The Effects of Reduced Analyst Coverage on Firm Equity Raising Behavior

Transcription

1 The Effects of Reduced Analyst Coverage on Firm Equity Raising Behavior Christy Huo University of Melbourne Chander Shekhar* University of Melbourne and Financial Research Network (FIRN) January 2016 *Corresponding author.

2 Abstract We exploit coverage terminations due to analyst closures and mergers as a source of exogenous increase in information asymmetry and its effect on firms equity raising behavior. Our results show that, on average, firms that lose analyst coverage do not alter their overall financing behaviours. After segregating equity raising choices into common stock seasoned equity offerings (SEOs) and private placements, we find that the proceeds raised from SEOs decrease. Furthermore, firms that are financially constrained or experience a loss in retail analyst coverage reduce the level of SEOs, both in terms of frequency and proceeds but no significant changes are found for placements. This supports the argument that analyst disappearances do not affect firms access to institutional investors, but causes retail investors to become less receptive to their seasoned equity offerings. 2

3 1 Introduction It is well established that information asymmetry plays an important role in shaping corporate decisions. Additionally, it impacts the interaction between various stakeholders in the market. The aim of this paper is to examine how financial intermediaries affect the information environment, and the subsequent responses of firms. Notably, this paper studies the causal effect of an exogenous decrease in broker coverage on capital raising behaviours in terms of debt and equity issuances, common stock seasoned equity offerings (SEOs) and private placements. Our results suggest that a variation in information asymmetry affects both the cost of capital of retail investors and the optimal level of SEOs. 1 The exogenous increase in information asymmetry in this natural experiment is induced by broker mergers (Hong and Kacperczyk, 2010) and broker closures (Kelly and Ljungqvist, 2012). The exogenous shock serves to minimise endogeneity biases present in many traditional corporate finance research. This allows us to address the following questions: do firms with greater information asymmetry find it more difficult to access the public market, leading them to issue less SEOs? Or do factors, such as size, simultaneously determine the two factors, so there is no causal link? These questions illustrate the complex interrelations between investors, financial intermediaries and firms, and why simple regressions may give rise to endogeneity 1 We use broker and analyst interchangeably throughout the paper. 3

4 issues. As such, a natural experiment approach is preferred when studying equity raising policies. Our research question is built on the analysis conducted by Derrien and Kecskes (2013). The authors study the effects of broker mergers and closures 2 on four groups of corporate policies: investment, financing, payout and change in cash. We extend their findings on firms financing behaviours, focusing particularly on equity raising behavior. Furthermore, we draw upon conclusions from past literature on equity raising when forming the hypotheses. Many papers consistently find that firms with high levels of information asymmetry tend to choose private placements over SEOs (Chemmanur and Fulghieri, 1999; Wu, 2000; Chen, Dai and Schatzberg, 2010). Based on the assumption that exogenous broker disappearances increase information asymmetry between retail and institutional investors 3, we hypothesise that the level of SEOs decreases after broker disappearances. Next, we hypothesise that the effect is stronger for financially constrained firms, because changes in the external financing market are more likely to affect firms with insufficient internal funds. Lastly, we hypothesise that the effect is stronger if the events are related to retail brokers compared to institutional brokers. The last hypothesis is based on the argument that retail investors cannot access institutional broker reports, thus they are not affected by institutional broker disappearances (Kelly and Ljungqvist, 2012). Institutional investors are also unaffected 2 Collectively broker disappearances or broker events hereafter. 3 See Brennan and Subrahmanyam (1995) and Ellul and Panayides (2009) for evidence that a decrease in analyst coverage increases information symmetry. Kelly and Ljungqvist (2012) also provide empirical support for this finding in their natural experiment. 4

5 as they can reproduce the lost signals. As such, there is no change in information asymmetry associated with institutional broker events. Therefore, we expect insignificant changes for the firms in the institutional broker subsample. Furthermore, the loss of retail broker signals leads to a more significant fall in optimism bias relative to the loss of institutional broker signals (Cowen, Groysberg and Healy, 2006). As such, the firms in the retail broker subsample should experience more significant negative impact on SEOs than the institutional broker subsample. There are 40 broker events in our sample, which are derived from 20 closures and 20 mergers. The broker events comprise 21 retail and 19 institutional brokerage firms. Our sample of firms is restricted to U.S. publicly traded firms that have data in Compustat. This leads to 1,085 exogenous coverage terminations, corresponding to 1,009 unique treatment firm-years or 809 unique treatment firms. Each treatment-year is matched to a control firm of the same year and industry, based on total assets, Q, cash flow and broker coverage. We use a propensity score matching mechanism to construct the control sample. This leads to 1,009 unique control firm-years, which correspond to 818 unique control firms. The unmatched treatment firm-years are not included in the final sample. Next, we use a difference-in-differences (DiD) approach to measure the change in capital raising variables for treatment firms relative to control firms caused by broker disappearances. This allows us to isolate the impact of broker events by removing market-wide trends that affect similar firms simultaneously. We note that some of the results in this analysis are inconsistent with previous research. We find that broker disappearances have no significant impact on debt issuances, equity issuances, total financing or change in cash. Insignificant results hold even when we test subsamples based on market capitalisation and broker coverage. An 5

