Journal of Economics and Finance

, Volume 36, Issue 3, pp 634–661

Explanation for market response to seasoned equity offerings

Authors

    • School of BusinessWashburn University
  • Sungkyu Kwak
    • School of BusinessWashburn University
  • Rosemary L. Walker
    • School of BusinessWashburn University
Article

DOI: 10.1007/s12197-010-9139-6

Cite this article as:
Hull, R.M., Kwak, S. & Walker, R.L. J Econ Finan (2012) 36: 634. doi:10.1007/s12197-010-9139-6

Abstract

In this paper, we use a multivariate framework to extend the recent univariate seasoned equity offering (SEO) research that investigated the valuation impact of inside ownership. Our multivariate findings re-enforce and add to the univariate findings as we show that the inside ownership level is a consistent factor in accounting for short-run and long-run returns around SEOs, while the decrease in inside ownership has no impact on short-run returns but influences long-run returns in a manner inconsistent with signaling theory. Compared to prior research, our regression tests do a much better job of accounting for returns associated with SEO announcements. For short-run regression tests, the four major factors associated with superior stock returns are: lower underpricing; greater profitability prior to SEO; lower inside ownership level; and, less stock price variability prior to SEO. For long-run regression tests, the four major conditions linked to superior returns are: greater profitability prior to SEO; smaller inside ownership level; relative size of the offering; and, greater decrease in inside ownership level.

Keywords

Inside ownershipSeasoned equity offeringSignaling theory

JEL Classification

D82G14G32

1 Introduction

This paper uses firms undergoing seasoned equity offerings (SEOs) to investigate the influence of both the level of inside ownership and the change in this ownership level within a regression framework. Myers and Majluf (1984) argue that an SEO signals negative news about overvaluation with the strength of the negative signaling increasing as the level or proportion of inside ownership increases. Leland and Pyle (1977) predict that insiders signal negative information to the extent insiders lower their ownership levels by (i) not buying primary SEO shares or (ii) selling shares if secondary shares are involved in the SEO.

SEOs in our data set cover 1999 through 2005. We divide these seven years into two periods: the bubble period covering 1999–2001 and the post-bubble period covering 2002–2005. The bubble period contains both the height of the internet-technology bubble and also the bursting of this bubble. SEOs occurring during this bubble period are characterized by greater price volatility attributable to the greater pre-SEO price run-ups.

From our short-run tests that extend to about a calendar month surrounding SEOs, we find superior (or less inferior) stock returns for these four major conditions: lower underpricing; greater profitability prior to SEO; lower inside ownership level; and, less stock price variability prior to SEO. For our long-run tests, we find superior stock returns for the four years surrounding SEOs for these four major conditions: greater profitability prior to SEO; smaller inside ownership level; relative size of the offering; and, greater decrease in inside ownership level. If we focus on either the pre-SEO or the post-SEO returns, two other variables can be deemed important in explaining both short-run and long-run returns surrounding SEOs. These two variables are pre-SEO price variability and time period of occurrence. Because coefficient signs for these variables are opposite for the pre-SEO and post-SEO tests, these variables do not consistently influence the stock return for the four years around SEOs for all tests. For long-run tests, leverage and Tobin Q variables give opposite pre-SEO and post-SEO effects.

In terms of overall statistical significance for all short-run and long-run tests, the following six variables are the most important: pre-SEO profitability, inside ownership level, pre-SEO price variability, period of occurrence, underpricing, and change in inside ownership. The first two variables are the two most important variables and owe their greater importance to their significant pre-SEO and post-SEO coefficients that have the same predicted coefficient signs for both short-run and long-run returns. The two inside ownership variables (inside ownership level and change in inside ownership level) are included among the six most important variables. However, the significant results for the change in inside ownership are inconsistent with signaling theory based in Leland and Pyle (1977). The inconsistency results because greater insider selling does not signal poorer post-SEO performance. Underpricing is one of the six most important explanatory variables but owes its importance strictly to its impact on short-run price behavior. Two variables (pre-SEO price variability and period of occurrence) owe their importance to their opposite influence on pre-SEO and post-SEO returns.

Our regression tests compare favorably with extant research in terms of accounting for the stock price behavior that accompanies SEOs. For example, our short-run three-day return has an R2 of 0.20. This is about three times greater than prior short-run SEO regression studies. To illustrate, Asquith and Mullins (1986) examine 140 SEOs with primary and secondary components and find an R2 of 0.064; Hull and Moellenberndt (1994) report an R2 of 0.06 for 496 SEOs; Hull and Mazachek (2001) find an R2 value of 0.07 for 455 observations; and, Errunza and Miller (2003) report an R2 of 0.07 for 123 U.S.A. and foreign SEOs. The SEO regression research on long-run price behavior is sparser than the short-run regression research. This makes it difficult to find comparable studies in terms of the number of observations used in tests. Kahle (2000) and Gao and Mahmudi (2008) perform long-run regression tests but analyze much greater sample sizes. They both report an adjusted R2 of 0.02 for a post-SEO three-year return, which is below our R2 of 0.14 for both of our post-SEO two-year tests. While not reported in table format, we find an adjusted R2 of 0.23 when raw returns are used for this test. While our pre-SEO long-run returns yield an R2 of 0.28 for one test, the independent variables included for this test cannot be classified as strictly causal because they occur during or immediately after the time period associated with our pre-SEO dependent return variable. In conclusion, our regression tests do a much better job than prior research when accounting for returns that accompany SEO announcements.

We organize the remainder of our paper as follows. Section 2 gives a literature review. Section 3 presents our sample, methodology and summary statistics. Section 4 offers our general regression model and predictions for independent variables. Section 5 provides empirical results for short-run tests, while Section 6 reports results for long-run tests. Section 7 overviews other tests and future research. Section 8 gives a summary.

