 4 show the results. In no case do we observe the predicted sign reversal and the difference between the coefficient on Repurchase Intensity and the coefficient on the interactive term is not statistically significant for Spread and Amihud, as indicated by the Wald test. If we repeat this exercise with more stringent definitions of End of Program, in which the program is within one month or two months of expiry, the same result obtains (see the Internet Appendix, Table A7). This result is not surprising, because repurchase programs do not imply any commitment on the part of the firm. Instead of increasing their repurchase intensity, firms can simply let a part of the program expire uncompleted and then start a new program. Hence, being close to the expiry of a repurchase program does not turn firms into impatient traders with a high demand for immediacy.

Overall, we cannot identify observable characteristics of firms or repurchase programs that would indicate situations in which they demand liquidity when repurchasing shares. While we did not anticipate this outcome, it is consistent with evidence on the liquidity supplied by hedge funds. During activist campaigns, hedge funds purchase proportions of the outstanding shares that are about the same magnitude as a repurchase program. According to Gantchev and Jotikasthira (2014), hedge funds purchase on average 4.25% of the outstanding shares during the 60day period reported in their 13D-filing until the filing date, which compares with 3.93% of the shares outstanding firms repurchase during the first 12 months of a repurchase program. Here, the value for repurchases of 3.93% is calculated as the sum of the monthly average of Repurchase Intensity over 12 months. Unlike firms, hedge funds do not disclose their trades before they have to file, which allows them to conceal their intentions. Compared with firms, this probably forces hedge funds to trade more shares in a shorter period of time, which should increase their demand for liquidity. Yet, hedge funds seem to generally supply liquidity (Franzoni and Plazzi, 2013; Jylhä, Rinne, and Suominen, 2014). However, this evidence is only indicative, because we are not aware of any evidence on the liquidity impact of hedge fund trading specifically in the period before 13D-filings. 4.3. GMM versus OLS We compare our IV analysis with the corresponding OLS results, which we report in Columns 1 to 3 of Table 5. The specification is identical to that in Table 3, except that we do not use instrumental variables. The bias is positive. The coefficient estimates are all negative, but only about 30% in absolute value of the corresponding GMM-IV coefficient. We show later that the reverse relation is also negative, i.e., higher spreads (price impact, Amihud illiquidity) reduce Repurchase Intensity. Hence, reverse causality would result in a negative bias. Because we find a positive bias, OLS must suffer from the presence of unobserved factors that move spreads and repurchases in the same direction. We provide more evidence on the likely characteristics of the omitted variables after discussing the reverse relation between repurchases and liquidity. 4.4. Contemporaneous control variables For better comparison, Table 5 reproduces the conventional specification in the literature with contemporaneous volatility and contemporaneous trading volume as controls. These controls have a long tradition in microstructure research going back to Demsetz (1968).10 Following Brockman and Chung (2001), they were included in all contributions on the repurchase-liquidity relation. (See Footnotes 1 and 2 for the studies on repurchases and liquidity.) However, volatility and trading volume are both endogenous to the trading process, so an analysis 10 See also Tinic (1972), Tinic and West (1974), and Stoll (1978). Stoll's 1999 presidential address (Stoll, 2000) canonized this specification.

Please cite this article as: Hillert, A., et al., Stock repurchases and liquidity. Journal of Financial Economics (2015), http: //dx.doi.org/10.1016/j.jfineco.2015.08.009i

A. Hillert et al. / Journal of Financial Economics ∎ (∎∎∎∎) ∎∎∎–∎∎∎

13

Table 5 The influence of repurchases on liquidity: alternative ordinary least squares (OLS) specifications. The table presents OLS regressions of liquidity on Repurchase Intensity and control variables. In Columns 1 to 3, we include the same controls as in Table 3. In Columns 4 to 6, we measure volatility and trading volume contemporaneously and use dollar trading volume instead of trading volume scaled by shares outstanding. The remaining controls are identical to the ones in Table 3. The sample is restricted to the first 12 months of a repurchase program. Appendix A.2 provides definitions of the liquidity measures. The repurchase variables and the control variables are defined in Table 1. Standard errors are clustered at the firm level. t-Statistics are provided in parentheses. n, nn, and nnn indicate significance at the 10%, 5%, and 1% level, respectively, using t-tests.

Spread (1) Repurchase Intensityt Returnt  1 Volatilityt  1

nnn

 2.4736 (  16.44)  0.0995nnn (  8.68) 0.0166nnn (3.78)

Price Impact (2) nnn

 1.5324 (  7.79)  0.0859nnn (  5.45) 0.0019 (0.27)

Dependent variable Amihud Spread (3)  6.0063 (  17.24)  0.2966nnn (  8.99)  0.0562nnn (  3.39)

 0.0889nnn (  4.38)

 0.1195nnn (  4.56)

 0.4427nnn (  4.99)

 0.1610nnn (  18.35) 0.6709 62,237 Yes Month

 0.1124nnn (  8.50) 0.5386 61,903 Yes Month

 0.6691nnn (  20.49) 0.5643 62,253 Yes Month

Trading Volumet (ln) Pricet R2 Number of observations Firm fixed effects Time fixed effects

including these controls can establish only correlations and several earlier papers were careful to emphasize this limitation. Because we wish to make causal statements, we do not include the above controls. Instead, we view trading volume as another aspect of market liquidity, which is determined simultaneously with volatility and the other liquidity measures we use above. Our specification differs from that in the literature in more than one dimension. We lag trading volume, use a definition of trading volume net of the volume from share repurchases, and scale by the number of shares outstanding. In untabulated results, we show that using lags of trading volume as defined in Table 5 also leads to negative coefficient estimates. Hence, we focus on the question of whether trading volume is contemporaneous or lagged. We reproduce the standard specification with contemporaneous controls in Columns 4 to 6 of Table 5 to illuminate why some previous papers have found a negative relation between repurchases and liquidity. Our results in Table 4 are unequivocally positive.11 With this specification, the coefficient of interest is always positive. Spread, Price Impact, and Amihud all increase with repurchases in this specification, and the relation is statistically significant for Spread and Amihud. Columns 4 to 6 differ from Columns 1 to 3 also because they do not include firm and time fixed effects. Including these fixed effects reduces the size of the coefficients on Repurchase Intensity, but the

Amihud

0.3196 (1.13)

3.3422nnn (10.11)

0.3827nnn (106.77)

0.5957nnn (123.75)

0.9238nnn (200.59)

 0.4851nnn (  472.27) 0.0772nnn (23.94) 0.9151 62,237 No No

 0.2497nnn (  174.57) 0.0338nnn (8.06) 0.6468 61,903 No No

 1.1336nnn (  709.28) 0.0418nnn (11.47) 0.9761 62,253 No No

(4) nnn

Volatilityt Trading Volumet  1 (scaled)

Price Impact (5)

nn

0.4996 (2.06)

(6)

coefficient remains positive for Spread and positive and significant for Amihud. The specification in the literature is the one reported in the table. This finding is important, because it could explain the inconsistent results found in the literature on the liquidityrepurchase relation. If we include volatility and trading volume, we ultimately control for the effect we want to measure and bias our results. The main qualitative difference between our results and those in some of the prior literature, therefore, do not result from reverse causality or omitted variables, but from using different control variables.

5. Trading strategies and the influence of liquidity on repurchases In this section, we discuss firms' repurchase behavior and relate it to their characteristics as traders in their own stock in terms of the framework developed in Section 2. In Subsection 5.1, we argue that firms minimize their transaction costs when trading in their own stock, to avoid a positive price impact. In Subsection 5.2, we argue that firms should take a more active stance when they observe that other investors sell the stock and exert downward price pressure, and then they should start to actively supply liquidity. 5.1. Minimizing transaction costs

11

See Barclay and Smith (1988), Brockman and Chung (2001), and Ginglinger and Hamon (2007) for studies that find a harmful effect of repurchases on liquidity.