6 explanation for this is that firm disclosure and broker research are substitutes (Balakrishnan et al., 2014). Consequently, an exogenous decrease in coverage leads to firms increasing their voluntary disclosure. As such, the information environment is largely unaffected, and the broker events have no effect on capital raising choices. When separating equity raisings into common stock SEOs and common stock private placements, the results support the hypotheses. We find that the proceeds from SEOs decrease. Next, we show that both the number of SEOs and the proceeds from SEOs decrease for firms that are financially constrained or experience retail broker disappearances. There is no significant effect on placements, consistent with the assumption that institutional investors can reproduce the lost signals and are therefore unaffected by broker events. As placements do not increase in response to a fall in SEOs, there is no substitution effect. Our results are robust to various specifications and subsamples. This paper contributes to the stream literature on information asymmetry and corporate policies, such as Myers and Majluf (1984), Diamond (1985), Merton (1987), Diamond and Verrecchia (1991), and Easley and O'Hara (2004). Moreover, we illustrate how the effect on equity raising behavior differs conditional upon financial constraint and the type of broker disappearances. The remainder of the paper is organised as follows. Section 2 discusses the relevant literature. Section 3 describes the identification strategy and methodology. Section 4 presents the data. Section 5 reports and explains the main results. Section 6 concludes. 6

7 2 Literature Review Recent corporate finance research has an increasing focus on identifying causal links. By using instrumental variables or natural experiments, researchers are able to identify how changes in certain variables can impact firm value or firm decision. The natural experiment in this paper exploits two exogenous events, which result in a decrease in analyst coverage. The first event is broker mergers, proposed by Hong and Kacperczyk (2010). The authors measure the causal effect of competition in the analyst market on forecast bias. In their sample, the treatment firms are covered by both of the merging brokers before the events. They argue that the subsequent coverage terminations are used to remove overlapping research, and are therefore exogenous to the firms prospects. They also provide evidence that competition in the analyst market can discipline the suppliers of information. Therefore, a fall in competition due to coverage termination leads to an increase in optimism bias. The difference-indifferences estimates from the natural experiment are significantly stronger than estimates from OLS regressions, which face a number of endogeneity problems. The second event is broker closures proposed by Kelly and Ljungqvist (2012). The paper also uses broker mergers as part of the identification strategy 4. The authors argue that the subsequent coverage terminations lead to a loss of public signals. Since institutional investors can reproduce some of the lost signals in-house, whilst retail investors cannot, the information asymmetry between them increases. The authors then show that in the context of asymmetric-information asset pricing models, demand of 4 Kelly and Ljungqvist (2012) redefine the criteria of treatment firms resulting from broker mergers relative to Hong and Kacperczyk (2010). The difference is discussed in section 4. The methodology in this paper follows Kelly and Ljungqvist (2012) more closely. 7

8 retail investors decreases on announcement. This is because after the broker events, retail investors face smaller information set, which increases the conditional payoff variance, whilst expected return remains constant. This induces the risk-averse retail investors to sell, causing prices to fall. At a lower price, institutional investors increase their demand until the equilibrium price is reached. A subsequent paper by Derrien and Kecskes (2013) extend the natural experiment of broker closures and broker mergers to a corporate finance context. They find that investment, financing and cash holdings decrease by 1.9%, 2.0% and 1.1% of total assets respectively. This is because analyst disappearances lead to an increase in information symmetry and cost of capital. Another reason is that treatment firms switch to sources of financing that are less sensitive to information asymmetry, such as reducing equity issuances and increasing use of cash. By dividing the data into subsamples, they also find that these results are stronger for firms that are smaller, have less analyst coverage and are more financially constrained. The aforementioned identification strategy results in a new group of literature that studies the causality between analysts and firm behaviours. Here, we provide a brief summary of some related work. Firstly, Irani and Oesch (2013) find that an exogenous reduction in coverage leads to a deterioration in financial reporting quality. This is consistent with models of managerial misbehaviour and earnings manipulation in the absence of analyst monitoring. Similarly, findings from Chen, Harford and Lin (2014) indicate that exogenous decrease in analyst monitoring results in a higher level of managerial expropriation of outside shareholders. Additionally, He and Tian (2013) identify a negative relation between analyst coverage and innovation, as measured by the number and quality of patents generated by firms. The authors also used an 8