Two hypotheses follow from the premise that firms try to minimize their trading costs when buying back their

Please cite this article as: Hillert, A., et al., Stock repurchases and liquidity. Journal of Financial Economics (2015), http: //dx.doi.org/10.1016/j.jfineco.2015.08.009i

A. Hillert et al. / Journal of Financial Economics ∎ (∎∎∎∎) ∎∎∎–∎∎∎

14

own stock. First, we follow Brockman, Howe, and Mortal (2008) and argue that managers try to minimize the costs of executing stock repurchase programs and avoid repurchasing shares when the market for their firms' stock is illiquid. Managers can observe the liquidity of their own stock as well as that of other stocks, and they can adapt their repurchase strategies to the current observed liquidity. Such a timing ability is different from market timing, which would require the ability to forecast future stock

prices. Liquidity timing has been observed for hedge funds (Cao, Chen, Liang, and Lo, 2013) and activist shareholders (Norli, Ostergaard, and Schindele, 2015). Busse (1999) makes a similar observation for mutual funds' ability to time the volatility of their portfolio stocks. This result is related, because volatility and liquidity are closely linked. Second, we observe that firms' objective is very similar to that of investors who want to liquidate large blocks, a problem that has been the focus of a large literature on

Table 6 The influence of liquidity on repurchases: generalized method of moments (GMM). The table presents GMM regressions of Repurchase Intensity on three different liquidity measures as specified in the table heading and control variables. All variable definitions can be found in Appendix A.2 and Table 1. Columns 1 to 3 use Exogenous Trading Volume and Deviation from $30 as instruments for liquidity and control for year fixed effects. Columns 4 to 6 use lagged Trading Volume and Deviation from $30 as instruments for liquidity and control for month fixed effects. Liquidityt is an instrumented variable. The sample is restricted to the first 12 months of a repurchase program. Standard errors are clustered at the firm level. t-Statistics are provided in parentheses.n, nn, and nnn indicate significance at the 10%, 5%, and 1% level, respectively. The Hansen J statistic tests for the validity of the overidentifying restrictions. The Kleibergen-Paap test is for underidentification and tests for the full rank of the reduced-form coefficient matrix following Kleibergen and Paap (2006). The table reports the test statistics and the p-values for both tests. The t-tests for the instruments are from the first stage regressions. The test suggested by Stock and Yogo (2005) rejects the hypothesis that the bias exceeds the ordinary least squares bias by more than 5% in all cases.

Spread Liquidity measure

(1)

 0.0028nnn (  3.44) Repurchase Intensityt  1 0.1189nnn (10.56) Returnt  1  0.0063nnn (  10.62) Returnt  2  0.0037nnn (  7.58) Returnt  3  0.0010nnn (  3.09) Total Assetst  3 0.0005 (0.87) Cash to Assetst  3 0.0034nnn (3.82) EBITDA to Assetst  3  0.0049nn (  2.21) Dividends to Assetst  3 0.0035 (0.76) Leveraget  3  0.0054nnn (  3.89) Options Exercisedt 0.0457nn (2.52) Book to Markett  3 0.0021nnn (4.73) Acquiror Dummyt  0.0003nn (  2.09) Target Dummyt  0.0005 (  0.72) Program Montht  0.0016nnn (  20.08) Program Sizet 0.0153nnn (7.40) Number of observations 51,652 Macro variables Yes Firm fixed effects Yes Time fixed effects Year Hansen J (test) 1.1 Hansen J (p-value) 28.57% Kleibergen-Paap (test) 224.0 Kleibergen-Paap (p-value) 0.00% First stage t-Statistics of included instruments Exogenous Trading Volume  15.32 Lagged Trading Volume Deviation from $30 6.77 Liquidityt

Price Impact (2)

Dependent variable: Repurchase Intensity Amihud Spread (3)

(4)

Price Impact (5)

Amihud

 0.0037nnn (  3.66) 0.1180nnn (10.41)  0.0066nnn (  10.36)  0.0034nnn (  8.13)  0.0007nn (  2.38) 0.0009n (1.84) 0.0032nnn (3.51)  0.0026 (  1.12) 0.0039 (0.82)  0.0060nnn (  4.57) 0.0534nnn (2.92) 0.0021nnn (4.92)  0.0004nnn (  2.92)  0.0002 (  0.32)  0.0017nnn (  20.06) 0.0147nnn (6.94) 51,477 Yes Yes Year 0.1 76.13% 149.0 0.00%

 0.0013nnn (  3.89) 0.1220nnn (11.05)  0.0066nnn (  10.74)  0.0040nnn (  7.71)  0.0012nnn (  3.54) 0.0001 (0.23) 0.0035nnn (4.03)  0.0056nn (  2.57) 0.0042 (0.94)  0.0045nnn (  3.09) 0.0422nn (2.31) 0.0021nnn (5.12)  0.0003nn (  2.21)  0.0005 (  0.78)  0.0016nnn (  20.11) 0.0160nnn (7.92) 51,704 Yes Yes Year 0.0 88.89% 362.7 0.00%

 0.0028nnn (  7.29) 0.1264nnn (11.12)  0.0054nnn (  11.13)  0.0040nnn (  10.32)  0.0012nnn (  3.71) 0.0004 (0.97) 0.0035nnn (3.92)  0.0045nn (  2.12) 0.0018 (0.40)  0.0052nnn (  4.68) 0.0422nn (2.38) 0.0021nnn (6.68)  0.0003n (  1.91)  0.0004 (  0.70)  0.0015nnn (  19.24) 0.0146nnn (7.20) 51,652 No Yes Month 0.6 45.24% 368.3 0.00%

 0.0063nnn (  5.77) 0.1217nnn (10.29)  0.0061nnn (  10.55)  0.0046nnn (  9.67)  0.0013nnn (  3.75) 0.0001 (0.16) 0.0029nnn (2.92)  0.0014 (  0.56) 0.0019 (0.36)  0.0040nnn (  2.86) 0.0508nnn (2.74) 0.0029nnn (6.22)  0.0005nnn (  3.40)  0.0002 (  0.23)  0.0015nnn (  17.51) 0.0127nnn (5.48) 51,477 No Yes Month 0.5 47.21% 69.7 0.00%

 0.0011nnn (  7.58) 0.1291nnn (11.49)  0.0055nnn (  11.20)  0.0041nnn (  10.49)  0.0013nnn (  4.09) 0.0004 (1.03) 0.0036nnn (4.15)  0.0047nn (  2.27) 0.0027 (0.62)  0.0051nnn (  4.59) 0.0407nn (2.30) 0.0019nnn (6.51)  0.0003nn (  1.99)  0.0004 (  0.71)  0.0015nnn (  19.49) 0.0155nnn (7.76) 51,704 No Yes Month 0.1 70.37% 436.9 0.00%

 29.60

 21.63

6.52

3.57

 12.74 2.11

 8.40 1.87

 45.5 3.06

(6)

Please cite this article as: Hillert, A., et al., Stock repurchases and liquidity. Journal of Financial Economics (2015), http: //dx.doi.org/10.1016/j.jfineco.2015.08.009i

A. Hillert et al. / Journal of Financial Economics ∎ (∎∎∎∎) ∎∎∎–∎∎∎

15

Fig. 1. Repurchase Intensity and Program Month. The figure plots Repurchase Intensity, calculated as the average number of shares repurchased scaled by the number of shares outstanding, against Program Month, expressed as the number of calendar months since the inception of the program. The broken line shows the actual average. The solid line shows the estimates from the regression in Column 1 of Table 6, assuming that all control variables are at their means. Because we include the natural logarithm of Program Month in this regression, the solid line in the figure is nonlinear.

block trading strategies and which easily extends to purchases.12 The generic setup of these models involves a risk-averse investor who tries to sell a large block in a sequence of smaller transactions. Each transaction has a temporary price impact for liquidity reasons and a permanent price impact for informational reasons. The key trade-off concerns the speed at which investors trade, but models differ on what determines this trade-off. Trading faster creates benefits through reducing the exposure to the uncertain market price of the stock and through improving risk sharing. The costs of trading faster can involve a higher adverse price impact (Almgren and Chriss, 2001) or less efficient exploitation of traders' private information about their endowment (Vayanos, 2001). We can view firms as block traders who wish to buy a large number of shares within a limited time horizon. The testable prediction from this trade-off concerns the patterns of trades. From the models of Almgren and Chriss (2001) and Vayanos (2001), we expect firms to front-load their trades and repurchase larger amounts at the beginning of a repurchase program and smaller amounts later on. In Almgren and Chriss (2001) this conclusion requires risk aversion (see their Fig. 2 and Subsection 2.4). In Vayanos (2001), this conclusion requires parametric restrictions that prevent the firm from optimally manipulating the market, i.e., first repurchasing to drive up the price and later reissuing securities at a higher price. We test both predictions in Table 6, which reports regressions for each liquidity measure as the independent variable. The dependent variable is Repurchase Intensity, and all regressions are estimated using GMM-IV. Columns 1 to 3 use the specification as in Eq. (1) with Exogenous Trading Volume and Deviation from $30 as instruments (see Section 12 Major contributions include Bertsimas and Lo (1998), Almgren and Chriss (2001), Vayanos (2001) and He and Mamaysky (2005). This question has also been the object of an extensive practitioner literature, e.g., Dubil (2002) and Moench (2009).