9 instrumental variable approach to confirm the robustness of the results obtained from the natural experiment. Another recent work by Balakrishnan et al. (2014) demonstrates that firms augment their voluntary disclosure when facing an exogenous increase in information asymmetry. This is to compensate the less informed retail investors for the loss of public signals. As a result, information asymmetry and cost of capital decrease, and liquidity and firm value improve. We also study the stream of literature that brings together equity raising and information asymmetry. Firstly, Eckbo and Masulis (1992) argue that information asymmetry can influence managers expectations about shareholders willingness to participate in equity raisings, thereby affecting the actual equity raising choices. Their analysis also indicates that firms with transparent production technology or greater level of mandatory disclosure experience lower information asymmetry. This causes the marginal benefit of quality certification and the resulting probability of employing an underwriter in an equity offering to be lower. Additionally, findings from Chemmanur and Fulghieri (1999) suggest that firms with the highest degree of information asymmetry, similar to that of a private firm, face substantial costs when providing information to investors. The information production costs are found to increase with the number of investors. Therefore, firms with high level of information asymmetry tend to choose private placements over SEOs to reduce information production costs. However, the trade-off is that large investors have more bargaining power against the firms and can demand a lower offer price. This is because institutional investors are less diversified and they contribute the entire portion of the capital of the issuance. In the case of public offerings, equity is priced in a competitive market, so investors have less bargaining power. 9

10 More recently, there are empirical works that support Chemmanur and Fulghieri s (1999) theoretical framework. Firstly, Wu (2004) shows that on average, private placement firms have high information asymmetry. They are characterised by low analyst coverage, low trading volume and large bid-ask spreads. However, the sample set only consists of high-technology post-ipo firms, which pose a potential sample selection bias. The consequential empirical results have limited implication for the wider market. Chen et al. (2010) provide similar insight into firms that employ private investment in public equity (PIPE). PIPEs are associated with shorter trading restriction periods than traditional private placement, but overall they are similar in nature. As such, similar results can be expected when compared to Wu (2004). The authors find that firms with higher information asymmetry and weaker operating performance tend to choose PIPEs, as they cannot efficiently access the SEO market. Consistent with Chen et al. (2010), Floros and Sapp (2012) study the characteristics of firms that issue multiple PIPEs and find that they tend to have high level of information asymmetry. As such, they are unable to access other forms of financing. PIPEs allow them to share private information with a small group of investors, mitigating information asymmetry problems, whilst avoiding sharing proprietary information with the public. The authors also find that information asymmetry remains high over the subsequent PIPEs. Other relevant equity raising papers include Cronqvist and Nilsson (2005), which examine equity issuances made by Swedish public firms. They identify, in the order of increasing information asymmetry, a financing hierarchy: uninsured rights offer, underwritten rights offer, and private placements. Next, Ang and Cheng (2011) demonstrate that firms increase direct communication with the market prior to SEOs, which decreases information asymmetry. The resulting announcement return is less 10

11 negative with an increasing level of communication. Furthermore, Jeppsson (2013) argues that information asymmetry plays a greater role than institutional monitoring in determining the choice between private and public equity offerings in European public biotechnology firms. The author finds that firms tend to issue public equity when information asymmetry is low, as measured by research and development disclosure. More broadly, Gomes and Phillips (2012) find firms with greater asymmetric information prefer to issue securities privately than publicly. The preference is most sensitive for equity raising. There is also evidence that firms switch from public to private market when information asymmetry increases, and vice versa. Lastly, we draw upon literature on analyst bias when examining the difference in the results between firms in the retail and the institutional broker subsamples. The analysis by Cowen et al. (2006) indicates that retail brokers produce more buy recommendation and more positive long-term forecasts than institutional brokers. This is because optimistic retail reports drive trading volume and commissions for retail analysts, as their clients (retail investors) are unable to judge the quality of research and recognise the optimism bias. On the other hand, institutional clients compensate institutional brokers for the quality of their research accordingly, thus there is less incentive to provide optimistic but low-quality research. Additionally, Iskoz (2003) suggests that institutional investors are aware of the potential conflicts of interest in analyst recommendations. Then Malmendier and Shanthikumar (2007) examine how investors address these biases. They find that institutional investors display less optimism in their trades than recommended by their brokers, but small investors make no adjustments and trade according to recommendations. 11

12 3 Identification Strategy This paper employs a natural experiment approach to examine the causal effect of an exogenous variation in information asymmetry on equity raising behavior of firms. This identification strategy is used to overcome certain endogeneity issues that arise in traditional corporate finance research. Namely, the use of proxies for information asymmetry in equity raising analysis 5 may lead to the problem of omitted variable bias. To illustrate this problem, consider regressing the number of private placements on a proxy of information asymmetry. Unobserved and unmeasurable factors such as a change in the company s effort to attract institutional investors can affect both the number of placements and the level of information asymmetry between institutional and retail investors. Without including variables that are correlated with the independent variables, the regression produces biased estimates. Additionally, there is the issue of reversed causality. Companies with higher levels of information asymmetry prefer issuing equity via private placements to SEOs, in order to reduce information production costs or because they cannot access the public market. On the other hand, private placements require disclosure to institutional investors, but not to retail investors, which in turn elevate the level of asymmetric information. Apart from potential biases in the estimates, past literature is also limited to identifying the association between variables, rather than the causal link. This is because panel regressions are performed on corporate policies that are already at equilibrium. On the other hand, the event-based approach of this paper provides a clearer illustration of the cause and consequence. 5 Common proxies include analyst coverage, the proportion of institutional investors and bid-ask spread (see Wu, 2004; Chen et al., 2010). 12