3). This specification uses year fixed effects instead of month fixed effects and controls for within-year fluctuations in the macroeconomic environment by using the four continuous macroeconomic control variables suggested by Baker and Wurgler (2006). Columns 4 to 6 use Lagged Trading Volume and Deviation from $30 as instruments and control for the macroeconomic environment with month fixed effects. We perform the same specification tests as before. We find that, for all liquidity measures, lower liquidity leads to a lower Repurchase Intensity. Estimates for the coefficient of interest are virtually identical across the two alternative specifications for Spread (compare Column 1 with Column 4) and Amihud (compare Column 3 with Column 6) and somewhat different in magnitude for Price Impact (compare Column 2 with Column 5). In Table A9 of the Internet Appendix we show the within-firm standard deviation for the main independent variables, which is 0.46 for the logarithm of the spread. Increasing the Spread by one within-firm standard deviation leads on average to a reduction in Repurchase Intensity of 0.13% (¼0.46  0.0028; 0.0028 is the coefficient from Table 6). The same calculation gives 0.12% for Amihud and 0.21% for Price Impact. From Table 2, the median of Repurchase Intensity is 0.35%. Increasing liquidity by one within-firm standard deviation leads to a reduction in Repurchase Intensity between one-third and one-half of its median value and is, therefore, economically significant. Thus, repurchase volumes respond strongly to changes in the liquidity of the stock, and we conclude that firms engage in liquidity timing. Next, we test whether firms behave like risk-averse block traders who front-load their trades. Table 6 reports the test results. The coefficient on Program Month is negative and highly significant. Trade size thus is declining during the execution of the program, consistent with the model prediction that firms front-load their trades. In Fig. 1, we plot Repurchase Intensity as a function of Program Month (dashed line) and the relation from Column 1 in Table 6 under the assumption that all control variables are

Please cite this article as: Hillert, A., et al., Stock repurchases and liquidity. Journal of Financial Economics (2015), http: //dx.doi.org/10.1016/j.jfineco.2015.08.009i

A. Hillert et al. / Journal of Financial Economics ∎ (∎∎∎∎) ∎∎∎–∎∎∎

16

Program Size has the predicted positive sign (see Section 3) and is highly significant. Jensen (1986) and Stephens and Weisbach (1998) find that firms tend to repurchase more shares if they have stronger cash flows. We measure operating cash flows as the ratio of EBITDA (earnings before interest, taxes, amortisation, and depreciation) to total assets and corroborate the findings of the previous literature, but results are significant only if Spread and Amihud are used as liquidity measures. Dividends seem to have no impact on repurchases, consistent with the notion that firms view repurchases as complementary to dividend payments instead of as substitutes. Dittmar (2000) shows that firms use repurchases to increase leverage, which is consistent with our finding that firms with a higher leverage conduct fewer repurchases. Value firms with a higher book-to-market ratio consistently conduct more repurchases, probably because they have fewer growth opportunities to reinvest their free cash flow. Options Exercised is the scaled number of shares obtained by employees through stock option exercises and has a positive impact on repurchases, most likely because firms want to hold the number of shares outstanding constant and avoid dilution from option exercises, confirming earlier results (Dittmar, 2000). Babenko (2009) argues that there could be a reverse effect, because repurchases can trigger lower incentive grants if repurchases and grants are substitutes. The coefficients of interest remain unchanged if we remove this control. Hence, this potential endogeneity of Options Exercised is not a concern here. The dummy variable Acquiror indicates acquiror status in a takeover and has a negative, but economically negligible, impact on repurchases. The dummy variable Target indicates target status in a takeover attempt and equals one from the time of the announcement until the completion or cancellation of the takeover. Bagwell

at their means (solid line). The relation is statistically and economically highly significant with t-statistics above 17. Predicted (actual) Repurchase Intensity falls by 45% (60%) from 0.54% (0.54%) in the first program month to 0.29% (0.22%) in the last program month. 5.1.1. Controls The regressions include a range of control variables. We use lagged stock returns for the past three months. Based on the prior literature, we expect a negative sign, because firms tend to repurchase more shares after their stock has declined (Brav, Graham, Harvey, and Michaely, 2005; Stephens and Weisbach, 1998). We find that repurchases respond strongly to past stock returns and all lags have the predicted negative signs. Table 7 The influence of liquidity on repurchases: ordinary least squares (OLS). The table presents OLS regressions of Repurchase Intensity on the liquidity measures as indicated and control variables. We include the same controls as in Table 6. Appendix A.2 provides definitions of the liquidity measures. The repurchase variables and the control variables are defined in Table 1. Standard errors are clustered at the firm level. tStatistics are provided in parentheses. n, nn, and nnn indicate significance at the 10%, 5%, and 1% level, respectively, using t-tests.

Liquidity measure Liquidityt R2 Number of observations Firm fixed effects Time fixed effects

Dependent variable: Repurchase Intensity Spread Price Amihud Impact (1) (2) (3)  0.0015nnn (  13.08) 0.0828 51,683

 0.0004nnn (  5.65) 0.0796 51,508

 0.0009nnn (  15.09) 0.0848 51,735

Yes Month

Yes Month

Yes Month

Table 8 Comovement of repurchases and liquidity. In this table, we form two groups according to whether an observation is above both median Repurchase Intensity and median Spread (High–High) or below both median Repurchase Intensity and median Spread (Low–Low). Differences in means are tested using a standard two-tailed t-test. Differences in medians are tested using the Wilcoxon-Mann-Whitney test (two-tailed). n, nn, and nnn indicate significance at the 10%, 5%, and 1% level, respectively, using t-tests. Panel A: Comparison of means Low–Low

Short Interest Return Abnormal Return Volatility Trading Volume

High–High

N

Mean

N

Mean

Difference

t-Statistic

11,536 11,550 11,200 11,550 11,550

3.90% 1.03%  0.33% 1.79% 3,202

11,547 11,546 9,998 11,550 11,550

4.74%  0.74%  0.47% 2.81% 120

 0.84%nnn 1.76%nnn 0.13%  1.03%nnn 3,081nnn

 13.34 12.39 0.96  48.05 56.95

Median

Difference

ranksum

3.14%  0.37%  0.65% 2.27% 37

nn

Panel B: Comparison of medians Low–Low N Short Interest Return Abnormal Return Volatility Trading Volume

11,536 11,550 11,200 11,550 11,550

High–High Median 2.43% 1.07%  0.31% 1.51% 1,222

N 11,547 11,546 9,998 11,550 11,550

 0.71% 1.44%nnn 0.33%nnn  0.77%nnn 1,185nnn

 2.21 14.66 2.82  53.30 120.55

Please cite this article as: Hillert, A., et al., Stock repurchases and liquidity. Journal of Financial Economics (2015), http: //dx.doi.org/10.1016/j.jfineco.2015.08.009i

A. Hillert et al. / Journal of Financial Economics ∎ (∎∎∎∎) ∎∎∎–∎∎∎

17

Table 9 Firms as providers of liquidity. The table presents generalized method of moments regressions of Repurchase Intensity on either Order Imbalances, Fund Trading, ΔShort Interest, Market Return, or VIX and controls. Instrumented variables are Liquidityt, Order Imbalancest, Fund Tradingt, and ΔShort Interestt. The sample is restricted to the first 12 months of a repurchase program. We include the same control variables as in Table 6 but omit them from the table for readability. Standard errors are clustered at the firm level. t-Statistics are provided in parentheses. n, nn, and nnn indicate significance at the 10%, 5%, and 1% level, respectively. The Hansen J statistic tests for the validity of the overidentifying restrictions. The Kleibergen-Paap test is for underidentification and tests for the full rank of the reduced-form coefficient matrix following Kleibergen and Paap (2006). The table reports the test statistics and the p-values for both tests. The t-tests for the instruments are from the first stage regressions. The test suggested by Stock and Yogo (2005) rejects the hypothesis that the bias exceeds the ordinary least squares bias by more than 5% in all cases.