13 Coverage terminations caused by broker closures and broker mergers serve as the exogenous event in our identification strategy. The underlying assumption is that investors have heterogeneous information: retail investors trade on public signals, whereas institutional investors possess additional private signals. Findings in a number of papers suggest that institutional investors are better informed than retail investors (see Szewczyk, Tsetsekos and Varma, 1992; Alangar, Bathala, and Rao 1999; Dennis and Weston, 2001). This is because institutional investors have better access to management, greater incentive to collect information, better processing capabilities and a large pool of analysts (Bushee and Goodman, 2007). The exogenous loss of analyst coverage due to broker disappearances leads to a decrease in public signals. As a result, retail investors have less information. On the other hand, institutional investors ability to access private information is unlikely to be affected by a decrease in external analyst reports. Furthermore, their in-house research departments can reproduce these lost public signals. This increases the degree of asymmetric information between the two groups of investors. Kelly and Ljungqvist (2012) provide empirical evidence to support the contention that information asymmetry increases after an exogenous decrease in broker coverage. This satisfies the relevance condition of natural experiments. For the shock to be suitable for a natural experiment, it must also cause a change in firm behaviours only through a change in information asymmetry. The relevance condition already demonstrates that broker disappearances are correlated with an increase in information asymmetry. However, they do not directly affect firm behaviours, because the resulting coverage terminations are not selective by brokers. As such, the loss in broker coverage is reasonably exogenous to future prospects and characteristics of the firm. Subsequent changes in corporate policies are caused by an 13

14 increase in information asymmetry. This satisfies the exogeneity and only through conditions of natural experiments. To measure the effect of the broker events on capital raisings, we employ difference-in-differences tests. They rely on the assumption of parallel trends, where treatment and control firms would behave similarly in the absence of the shock. For the assumption to hold, the treatment firms need to be similar to the control firms, except the latter is not subject to the shock. Our treatment sample is matched to a control sample based on total assets, Q, cash flow and coverage. In the robustness test section, we also show that this assumption holds by doing a placebo test, where we move the analysis backward by two years. We find no non-parallel trends, thus infer that any significant changes are driven by the shock rather than the systematic differences between the two groups of firms. Subsequently, we perform the difference-in-differences tests. For each dependent variable, the treatment difference and control difference are calculated as the change from pre-treatment window to post-treatment window for treatment firms and control firms, respectively. Then we compute the DiD as the difference between the matched treatment difference and control difference: DDDDDD = yy pppppppp,tt yy pppppp,tt (yy ppoooooo,cc yy pppppp,cc ) (1) where y denotes the dependent variable, pre and post refers to the pre-treatment window and post-treatment window respectively, and treatment firm t is matched to control firm c. We then obtain the mean difference-in-differences by averaging the DiDs across all of the pairs of firms. 14

15 Difference-in-differences tests minimise the time-series effect. The control difference removes market-wide trends that affect similar firms simultaneously and cause firms to experience non-zero changes even without the event. It eliminates the change in treatment group that is not related to the exogenous shock. The net difference between the two groups of firms illustrates the impact of the exogenous shock on treatment firms over and above contemporaneous trends. Throughout the paper, w also obtain difference-in-differences estimation from pooled regressions to demonstrate the statistical robustness of the DiD specification. Firstly, we retain the observations in the pre-treatment and post-treatment periods for each treatment and control firm. Next, we establish the baseline model yy ii = αα + ββ 1 PPPPPPPP ii + ββ 2 TTTTTTTTTTTTTT ii + ββ 3 PPPPPPPP ii TTTTTTTTTTTTTT ii (2) where y i denotes the dependent variable, POST i denotes a dummy variable that is equal to one in the post-treatment period and zero in the pre-treatment period, and TREATED i is a dummy variable that is equal to one if firm i is a treatment firm and zero otherwise. The variable of interest is the interaction term, POST x TREATED, which captures the difference-in-differences effect caused by the shock. When the treatment and control groups are matched using propensity score matching, the systematic differences between the two groups should be eliminated. However, some may remain due to unobserved flaws in the matching procedure. Consequently, the difference-in-differences estimator may capture non-event related variation in the dependent variable caused by these systematic differences. To address this concern and isolate the effect of the shock, the augmented specification adds control variables to the baseline model. The control variables account for variation in the 15

16 dependent variables that is not explained by the event. The augmented specification is as follows yy ii = αα + ββ 1 PPPPPPPP 1 + ββ 2 TTTTTTTTTTTTTT ii + ββ 3 PPPPPPPP ii TTTTTTTTTTTTTT ii + γγ XX ii + εε ii (3) where X i is a vector of control variables. In this paper, we use total assets, Q, cash flow and coverage as control variables. The augmented regression is preferred over the baseline model if they present different results. 4 Data We start by identifying broker disappearance events and the affected firms. Broker disappearance events are obtained by using Factiva for the announcements of brokerage firm closures and SDC Platinum for brokerage firm mergers. We initially identify 44 broker disappearance events, from the first quarter of 2000 to the first quarter of The list of events includes 21 closures and 23 mergers. This includes all of the mergers in Hong and Kacperczyk (2010) for the corresponding sample period and the broker events in Kelly and Ljungqvist (2012). To identify the firms affected by brokerage firm closures and mergers, we obtain the Institutional Brokers Estimate System (I/B/E/S) Stopped Estimate file, which details the announcement date of when brokers discontinue their coverage and the corresponding stocks. The file only provides the estimator codes, which are unique identification numbers for brokers, so we use the I/B/E/S Broker Translation file to determine the broker names from the estimator codes. We then cross-reference the brokers from the Stopped Estimate file with the list of brokers that experience closures or mergers to identify the treatment firms. For closures, we use the stop date in the 16

17 I/B/E/S Stopped Estimate file, which is the last date that a brokerage firm terminates coverage, as the event date. In this sample, we exclude the firms that have their coverage terminated before the event date to avoid endogenous terminations. For mergers, the stop date on I/B/E/S and merger date on SDC do not always correspond, so we use the earlier date as the event date. Firms that are covered by both merging brokerage firms in the 12 months prior to the event date and are still covered by the surviving brokerage firm in the 12 months after the event date are retained. This differs to Hong and Kacperczyk (2010), who do not restrict treatment firms to be covered by the surviving brokerage firm afterwards. As such there are firms that lose coverage from both brokerage firms, which may be due to endogenous reasons. The treatment firms should only experience one broker coverage termination due to overlapping coverage after each merger. This is to ensure that the observations are exogenous coverage terminations. Again, we exclude the firms that have their coverage terminated before the event date. This procedure forms the preliminary set of sample firms. The treatment firms are U.S. operating firms that are publicly traded one year before and after the broker disappearance date. We then exclude firms that are financials and utilities (2-digit SIC codes of 49 and 60-69), and have share codes greater than 11 (REITs, ADRs, unit investment trusts, beneficial interests, closed-end funds, etc.). This leads to a reduction in the number of broker disappearance events to 41. The subsequent set of brokerage firms eliminates Schroders, Cohen Bros. & Co, and Cochran and Caronia Securities, which are listed in Kelly and Ljungqvist (2012). This is because the authors do not exclude financials and utilities. We also classify the brokerage firms into the types of clients they primarily serve, resulting in 21 retail brokerage firms and 20 17

18 institutional brokerage firms. The remaining firms form the initial treatment sample of 2,849 analyst coverage terminations, corresponding to 1,362 unique firms. The treatment firm-years are subsequently matched to control firm-years. Like Derrien and Kecskes (2013), we first construct a set of matching variables: total assets, Tobin s Q, cash flow and coverage. The accounting data is obtained from Compustat and the broker information from I/B/E/S Detail History. Q is defined as the sum of market value of equity and book value of liability, all divided by book value of assets. Cash flow is defined as earnings before extraordinary items plus depreciation and amortization, scaled by total assets. Coverage is calculated by counting the number of brokerage firms that release at least one forecast about the stock during a certain year. Firms with missing data for the matching variables are eliminated. For the accounting and analyst measures, the treatment year is the fiscal year that incorporates the broker disappearance date. Analogously, the pre-treatment window is the period that ends before the event date and the post-treatment window is the period that starts after the event date. The candidates for control firms are required to not have experienced exogenous coverage termination in the one year before the broker event date. We also restrict the control firms to be covered by at least one broker in the one-year pre-treatment window (similar to Irani and Oesch, 2013). We then run a logit regression to estimate the propensity scores. A dummy variable that is equal to one if a particular firm-year is treated and zero otherwise, is regressed on total assets, Q, cash flow and coverage. Finally, we use a one-to-one propensity score matching scheme that matches each treatment firm-year to a control firm with the closest propensity score in the same industry and calendar year. 18