Liquidity measure

Not included (1)

Liquidityt Order Imbalancest

 0.0475nnn (  6.12)

Spread (2)  0.0031 (  1.46)  0.0800nnn (  3.78)

Fund Tradingt

Not included (3)

Dependent variable: Repurchase Intensity Spread Not Spread included (4) (5) (6)  0.0034nnn (  3.79)

 0.7179nn (  1.98)

 0.0780 (  0.41)

ΔShort Interestt

 0.0025nnn (  2.85)

0.0596nnn (3.65)

47,867 0.940 33.23% 62.978 0.00% No Month Yes

4.81

7.51

(1991) develops a theoretical model to show that repurchases can serve as a takeover defense, and Dittmar (2000) finds supporting evidence for this hypothesis by showing that a positive relation exists between takeover attempts or takeover rumors and share repurchases. We find a negative coefficient on Target, which is never significant. We do not include a control for convertible issues, which seem to be associated with repurchases (see de Jong, Dutordoir, and Verwijmeren, 2011), because only 115 (0.2%) of the repurchase months in our sample are associated with convertible bond issues. 5.1.2. GMM-IV versus OLS To better understand the impact of endogeneity, we compare the GMM results from Table 6 with the same estimations using OLS, which we report in Table 7. Table 7 has the same format and includes the same variables as Table 6, but the control variables are not reported. All OLS regressions show a negative sign on the coefficients of the liquidity measures, but the size of the OLS coefficients is numerically much smaller. The OLS regressions predict that doubling our measures of liquidity would result in reductions of Repurchase Intensity by 0.15% for the Spread, by 0.04% for Price Impact, and by 0.09% for Amihud. Hence, the IV specifications predict

 14.38 6.55

 0.0033nnn (  3.53)

 0.0050nnn (  3.67)

 0.0015nnn (  4.13)

0.0013nnn (7.98) 0.0019nnn (4.41) 51,477 0.183 66.88% 88.685 0.00% Yes Year Yes

0.0009nnn (11.85) 0.0011nnn (4.63) 51,665 0.027 86.97% 311.551 0.00% Yes Year Yes

 9.58 1.92

 19.65 3.37

51,608 0.000 0.000 493.222 0.00% Yes Year Yes

51,592 1.001 31.70% 222.000 0.00% Yes Year Yes

0.0009nnn (11.04) 0.0011nnn (4.31) 51,652 1.672 19.60% 185.136 0.00% Yes Year Yes

27.58

27.76  13.69 6.77

 13.53 6.64

VIXt 47,880 0.000 0.000 22.653 0.00% No Month Yes

Amihud

(7)

Price Impact (8)

(9)

0.0341n (1.85)

Market Returnt

Number of observations 51,639 51,636 Hansen J (test) 0.000 0.796 Hansen J (p-value) 0.000 37.23% Kleibergen-Paap (test) 67.551 17.275 Kleibergen-Paap (p-value) 0.00% 0.02% Macro variables Yes Yes Time fixed effects Year Year Firm fixed effects Yes Yes First stage t-Statistics of included instruments Exogenous Order Imbalances 8.42 8.70 Industry Fund Trading Exogenous Short Interest Exogenous Trading Volume  8.32 Deviation from $30 6.60

Spread

that the impact of a reduction in Repurchase Intensity from a less liquid market is much larger compared with OLS. Again, the OLS bias cannot result form reverse causality, because the impact of repurchases on spreads is also negative, so that reverse causality would reduce the OLS coefficient, rendering the bias negative. We find a positive bias and attribute it to unobserved factors that simultaneously move spreads and repurchases in the same direction. One possible explanation could be that investors sometimes trade on private negative information they have obtained and that firms react to negative information about their price by increasing Repurchase Intensity. While this hypothesis is untestable with our data, because the omitted variables are not related to any observables we have access to, we provide some indirect evidence here. In Table 8, we show univariate statistics on several variables that should be correlated with negative perceptions regarding a particular stock. We compare them across two scenarios. In the Low–Low scenario, Repurchase Intensity and Spread are both below their medians; in the High– High scenario, both variables are above their medians. If spreads and repurchases are both driven by negative perceptions of the firm by some investors, then we expect higher short interest, higher volatility and lower stock

Please cite this article as: Hillert, A., et al., Stock repurchases and liquidity. Journal of Financial Economics (2015), http: //dx.doi.org/10.1016/j.jfineco.2015.08.009i

18

A. Hillert et al. / Journal of Financial Economics ∎ (∎∎∎∎) ∎∎∎–∎∎∎

returns in the High–High scenario compared with the Low–Low scenario. In Table 8, we compare one-month returns, abnormal returns, return volatility, trading volume, and short interest between the High–High and Low– Low scenarios. Univariate comparisons of means and medians show that short interest and volatility are both significantly higher and that stock returns and trading volume are both significantly lower when repurchases and spreads are both above their sample medians, compared with situations in which both are below their sample medians. Results for abnormal returns are significant for medians, but not for means. These results are consistent with the hypothesis that during high repurchase–low liquidity months, some investors take a negative view of the stock, short sell it, and move returns downward, while firms provide some price support by repurchasing the stock. These results are only indicative, and we cannot use any of the contemporaneous variables from Table 8 in our regressions, because they are all simultaneously determined with the repurchase and liquidity measures. 5.2. Liquidity provision In the previous subsection, we ask how firms respond to exogenous variation in the liquidity of their own stock. In this subsection, we go one step further and investigate whether firms also actively attempt to influence the liquidity of their stock. Hong, Wang, and Yu (2008) argue that firms provide liquidity in their own stock and act as “buyers of last resort” in crisis periods. They build on the model of Grossman and Miller (1988), which implies that the market participants with a higher capacity to bear inventory risk should provide liquidity. Unlike traditional market makers, firms do not have to resell repurchased shares and their position in their own stock does not create additional inventory risk for them. In addition, they are less likely to be the uninformed investors whose limit orders are vulnerable to the trades of better informed investors (see Bloomfield, O'Hara, and Saar, 2005). As such, firms are perfectly suited to provide liquidity. Based on this reasoning, we hypothesize that firms buy their own shares when other investors sell them and thereby act as contrarian traders. In this respect, they are similar to hedge funds (Gantchev and Jotikasthira, 2014), who purchase shares in their target firms when institutional investors sell but who have less vulnerable balance sheets. In Table 9, we perform several tests to evaluate this hypothesis. In Columns 1 to 6 of Table 9, we use three different measures of investor selling. In Columns 2, 4, and 6 we use Spread to control for liquidity, and in Columns 1, 3, and 5 we do not control for liquidity. We obtain qualitatively similar results with Price Impact and Amihud as controls for liquidity and report these results in Table A8 in the Internet Appendix. The first tests in Columns 1 and 2 use Order Imbalances as an additional explanatory variable, which is defined as the difference between buys and sells, divided by the combined volume of buys and sells. Our hypothesis is that Order Imbalances carries a negative sign, because a decrease in order imbalances indicates that other investors

sell and market makers are long in the stock of the firm. We instrument for Order Imbalances by using the monthly median order imbalances of all firms in the sample that never repurchase shares during our sample period. According to liquidity models such as Grossman and Miller (1988), Hong, Wang, and Yu (2008) and Nagel (2012), higher order imbalances indicate that the stock price is temporarily depressed and the returns to providing liquidity are higher. Columns 1 and 2 of Table 9 show that the coefficient on Order Imbalances is negative as predicted and statistically highly significant. To see the economic significance, consider a situation in which buy orders and sell orders are initially equal, but then buy orders decrease by one time-series standard deviation of Order Imbalances (15.96%) and sell orders also increase by 15.96%, so that Order Imbalances decreases from zero to  0.1596. (For time series standard deviations see Table A9 in the Internet Appendix.) Then the estimates in Column 1 (Column 2) suggest that Repurchase Intensity increases by 76 (128) basis points, a magnitude larger than its monthly mean of 66 basis points (see Table 2). Columns 3 and 4 of Table 9 build on the same idea but use the trades of mutual funds as an indication of investor buying or selling pressure. Again, we predict a negative sign on the variable Fund Trading, which is positive if mutual funds are net buyers and negative if mutual funds are net sellers. We adapt the approach of Gantchev and Jotikasthira (2014) and instrument Fund Trading with the net trades of all mutual funds in all other firms in the same Fama and French 48 industry in the same month. The coefficients on Fund Trading are negative as predicted. The coefficient in Column 3 is significant and supports the hypothesis that firms tend to provide liquidity by repurchasing more of their shares when other investors sell than when other investors buy their stock. However, once we control for the instrumented Spread in Column 4 this effect vanishes. Mutual fund trading is negatively correlated with Spread, and Spread seems to capture the relevant part of the exogenous variation in mutual fund trading. In Columns 5 and 6 of Table 9, we extend this argument by including ΔShort Interest as a control variable, which is defined as the first difference of shares sold short as of the 15th business day of each month, scaled by shares outstanding. We instrument for ΔShort Interest by the monthly median change in short interest of all firms that never repurchase shares. The theoretical argument is the same as for the other motivations for liquidity provision: Short selling exerts downward pressure on the stock price, and then the buyer-of-last-resort argument implies that firms increase their repurchase intensity when investors short sell their stock. We test for this hypothesis in Columns 5 and 6. We use the first difference of Short Interest on the premise that firms adapt their repurchases to the changes in the short-selling behavior of investors and expect a positive coefficient, i.e., firms should increase their repurchases whenever investors step up short sales of their stock. The results in Table 9 support this prediction. In Column 5, the coefficients on ΔShort Interest is significant at the 1% level. If ΔShort Interest is higher by one within-firm standard deviation, share repurchases are