19 Table 1 reports the treatment sample used in this paper with a breakdown of details for the subsamples of broker closures and mergers. The final sample only retains treatment and control firm-years that are matched by industry, total assets, Q, cash flow and broker coverage. Consequently, the final sample contains 1,009 treatment firmyears and the same number of control firm-years. As some firms experience multiple coverage terminations in one year, the sample actually contains 1,085 coverage terminations. This corresponds to 809 unique treatment firms and 818 control firms. There are more control firms than treatment firms, as some firms experience coverage terminations in multiple years and each firm-year can be matched to a different control firm. The final sample comprises 40 broker events, in which there are 21 retail and 19 institutional brokerage firms. The panel of events is broken down into 20 closures and 20 mergers. Table 1 also illustrates that the firms in our sample span across 25 of the 30 Fama-French sectors and do not cluster in industries. Utilities and Financials sectors are excluded by construction. The other industries omitted are Tobacco, Coal and Other. 19

20 Figure 1: Time distribution of broker disappearances and affected firms This figure presents the distribution in calendar time of broker events and treatment firms in Panel A and Panel B respectively. The sample comprises 40 broker disappearances and 1,009 corresponding treatment firm-years over the period between the first quarter of 2000 and the first quarter of The treatment sample is restricted to US operating firms that have been publicly traded for at least one year, are not financials or utilities, and are matched to control firms. 12 Panel A: Number of broker events in a calendar year Panel B: Number of treatment firms in a calendar year Panel A of Figure 1 illustrates the distribution of broker events across calendar years. There is slight clustering in 2000 and 2001, which comprises 48% of broker disappearances in the sample. Furthermore, Panel B shows more clustering in firm level, which occurs in 2000, 2002, 2005 and This is because a small number of 20

21 broker disappearances cause a large number of exogenous coverage terminations. For example in 2002, Robertson Stephens and ABN Amro are responsible for 160 and 133 coverage terminations, or 15% and 12% of the total number of observations in our sample, respectively. About 51% of coverage terminations are caused by the top five broker disappearances. To address the effect of clustering, we rerun the tests for the top 10 brokers separately in the robustness test section. In order to obtain unbiased difference-in-differences estimations, treatment firms and control firms need to be similar in characteristics before the broker disappearances. The propensity score matching scheme is used to ensure that the two groups are similar in terms of total assets, Q, cash flow and coverage. Table 2 illustrates that the matching mechanism performs well as the matching variables are similar between treatment firms and control firms in the fiscal year prior to the event. Additionally, other corporate policy variables are also similar between the two groups. Next we obtain equity raising data from Thomson Financial SDC new issues database for the years 1999 to 2009 (to cover the one-year pre-treatment and posttreatment periods). We retain primary offerings of common stock for the firms in the treatment and control samples. We exclude offerings from SEC Rule 415 shelf registrations and rights issues and classify an offering as a private placement if it is flagged by SDC as issued privately. Analogously, we classify an offering as a SEO if it is issued in the public market. Firm-years without SDC observations are assumed to not have raised any common equity. Common stock SEOs and placements are the main focus of the paper. Table 2 highlights that these variables are also similar between the two groups. 21

22 In summary, there are no systematic differences in corporate policies, except for the change in long term debt, that drive the results in the subsequent sections. The equality of distributions and equality of medians are tested using the Kolmogorov Smirnov test and the sign test respectively. Most of the corresponding p-values are high, suggesting that the two groups are similar. 5 Results We begin by confirming the relevance of the natural experiment by testing the change in broker coverage. Next we examine the change in capital issuance policies due to coverage terminations. This leads to some unexpected results. To identify the drivers of these findings, we break down equity issuance into its components. Notably, we analyse the change in common stock SEOs and placements and conclude with a series of robustness tests to validate the results. 5.1 The effect on broker coverage In this section, we test the validity of the natural experiment given the sample set. Namely, we examine whether or not broker disappearances lead to a fall in broker coverage in treatment firms relative to control firms. Following Irani and Oesch (2013), we test the difference-in-differences of broker coverage. Table 1 shows that for 1,009 treatment-firm years, there are 1,085 coverage terminations, therefore an average treatment firm loses 1.08 brokers per year. Consequently, we hypothesise the mean DiD term to be approximately -1 and statistically significant. We expect a similar result for the coefficient of the interaction term, POST x TREATED, in the pooled regressions, which is an estimator of the mean DiD. 22

23 Table 3 shows the results. The mean DiD term of specification (1) is and statistically significant. Similarly, the baseline model and the augmented model produce significant coefficients of and , respectively. Since all three specifications have significant and negative results, we can conclude that broker coverage decreases post-event. As recent literature identifies a negative relation between exogenous variation broker coverage and information asymmetry 6, we can infer that the treatment sample also experiences an increase in information asymmetry. Therefore, the resulting natural experiment and sample is relevant to the change in information asymmetry. 5.2 The effect on financing Next we examine the impact of a decrease in broker coverage on capital raising policies. There are two interrelated theories relating information asymmetry to financing choices. Firstly, according to the pecking order theory 7, in the presence of information asymmetry, firms prefer to use internal funds first, then issue debt, and then issue equity as a last resort. As a result, an increase in information asymmetry should lead to an even more pronounced order of preference. Namely, we expect a decrease in equity issuance, as firms move away from the source of financing that is the most sensitive to information asymmetry. Secondly, an increase in information asymmetry leads to a higher cost of capital 8, thus external financing becomes more costly. As a result, we hypothesise total financing (sum of short-term debt issuances, long term debt issuances 6 We do not explicitly demonstrate that our treatment sample experiences an increase in information asymmetry as the evidence is presented in other related work. This includes Hong and Kacperczyk (2010), who argue that the forecast error variance among analysts increases after broker mergers. Additionally, Kelly and Ljungqvist (2012) shows that after broker disappearances, there are increases in bid-ask spreads, Amihud illiquidity measure, number of days where no shares are traded, volatility at earnings announcements and earnings surprises. 7 See Myers and Majluf (1984) for detailed theoretical set up and evidence. 8 There is a large body of literature on the relation between information asymmetry and cost of capital, see Diamond and Verrecchia (1991) and Easley and O'Hara (2004). 23

24 and equity issuances, scaled by total assets) to decrease. There should also be a fall in cash holding, as firms use more internal funds to finance their projects to avoid external financing costs. The expected effect on change in short-term and long-term debt is less clear because debt issuance also depends largely on the availability of internal funds and debt capacity. Table 4 details the results on short-term debt financing, long-term debt financing, equity issuance, total financing and change in cash, all scaled by total assets. All of the capital raising variables, except long-term debt financing, have insignificant mean DiD terms in Panel A and insignificant coefficients on the interaction terms, POST x TREATED, in Panel B. Even for changes in long-term debt, the slightly significant coefficient may be driven by the systematic difference between the treatment group and the control group. As Table 2 shows that the medians of change in long-term debt is significantly different at a 10% significance level, caution needs to be exercised when interpreting its difference-in-differences estimators. Table 4 suggests that the firms financing behaviours are not affected by the variation in information asymmetry. This is inconsistent with Derrien and Kecskes (2012), who find that broker disappearances cause a decrease in long-term debt, equity issuance, total financing and cash holdings. This may be due to the fact that the two samples are systematically different. Derrien and Kecskes (2012) use a longer sample period of The firms not included in our sample may be causing the difference in the results. Apart from mechanical reasons, firms liquidity management also provides potential explanations. Previously, we posited that equity issuance and total external financing decrease after broker disappearances due to an increase in information 24

25 asymmetry and cost of capital. This assumes that the information environment around a firm is exogenous to firm behaviours and that managers have no influence over it. However, there is evidence to suggest that firms endogenously affect the level of information asymmetry, and ultimately change the cost of capital. By using a similar natural experiment, Balakrishnan et al. (2014) illustrate that firms choose to increase quarterly EPS guidance, which is a measure of voluntary disclosure. This compensates retail investors for the loss of public signals due to coverage terminations. As a result, liquidity partially recovers a quarter after broker disappearances, suggesting that information asymmetry decreases. The two opposing forces of broker disappearances and firms actively increasing disclosure imply that the net impact on information asymmetry and cost of capital is weakened. As such, it is reasonable to see insignificant change in external financing and cash holding after broker disappearances. Other related work also supports the contention that firms can actively manage their information environment through corporate disclosure. This includes Diamond and Verrecchia (1991), who find that firms with greater voluntary disclosure have lower information asymmetry and higher liquidity. This attracts demand from institutional investors, consequently reducing cost of capital. Graham, Harvey and Rajgopal (2005) also provide survey evidence that CFOs use voluntary disclosure in the form of press releases, investor meetings, monthly newsletters and conference calls to reduce the information risk surrounding the stock. This decreases the information risk premium demanded by investors, reducing the cost of equity. Furthermore, Dass, Nanda and Xiao (2011) analyse the disclosure behaviours of a set of innovative firms, which tend to rely on equity markets for funding. The authors argue that more frequent earnings guidance reduce information asymmetry and increase stock liquidity, subsequently decreasing 25

26 cost of capital. The need to manage liquidity decreases for firms with less reliance on equity markets, which are financially unconstrained firms or firms with access to other sources of financing. More specifically to equity raising, firms are found to increase frequency of disclosure using press releases around a SEO to reduce information asymmetry (Jo and Kim, 2007). This reduces any temporary mispricing of the SEO and reduces post-issue SEO underperformance associated with asymmetric information. Similarly, Ang and Cheng (2011) argue that firms shape the information environment through an endogenous process by choosing the level and channels of communication with the capital market. The authors illustrate that firms that issue SEOs tend to communicate more with external market through company reports and interactions with analysts. Overall, as managers actively increase disclosure to decrease information asymmetry, the net effect of broker disappearances may be insignificant on the information environment of the stock. As a result, there are no significant changes in the capital issuance variables. 5.3 The effect on financing conditional upon size and coverage Another potential explanation to the results above is that large firms or firms covered by more brokers are dominating the effect of the coverage shock. The idea is that larger firms tend to have less information asymmetry, because they disclose more information voluntarily (Diamond and Verrecchia, 1991). Consequently, a decrease in coverage would cause a less substantial increase in information asymmetry for larger firms than smaller firms. Moreover, firms covered by more brokers experience a weaker increase in information asymmetry after broker disappearances, because they still have 26