Please cite this article as: Hillert, A., et al., Stock repurchases and liquidity. Journal of Financial Economics (2015), http: //dx.doi.org/10.1016/j.jfineco.2015.08.009i

A. Hillert et al. / Journal of Financial Economics ∎ (∎∎∎∎) ∎∎∎–∎∎∎ Table 10 Measuring information using filing-day returns. The table presents regressions of the liquidity measures on Repurchase Dummy, cumulative abnormal filing date returns, their interaction, and controls. We include the same controls as in Table 3. Appendix A.2 provides definitions of the liquidity variables. All control variables are defined in Table 1. Filing CAR is the cumulative abnormal return (CAR) of the respective stock from t ¼  1 to t¼ þ 1 relative to the filing day of the 10-Q or 10-K report. The CARs are subsequently matched to the months covered by the report. CARs are computed with the market model using the Center for Research in Security Prices equally weighted index. The estimation window ends 31 days prior to the event day. The estimation length is two hundred days with a minimum of one hundred days being required. Standard errors are clustered at the firm level. t-Statistics are provided in parentheses.n, nn, and nnn indicate significance at the 10%, 5%, and 1% level, respectively, using t-tests. Spread (1)

Price Impact (2)

Amihud (3)

 0.015nnn (  8.29) 0.036nnn (6.11)  0.057nnn (  2.63) 0.798 333,559 Yes Month

 0.012nnn (  4.16) 0.019nnn (2.66)  0.049 (  1.54) 0.623 331,785 Yes Month

 0.026nnn (  6.05) 0.143nnn (8.55)  0.107n (  1.94) 0.738 333,801 Yes Month

Table 11 Measuring information using abnormal returns. The table presents regressions of the liquidity measures on Repurchase Dummy, cumulative abnormal returns (CARs), their interaction, and control variables. We include the same controls as in Table 3. Appendix A.2 provides definitions of the liquidity variables. The repurchase variables and the control variables are defined in Table 1. CARs are computed with the market model using the Center for Research in Security Prices equally weighted index. The estimation window ends six months prior to the event month. The estimation length is 60 months with a minimum of 36 months. Standard errors are clustered at the firm level. t-Statistics are provided in parentheses. n, nn, and nnn indicate significance at the 10%, 5%, and 1% level, respectively, using t-tests.

Repurchase Dummyt CAR (six months)

Repurchase Dummyt Filing CAR Repurchase Dummyt  Filing CAR R2 (adjusted) Number of observations Firm fixed effects Time fixed effects

higher by 9 basis points ( ¼0.0596  1.44%), about one-fifth of median Repurchase Intensity (0.35%). Once we control for Spread, economic significance declines and statistical significance becomes marginal (p-value: 6.5%), which suggests that a significant portion of the impact of ΔShort Interest is absorbed by its impact on liquidity. In Columns 7 to 9 of Table 9, we build on Hong, Wang, and Yu (2008) and ask if firms repurchase more shares in times when the returns to providing liquidity are high. Based on Nagel (2012), we associate returns to liquidity provision with falling market prices and increases in volatility, which he measures with the level of the VIX (Chicago Board Options Exchange Market Volatility Index), an index of implied option volatilities on the S&P 500. We therefore define two dummy variables, Market Return and VIX, to indicate that the value-weighted CRSP market return in that month is below the median of all sample months and the level of the VIX is above its median level. We expect a positive coefficient on both dummy variables if firms repurchase more when the returns to liquidity provision are higher. The table uses three different specifications, which differ only by the particular liquidity measure we use to control for liquidity, but the results are largely unaffected by this control. In all three cases the estimates of both Market Return and VIX are positive and highly significant, with coefficient estimates of around 0.0009 to 0.0013 and 0.0011 to 0.0019, respectively. Hence, firms increase Repurchase Intensity in months when the market return is below its median and the VIX is above its median by 20 to 32 basis points, or by about a quarter to a half of its average value. Overall, we find that firms provide liquidity when other investors sell and in times of crisis, when stock returns are low and volatility is high. This discussion complements the

19

Repurchase Dummyt  CAR (six months) R2 (adjusted) Number of observations Firm fixed effects Time fixed effects

Spread (1)

Price Impact (2)

Amihud (3)

 0.013nnn (  7.37) 0.009nnn (5.69)  0.003 (  0.68) 0.793 291,635 Yes Month

 0.012nnn (  3.96) 0.005nn (2.50)  0.021nnn (  3.19) 0.624 290,029 Yes Month

 0.019nnn (  4.48) 0.047nnn (10.56)  0.014 (  1.29) 0.736 291,815 Yes Month

discussion of liquidity timing in Table 6. It seems that, other things being equal, firms avoid a positive price impact and are careful to reduce transaction costs and not to drive up the price and consume liquidity. However, whenever investors exert downward price pressure by increasing (short) sales, firms become more active providers of liquidity.

6. Liquidity, information and repurchases In this section, we analyze how the information content of share repurchases affects how repurchases influence liquidity. According to the results of Kaniel and Liu (2006) and Bloomfield, O'Hara, and Saar (2005), informed investors demand liquidity whenever the information they trade on is short-lived, but otherwise they supply liquidity. We use two different methods to analyze the relation between information and liquidity. The first is an eventstudy analysis of disclosure-date returns, and the second is an investigation into how liquidity in the repurchase month depends on information as measured by subsequent cumulative abnormal stock returns. In both cases, we rely on time horizons of several weeks or months, which should correspond to the theoretical notion of longlived information. 6.1. Event studies around the filing date The first method we employ to identify informed trading builds on the premise that the information content of repurchases should be revealed in stock price changes at the disclosure date. If the market interprets repurchases as signals about insiders' information, then the disclosure of actual repurchases in 10-Q and 10-K filings should cause positive stock price reactions. We assume that the filing date is also the date around which the information about

Please cite this article as: Hillert, A., et al., Stock repurchases and liquidity. Journal of Financial Economics (2015), http: //dx.doi.org/10.1016/j.jfineco.2015.08.009i

20

A. Hillert et al. / Journal of Financial Economics ∎ (∎∎∎∎) ∎∎∎–∎∎∎

actual repurchases becomes public and perform a standard event study. Cumulative abnormal returns (CARs) around the filing date are calculated from a market model using daily data with an estimation window of two hundred days and a minimum of one hundred days of stock return data. Disclosure day returns are calculated around the filing date in which repurchases are published by the company. We calculate the CAR from one day before to one day after the filing. Filings are quarterly, and filing dates are typically about six weeks after the end of the quarter. For example, a firm could disclose share repurchases executed in January by mid-May, when it files the 10-Q statement for the first quarter. We then associate the CAR for the three days around the filing date in May with the average liquidity of the stock on all trading days in January. The dependent variables are the liquidity measures in the repurchase month, and the independent variables include Repurchase Dummy, Filing CAR, the interaction of Repurchase Dummy with Filing CAR, and the controls from Table 3. Table 10 presents the results. The coefficient of interest is the interaction between Filing CAR and Repurchase Dummy. Under the hypothesis that informed firms consume liquidity to exploit their information, the coefficient on the interaction should be positive. A higher value of Filing CAR shows that repurchases are more informative, and they should reduce liquidity, i.e., increase the spread. Under the alternative hypothesis that informed firms use limit orders to exploit their information, the coefficient on the interaction should be negative. The results in Table 10 lend support to another conclusion: For all liquidity measures, a higher abnormal filing day return is associated with more liquidity in the month of the actual repurchase. This observation is consistent with the notion that firms trade on long-lived information and supply liquidity. However, statistical significance is marginal for Price Impact and Amihud. 6.2. Measuring cumulative abnormal returns after repurchase months Our second approach to identifying informed trading is based on a standard procedure from the insider trading literature and infers the information content of repurchases by looking at abnormal stock returns associated with repurchases after the repurchase month.13 The insider trading literature uses abnormal announcement returns, typically beginning with the disclosure of insider trades. We include the period before the filing, assuming that information can become known to the market through other means, including repurchases themselves. We use the market model with the CRSP equally weighted index and estimate the parameters based on 60 months of monthly data. One difficulty is that we do not know when in the month a repurchase transaction took place. Hence, CARs that include the repurchase month itself could measure price changes before and after the repurchase. We exclude 13 For example, see Lakonishok and Lee (2001) and Fidrmuc, Goergen, and Renneboog (2006). Babenko, Tserlukevich, and Vedrashko (2012) apply a similar method to repurchase announcements.

the repurchase month itself from the calculation of CARs and, therefore, miss the abnormal stock returns between the repurchase transactions and the last day of the repurchase month. Hence, we potentially underestimate the impact of repurchases on CARs. This fact should at most create some noise in the CAR variable and give rise to attenuation bias, i.e., bias the coefficients toward zero. Table 11 presents results for CARs measured over the six months subsequent to the repurchase month. We include CAR, Repurchase Dummy, and the interaction of CAR with Repurchase Dummy. The coefficient on Repurchase Dummy has the same interpretation as before. The coefficient on CAR reveals how market liquidity responds to abnormal stock returns independently of whether these abnormal stock returns are related to repurchases or not. Our coefficient of interest is the one on the interaction term, which should be positive if firms execute repurchases with a higher informational content with liquidity-consuming market orders and negative if firms provide more liquidity if the informational content of repurchases is larger. The control variables are the same as in the previous regressions with the liquidity measures as dependent variables. They are always included but are not displayed in the table. Hence, the effect of repurchases that can be anticipated based on the control variables is removed. We observe that the coefficient on the interaction term Repurchase Dummy × CAR is negative in all specifications, but statistically significant only for Price Impact. Hence, if anything, higher subsequent abnormal returns are associated with larger liquidity improvements during the repurchase month itself. This finding is not in line with the notion that repurchases with larger information content reduce liquidity. The coefficient on CAR is positive and significant at the 5% level or better in all regressions. Informed trading thus is in general detrimental to liquidity. This insight highlights a difference between repurchasing firms and other informed traders: While informed firms do not consume liquidity, other informed traders probably do. 6.3. Conclusion We conclude from the discussion in this section that the information content of repurchases is not associated with a deterioration in liquidity. To the contrary, higher information content seems to be associated with improvements and not with deteriorations in liquidity at the time repurchases were executed. While theoretical models do not specify how the notions of long-lived information and short-lived information should be operationalized, it seems fair to assume that the typical three-month gap between repurchases and the subsequent filings as well as the six-month period over which we calculate CARs count as longer periods. Under this interpretation, the results support the predictions of the models of Harris (1998) and Kaniel and Liu (2006), who see patient informed traders as suppliers of liquidity. The findings in this section are therefore consistent with our conclusions from previous sections and characterize firms as patient traders. Unlike our previous analysis, the analysis in this section does not address concerns about causality. Instead, our

Please cite this article as: Hillert, A., et al., Stock repurchases and liquidity. Journal of Financial Economics (2015), http: //dx.doi.org/10.1016/j.jfineco.2015.08.009i

A. Hillert et al. / Journal of Financial Economics ∎ (∎∎∎∎) ∎∎∎–∎∎∎

research strategy relies on the correlation between spreads and subsequent abnormal returns. The hypothesis that informed trading consumes liquidity, which underlies much of the literature, unequivocally implies that this correlation is positive, i.e., higher returns are related to lower liquidity. We reject this hypothesis by showing that the correlation is in fact negative.

7. Discussion and conclusion We analyze how realized repurchases affect the liquidity of firms' market in their own stock and find that this influence is unequivocally positive. We interpret our findings in terms of the theory of limit order markets. Our overall conclusion is that firms are large, patient traders with a low demand for immediacy. Firms supply liquidity even when they have considerable cash to disburse and even when they find themselves close to the expiration date of their repurchase program. Moreover, repurchases do not reduce liquidity when firms appear to trade on nonpublic information. On average, firms seem to try to reduce transaction costs by buying back shares in times when liquidity is high. Firms sometimes take a more active approach and supply liquidity. They buy when other investors sell or short sell their stock, consistent with the notion that firms act as buyers of last resort in their own stock. They seem to supply liquidity together with price support. Firms also provide liquidity in times of crisis, i.e., when volatility is high and when market returns are low. We use an instrumental variables approach to control for reverse causality and omitted variables and find that different results in the prior literature can be explained by the use of endogenous control variables. Once we remove these, even OLS regressions give qualitatively unequivocal results. The difference between GMM and OLS estimates suggests that the main empirical issue is not reverse causality but the omission of an unobserved factor that drives repurchases and illiquidity measures in the same direction. We cannot identify this unobserved factor but suggest that it is information or sentiment about the stock. Looking at contemporaneous volatility, volume, returns, and short sales is consistent with this hypothesis.

Appendix A A.1. Data collection This section provides more detailed information on how we collected the information on share repurchases and how we constructed the final sample. A.1.1. Collection from filings We collected and edited data on repurchase programs from the Form 10-Q and 10-K filings. The difference between the total number of shares purchased and the number of shares purchased under a program is important. The total number of shares purchased includes, among other things: shares delivered back to the issuer for

21

the payment of taxes resulting from the vesting of restricted stock units, shares delivered back to the issuer by employees and directors for the payment of taxes and the exercise price of stock options, and the repurchase of unvested restricted stock units from employees whose employment terminated before their shares vested. In these cases, the employee—not the company—decides whether the company has to purchase shares. In a repurchase program, the purchase decision is made by the company and the shares are acquired at market prices. However, in transactions with employees, the price can be different from the current stock market price, e.g., if companies use their own fundamental valuation instead of the market price when purchasing shares from their employees.14 Repurchases of unvested restricted stock units from employees whose employment terminated before their shares vested are typically executed at the nominal share value, which is often just one cent. Therefore, repurchases of unvested restricted stock introduce a significant downward bias of the average purchase price.15 In addition to the above-mentioned more common repurchase activities outside of a program there are other, less common transactions, which lead to repurchases outside of active repurchase programs. One example is the repurchase of shares that were issued as acquisition currency when the target is later divested.16 Finally, some data corrections are necessary if companies report transactions under a repurchase program that were repurchased outside the program, e.g., when shareholders held put options against the company.17 In some cases, companies even report buybacks as repurchases under a program even though no program existed at the time.18 While misclassifications from put options, divestitures, and similar transactions are rare, the misclassification of 14 In October 2007 Morgan Stanley (CIK 895421) repurchased shares from employees at an average price of $66.34 while it purchased shares under the repurchase program at an average price of $63.32 (see form 10K filed on January 29, 2008). In October 2008 the difference became even more pronounced, when shares from employees were purchased at $36.13, while shares under the repurchase program were purchased in the open market at $15.09. 15 For example, in April 2006 Sun Microsystems Inc. (CIK 709519) purchased 188,675 shares at an average price of $0.09 although its stock price during that month was around $5. 16 One example is the Interpublic Group of Companies Inc. (CIK 51644), which recorded a repurchase of 15,325 shares in February 2010. The company writes in its 10-K that these shares consist “of our common stock that we received as consideration for the sale of our interest in a company that we previously had acquired (the ’Acquisition Shares’).” 17 In the Form 10-Q filing for the period from April to June 2006 Refac Optical Group (CIK 82788) reports repurchases under a program when shareholders exercised their right to sell their shares to the company at a predetermined price pursuant to a merger agreement. 18 Unit Corp (CIK 798949) records in its 10-Q filing for April to June 2008 all shares as purchased under a program although they were all related to the payment of taxes and to the payment of the exercise price of stock options and Unit Corp did not have any repurchase program at this time. Versata Inc. (CIK 1034397) reports in its 10-Q filing for February to April 2005 purchases from shareholders, who received the right to sell their shares back to the company at a premium in a security class action, as repurchases under the program. At this time Versata Inc. did not have a repurchase program.