27 an abundant amount of public signals from other brokers. If our sample is populated with treatment firms that are large and/or have many brokers covering them, then their information environment will not be affected much by coverage terminations. Subsequently, the effect on their behaviours will not be significant when we estimate the difference-in-differences using the whole sample. To test whether broker disappearances affect financing behaviours, at least for small firms or firms covered by fewer brokers, we separate sample firms into market capitalisation quintiles and coverage quintiles. The conditioning variables are applied to the fiscal year that ends prior to the event date, like Derrien and Kecskes (2012). Next we perform the difference-in-differences test as described in equation (1). Table 5 presents the results on the mean DiD. This provides no new insights. Panel A shows the difference-in-differences effect conditional upon market capitalisation. The debt and equity variables are insignificant for all of quintiles. The change in cash is significantly negative for the second quintile, but not for the smallest quintile, thus the result is not meaningful. Panel B shows the difference-in-differences effect conditional upon the number of brokers covering the firms. Again, most of the variables experience insignificant change for all quintiles. The exception is the change in long-term debt, which shows a significant and positive change in the second and the forth quintiles. However, there is no clear trend, so no conclusion can be drawn. On the other hand, the equity issuance variable produces some interesting findings. The mean DiD in the fifth quintile is positive and significant, suggesting that firms with the most broker coverage increase their equity issuances after broker disappearances. This is can be explained by Balakrishnan et al. (2014), who argue that firms covered by many brokers tend to actively increase disclosure in response to a loss of public signals. As they actively 27

28 counteract the effect of coverage terminations, they can actually decrease information asymmetries between retail and institutional investors. Consequently, the order of preference on financing sources becomes less prominent and cost of capital falls. This is consistent with an increase in equity issuances in the largest coverage quintile. We note that in Balakrishnan et al. (2014), increasing EPS guidance only partially counteracts change in information asymmetry caused by broker events. Also, firms that do not provide guidance do not increase voluntary disclosure, so their asymmetric information is elevated permanently. These firms tend to be smaller and covered by less coverage. On average, information asymmetry increases after broker disappearances, but to a lesser extent than expected. The question remains, why are the results showing no impact? We address this question in the next section. 5.4 The effect on the number of SEOs and placements The construction of the equity issuance variable using Compustat data is fundamentally unsuitable for this analysis. Compustat defines equity issuance as funds received from issuance of common and preferred stock. This includes conversion of special stock, preferred stock or debt into common stock; exercise of options and warrants; sale of common or preferred stock; stock issued for acquisitions; and other cash flow related to equity. The equity issuance dataset in Compustat picks up all cash flows related to stock issuances, including those unrelated to raising funds. Therefore, it is not representative of firms equity raising behavior. For the purpose of our analysis, the only relevant component is the sale of common stock. In order to segregate cash flows associated with equity raisings from total equity issuances, we obtain SEO and 28

29 placement data from SDC Platinum. This shows when companies directly access the public or private markets for funds. Table 2 shows that the mean cash flow from Compustat equity issuance for treatment firms and control firms account for 7.96% and 7.78% of total assets, respectively, for the year before the event date. However, SDC data shows that total proceeds from common stock equity raisings (sum of SEOs and placements) for treatment firms and control firms only account for 4.25% and 3.42% of total assets, respectively. About half of the funds received from equity issuance are from activities other than common stock equity raisings, and this component may not change when information asymmetry increases. In other words, the presence of this component in the Compustat equity issuance data dilutes the real change in equity raising. This would explain why the results on equity issuance in sections 5.2 and 5.3 are insignificant. In order to measure the real effect on equity raising choices more accurately, we use SDC equity data in all subsequent tests. Firstly, we test the yearly number of SEOs and placements separately. We estimate the baseline specification and the augmented specification as described in (1) and (2) respectively. The variable of interest is the interaction term, POST x TREATED, which estimates the difference-in-differences effect. Table 6 Panel A presents the results. The interaction terms for both specifications have insignificant coefficients for the number of SEOs and placements. Before claiming that the treatment firms do not change their equity raising behavior, we emphasise that the analysis is only performed on the change in the number of offerings. This is not meaningful on its own, so in section 5.5, we also examine the effect on proceeds from equity raisings to provide a more comprehensive analysis and conclusion. 29