Please cite this article as: Hillert, A., et al., Stock repurchases and liquidity. Journal of Financial Economics (2015), http: //dx.doi.org/10.1016/j.jfineco.2015.08.009i

22

A. Hillert et al. / Journal of Financial Economics ∎ (∎∎∎∎) ∎∎∎–∎∎∎

repurchases from employees as repurchases under a program is more common. It is generally not possible to determine the transaction price of the shares purchased under a program when the total number of shares purchased differs from the number of shares purchased under a program. Companies then provide the average purchase price, which corresponds to the total number of shares purchased, and therefore includes purchases outside the program that were conducted at different prices. We correct for these errors by manually checking the footnotes and remarks in the filings and setting the repurchases under a repurchase program to zero whenever such a misclassification took place. Furthermore, we manually adjusted the number of shares and the purchase price for stock splits and stock dividends when necessary. Usually, companies report the repurchase data during the period covered by the filing on a post-stock split basis even if the stock split took place not before the second or the third month of the quarter. For example, if a company repurchases one hundred shares at $10.00 in January and conducts a 2:1 stock split in February, then the company reports this transaction in its filings for the period from January to March as two hundred shares purchased at $5.00. This means that the repurchases in the first and in the second month of a quarter can be reported post-split, although they can take place pre-split. We always adjusted the repurchase data to match the stock market data from CRSP. A.1.2. Details on sample selection From our computerized download, we obtain 96,203 10-Qs and 34,589 10-Ks and extract the repurchase data from these filings. This procedure leaves 376,843 firmmonth observations. Among these are more than 20 thousand firm-months with missing CRSP data if the firms are no longer or not yet listed on Amex, Nasdaq, or NYSE at the time of the repurchase. From the 9,100 repurchase programs, we drop 167 programs with unknown announcement date; 1,587 programs, which were started before 2004; and a further 50, which were announced after 2010. Furthermore, 144 programs are excluded, because they are not executed in the open market (e.g., as private transactions or tender offers). About 3% of the programs in our sample have an unlimited or variable volume. We exclude these because Program Size is one of our instruments and needs to be determined. From CRSP we obtain closing prices, the number of shares outstanding, the number of shares traded, and daily and monthly stock returns. From Compustat we obtain data on total assets, book value of equity, book value of debt, operating income before depreciation, and S&P 500 membership. Data on analyst coverage are from I/B/E/S. All liquidity measures are obtained from TAQ. We eliminate all observations from the final sample for which the variables used in the baseline analysis are not available. These variables are listed in Table 2. A.2. Liquidity measures To calculate Spread and Price Impact, we use the NYSE TAQ database to extract the necessary intraday data. For

each trade, we assign the prevailing bid and ask quotes that are valid at least one second before the trade took place. If there is more than one transaction in a given second, the same bid and ask quotes are matched to all of these transactions. If there is more than one bid and ask quote in a given second, we assume that the last quote is the prevailing quote.19 For the calculation of Spread, we use the NBBO (National Best Bid and Offer) quotes. The NBBO offer size is computed by aggregating all offer sizes at the best bid and best offer ( ¼ask) over all US exchanges (see the Wharton Research Data Services website).20 The final data set contains five items for each transaction. 1. Date and time stamp (up to seconds). 2. Transaction price (Pt). 3. Transaction volume in shares (wt). 4. Prevailing bid quote (Bt). 5. Prevailing ask quote (At). We calculate the quote midpoint price (Qt) as the average of the prevailing bid and ask quotes (Q t = (At + Bt ) /2). We further use the algorithm of Lee and Ready (1991) to classify trades into buys and sells. We define trades with a transaction price above the quote midpoint (Pt > Q t ) as buys and those with a transaction price below the quote midpoint (Pt < Q t ) as sells. If a transaction price is equal to its quote midpoint, we compare the current transaction price with the previous transaction price. If Pt < Pt − 1, we consider a trade to be seller-initiated; if Pt > Pt − 1, we consider it to be buyer-initiated. Should the two prices be equal, we leave the trade unclassified. A.2.1. Spread We calculate the relative spread for each transaction as Spreadt = (At − Bt )/Q t . We further aggregate the relative spreads of all transactions within a day for a particular stock. Spread represents the daily average of all NBBO spreads for a given stock weighted by the time the quote is valid. A.2.2. Price impact We follow the approach of Riordan and Storkenmaier (2011) and define the price impact of each trade after five minutes as Price Impactt = 2 Q t + 5 − Q t /Q t , where Q t + 5 represents the quote midpoint price of the stock after five minutes (three hundred seconds). In the analysis, we use the daily average of this measure for all stocks. A.2.3. Amihud The Amihud illiquidity measure (Amihud, 2002) is calculated by dividing the absolute daily return by trading volume denoted in dollars: 19 Henker and Wang (2006) consider this procedure to be more appropriate compared with the classical Lee and Ready (1991) five-second rule. Bessembinder (2003) tries zero- to thirty-second delays in increments of five seconds and does not find any differences in the results. 20 See http://wrds-web.wharton.upenn.edu/wrds/research/applica tions/microstructure/NBBO%20derivation/.

Please cite this article as: Hillert, A., et al., Stock repurchases and liquidity. Journal of Financial Economics (2015), http: //dx.doi.org/10.1016/j.jfineco.2015.08.009i

A. Hillert et al. / Journal of Financial Economics ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Amihudt =

|rt| Pt⁎ Volt

(3)

where rt represents the daily holding period return, Pt represents the daily closing transaction price, Volt represents the daily transaction volume. Data for these calculations are from CRSP.

References Almgren, R., Chriss, N., 2001. Optimal execution of portfolio transactions. Journal of Risk 3, 5–39. Amihud, Y., 2002. Illiquidity and stock returns: cross section and time series effects. Journal of Financial Markets 5, 31–56. Anand, A., Chakravarty, S., Martell, T., 2005. Empirical evidence on the evolution of liquidity: choice of market versus limit orders by informed and uninformed traders. Journal of Financial Markets 8, 288–308. Aragon, G.O., Strahan, P.E., 2012. Hedge funds as liquidity providers: evidence from the Lehman bankruptcy. Journal of Financial Economics 103, 570–587. Babenko, I., 2009. Share repurchases and pay-performance sensitivity of employee compensation contracts. Journal of Finance 64, 117–150. Babenko, I., Tserlukevich, Y., Vedrashko, A., 2012. The credibility of open market share repurchase signaling. Journal of Financial and Quantitative Analysis 47, 1059–1088. Bagwell, L.S., 1991. Share repurchase and takeover deterrence. RAND Journal of Economics 22, 72–88. Baker, M., Wurgler, J., 2006. Investor sentiment and the cross section of stock returns. Journal of Finance 61, 1645–1680. Banyi, M.L., Dyl, E.A., Kahle, K.M., 2008. Errors in estimating share repurchases. Journal of Corporate Finance 14, 460–474. Barclay, M.J., Smith, C.W., 1988. Corporate payout policy: cash dividends versus open market repurchases. Journal of Financial Economics 22, 61–82. Bargeron, L., Kulchania, M., Thomas, S., 2011. Accelerated share repurchases. Journal of Financial Economics 101, 69–89. Ben-Rephael, A., Oded, J., Wohl, A., 2014. Do firms buy their stock at bargain prices? Evidence from actual stock repurchase disclosures. Review of Finance 18, 1299–1340. Bertsimas, D., Lo, A.W., 1998. Optimal control of execution costs. Journal of Financial Markets 1, 1–50. Bessembinder, H., 2003. Issues in assessing trade execution costs. Journal of Financial Markets 6, 233–257. Bloomfield, R., O'Hara, M., Saar, G., 2005. The “make or take” decision in an electronic market: evidence on the evolution of liquidity. Journal of Financial Economics 75, 165–199. Bonaimé, A.A., 2012. Repurchases, reputation, and returns. Journal of Financial and Quantitative Analysis 47, 469–491. Bonaimé, A.A., Hankins, K.W., Harford, J., 2014. Financial flexibility, risk management, and payout choice. Review of Financial Studies 27, 1074–1101. Brav, A., Graham, J.R., Harvey, C.R., Michaely, R., 2005. Payout policy in the 21st century. Journal of Financial Economics 77, 483–527. Brockman, P., Chung, D.Y., 2001. Managerial timing and corporate liquidity: evidence from actual share repurchases. Journal of Financial Economics 61, 417–448. Brockman, P., Howe, J.S., Mortal, S., 2008. Stock market liquidity and the decision to repurchase. Journal of Corporate Finance 14, 446–459. Busse, J.A., 1999. Volatility timing in mutual funds: evidence from daily returns. Review of Financial Studies 12, 1009–1041. Cao, C., Chen, Y., Liang, B., Lo, A.W., 2013. Can hedge funds time market liquidity? Journal of Financial Economics 109, 493–516. Chakravarty, S., Holden, C.W., 1995. An integrated model of market and limit orders. Journal of Financial Intermediation 4, 213–241. Choi, D., Getmansky, M., Tookes, H., 2009. Convertible bond arbitrage, liquidity externalities, and stock prices. Journal of Financial Economics 91, 227–251. Chung, D.Y., Isakov, D., Perignon, C., 2007. Repurchasing shares on a second trading line. Review of Finance 11, 253–285. Chung, K.H., Elder, J., Kim, J.-C., 2010. Corporate governance and liquidity. Journal of Financial and Quantitative Analysis 45, 265–291. Comerton-Forde, C., Hendershott, T., Jones, C.M., Moulton, P.C., Seasholes, M.S., 2010. Time variation in liquidity: the role of market-maker inventories and revenues. Journal of Finance 65, 295–331.

23

Cook, D.O., Krigman, L., Leach, J., 2004. On the timing and execution of open market repurchases. Review of Financial Studies 17, 463–498. Coval, J., Stafford, E., 2007. Asset fire sales (and purchases) in equity markets. Journal of Financial Economics 86, 479–512. De Cesari, A., Espenlaub, S., Khurshed, A., 2011. Stock repurchases and treasury share sales: Do they stabilize price and enhance liquidity?. Journal of Corporate Finance 17, 1558–1579. de Jong, A., Dutordoir, M., Verwijmeren, P., 2011. Why do convertible issuers simultaneously repurchase stock? An arbitrage-based explanation. Journal of Financial Economics 100, 113–129. Demsetz, H., 1968. The cost of transacting. Quarterly Journal of Economics 82, 33–53. Dittmar, A.K., 2000. Why do firms repurchase stock? Journal of Business 73, 331–355. Dubil, R., 2002. How quickly should you liquidate your vested stock? Financial Services Review 11, 65–84. Fidrmuc, J.P., Goergen, M., Renneboog, L., 2006. Insider trading, news releases, and ownership concentration. Journal of Finance 61, 2931–2973. Foucault, T., Kadan, O., Kandel, E., 2005. Limit order book as a market for liquidity. Review of Financial Studies 18, 1171–1217. Franz, D.R., Rao, R.P., Tripathy, N., 1995. Informed trading risk and bid–ask spread changes around open market stock repurchases in the Nasdaq market. Journal of Financial Research 18, 311–327. Franzoni, F., Plazzi, A., 2013. Do hedge funds provide liquidity? Evidence from their trades. Unpublished working paper. Swiss Finance Institute, Zurich, Switzerland. Gantchev, N., Jotikasthira, C., 2014. Hedge fund activists: do they take cues from institutional exit. Unpublished working paper. University of North Carolina, Chapel Hill, NC. Ginglinger, E., Hamon, J., 2007. Actual share repurchases, timing, and liquidity. Journal of Banking and Finance 31, 915–938. Goettler, R.L., Parlour, C.A., Rajan, U., 2009. Informed traders and limit order markets. Journal of Financial Economics 93, 67–87. Grossman, S.J., Miller, M.H., 1988. Liquidity and market structure. Journal of Finance 43, 617–633. Harris, L., 1998. Optimal dynamic order submission strategies in some stylized trading problems. Financial Markets, Institutions, and Instruments 7, 1–76. Hasbrouck, J., Sofianos, G., 1993. The trades of market makers: an empirical analysis of NYSE specialists. Journal of Finance 48, 1565–1593. He, H., Mamaysky, H., 2005. Dynamic trading policies with price impact. Journal of Economic Dynamics and Control 29, 891–930. Henker, T., Wang, J.-X., 2006. On the importance of timing specifications in market microstructure research. Journal of Financial Markets 9, 162–179. Hoffmann, P., 2014. A dynamic limit order market with fast and slow traders. Journal of Financial Economics 113, 156–169. Hong, H., Wang, J., Yu, J., 2008. Firms as buyers of last resort. Journal of Financial Economics 88, 119–145. Jain, P.K., 2005. Financial market design and the equity premium: electronic versus floor trading. Journal of Finance 60, 2955–2985. Jensen, M.C., 1986. Agency costs of free cash flow, corporate finance, and takeovers. American Economic Review 76, 323–329. Jylhä, P., Rinne, K., Suominen, M., 2014. Do hedge funds supply or demand liquidity? Review of Finance 18, 1259–1298. Kaniel, R., Liu, H., 2006. So what orders do informed traders use? Journal of Business 79, 1867–1913. Kleibergen, F., Paap, R., 2006. Generalized reduced rank tests using the singular value decomposition. Journal of Econometrics 127, 97–126. Lakonishok, J., Lee, I., 2001. Are insider trades informative? Review of Financial Studies 14, 79–111. Lee, C.M.C., Ready, M.J., 1991. Inferring trade direction from intraday data. Journal of Finance 46, 733–746. McNally, W.J., Smith, B.F., 2011. A microstructure analysis of the liquidity impact of open market repurchases. Journal of Financial Research 34, 481–501. Miller, J.M., McConnell, J.J., 1995. Open market share repurchase programs and bid–ask spreads on the NYSE: implications for corporate payout policy. Journal of Financial and Quantitative Analysis 30, 365–382. Moench, B., 2009. Liquidating large security positions strategically: a pragmatic and empirical approach. Financial Markets and Portfolio Management 23, 157–186. Nagel, S., 2012. Evaporating liquidity. Review of Financial Studies 25, 2005–2039.

Please cite this article as: Hillert, A., et al., Stock repurchases and liquidity. Journal of Financial Economics (2015), http: //dx.doi.org/10.1016/j.jfineco.2015.08.009i

24

A. Hillert et al. / Journal of Financial Economics ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Nayar, N., Singh, A.K., Zebedee, A.A., 2008. Share repurchase offers and liquidity: an examination of temporary and permanent effects. Financial Management 37, 251–270. Norli, O., Ostergaard, C., Schindele, I., 2015. Liquidity and shareholder activism. Review of Financial Studies 28, 486–520. Oded, J., 2009. Optimal execution of open market stock repurchase programs. Journal of Financial Markets 12, 832–869. Parlour, C.A., Seppi, D.J., 2008. Limit order markets: a survey. In: Boot, A., Thakor, A. (Eds.), Handbook of Financial Intermediation and Banking, vol. 5. , North-Holland, Amsterdam, pp. 63–95. Rasbrant, J., De Ridder, A., 2013. The liquidity impact of open market share repurchases. Unpublished working paper. Uppsala University, Uppsala, Sweden. Riordan, R., Storkenmaier, A., 2011. Latency, liquidity and price discovery. Journal of Financial Markets 15, 416–437. Stephens, C.P., Weisbach, M.S., 1998. Actual share reacquisitions in open market repurchase programs. Journal of Finance 53, 313–333. Stock, J.H., Yogo, M., 2005. Testing for weak instruments in linear IV regression. In: Andrews, D., Stock, J.H. (Eds.), Identification and

Inference for Econometric Models: Essays in Honor of Thomas Rothenberg, Cambridge University Press, Cambridge, England, pp. 80–108. Stoll, H., 2000. Presidential address: friction. Journal of Finance 55, 1479–1514. Stoll, H.R., 1978. The pricing of security dealer services: an empirical study of Nasdaq stocks. Journal of Finance 33, 1153–1172. Tinic, S.M., 1972. The economics of liquidity services. Quarterly Journal of Economics 86, 79–93. Tinic, S.M., West, R.R., 1974. Marketability of common stocks in Canada and the USA: a comparison of agent- versus dealer-dominated markets. Journal of Finance 29, 729–746. Vayanos, D., 2001. Strategic trading in a dynamic noisy market. Journal of Finance 56, 131–171. Wiggins, J.B., 1994. Open market stock repurchase programs and liquidity. Journal of Financial Research 17, 217–229.

Please cite this article as: Hillert, A., et al., Stock repurchases and liquidity. Journal of Financial Economics (2015), http: //dx.doi.org/10.1016/j.jfineco.2015.08.009i