2 Literature review

2.1 Inside ownership research

Researchers (Vermaelen 1981; John and Lang 1991; Hirschey and Zaima 1989; Seyhun 1990; Han and Suk 1998; Fried 2005; Firth et al. 2008) collectively show that inside ownership can significantly influence stock price behavior in a variety of ways including dividend initiations, repurchases, sell-offs, take-over bids, stock splits, and share repurchases. In regards to the study of inside ownership as applied to security offerings, there is a variety of research (Gerard and Nanda 1993; Lee 1997; Kahle 2000; Limpaphayom and Ngamwutikul 2004; Ching et al. 2006; Cornett and Travlos 1989; Hull and Mazachek 2001; Lundstrum 2009; Hull et al. forthcoming). This research covers the investigation of (i) insider trading before and after the time of the announcement, (ii) the inside ownership level at the time of the announcement, and (iii) the change in the inside ownership level at the time of a security offering announcement. Prior to Hull et al. (forthcoming), the change in inside ownership level research associated with SEOs was neglected other than attempts by researchers (Cornett and Travlos 1989; Hull and Mazachek 2001) to proxy for insider selling by using the percent change in common stock or the amount of secondary selling.

The findings of insider empirical research, especially as applied to SEOs, are often cited as being consistent with signaling theory originating with Myers and Majluf (1984) and Leland and Pyle (1977). Each of these two theories assumes that insiders have a private information advantage. Concerning this advantage, Myers and Majluf suggest that insiders-managers know more and can take advantage of this knowledge by issuing securities when they are overvalued. Because investors know that a firm does not have to issue equity (such as when it has financial slack), Myers and Majluf argue that issuing equity sends a strong negative signal about overvaluation. It follows that firms with higher levels of inside ownership (who have more to gain from issuing an overvalued security) should signal more negative news when they announce a negative corporate event such as an SEO.

The signaling theory of Leland and Pyle posits that the change in the inside ownership level impacts stock value but this theory makes no real mention of the role of any absolute level or proportion. Leland and Pyle suggest that insiders can signal that a project is good (bad) if they maintain or increase (decrease) their ownership level at the time of an equity offering. Thus, if markets are efficient and insiders have information not known to the market, then how insiders change their ownership levels at the time of an SEO should determine the market response.

2.2 Seasoned equity offering (SEO) research

SEOs provide an ideal experimental setting to examine signaling theories premised on superior insider knowledge since prospectuses can reveal both the level of inside ownership and the change in this inside level caused by the amount of insider selling and buying that will accompany the SEO. The SEO line of research has its origins in a number of short-run announcement period studies beginning with Masulis (1983) and Mikkelson and Partch (1986) who find negative announcement period returns. Regardless of the purpose of the offering, the consensus is that the negative market short-run market response occurs because SEOs signal negative news about stock overvaluation. If this is the case, then greater negative news should be signaled to the extent that insiders have more at stake. Hull and Mazachek (2001), Lundstrum (2009), and Hull et al. (forthcoming) all find that greater inside ownership levels cause greater negative announcement period returns for SEOs and thus greater signaling about overvaluation.

The long-run SEO research, like the short-run SEO research, is plentiful. For example, Spiess and Affleck-Graves (1995) analyze SEOs from 1975 through 1989 and find that their post-offering performance is worse than a sample of matched non-SEO firms that did not issue equity. This poorer performance suggests that managers undertake SEOs to take advantage of overvaluation. Rangan (1998) blames the poorer post-SEO stock price performance on overvaluation that occurs during the price run-ups prior to SEOs. Jegadeesh (2000) uses a number of benchmarks to examine the long-run performance of SEO firms and finds that SEO stock prices significantly underperform these benchmarks for long periods following SEOs. Clarke et al. (2001) examine insider trading surrounding SEOs and discover that the market fails to fully capitalize the negative information in the offering announcement and the insider trading that is taking place around this time. This finding indicates that insiders can use SEOs to change their ownership levels without revealing the extent of their motives. Much of the long-run SEO research (Ritter and Loughran 1995; Kothari and Warner 1997; Lyon et al. 1999; Panagiotis 2009; Lyandres et al. 2008) focuses on problems related to interpretation of this performance given the disagreements over methodologies. Thus, one must be cautious in interpreting long-run results because long-run abnormal return methodologies are viewed with skepticism and sharp disagreements are found (Kothari and Warner 1997; Barber and Lyon 1997; Lyon et al. 1999; Mitchell and Stafford 2000; Li and Zhao 2006). As argued by Hull et al. (forthcoming) univariate comparison tests can minimize methodology problems to the extent methodological errors are canceled out in the comparison process. Likewise, multivariate regression tests can also render correct results regardless of the long-run methodology used. This is because different methodologies should render long-run abnormal returns that are highly correlated since the main component of each methodology involves its cumulative long-run raw return.

2.3 Potential explanatory variables

Inside ownership variables are potential explanatory variables when trying to understand stock price behavior surrounding SEOs. In particular, Myers and Majluf (1984) theorize that the level of inside ownership is an important factor, while Leland and Pyle (1977) predict that it is the change in the inside ownership level that is important. Besides inside ownership effects, there are other effects that should be considered in a regression analysis that seeks to weigh possible factors capable of accounting for the market response to SEOs. For example, Bhushan (1989) advocates a size effect related to greater asymmetric information for smaller firms. Myers and Majluf (1984), Miller and Rock (1985), and Brennan and Kraus (1987) suggest that the purpose of the offering can have an impact. Jensen and Meckling (1976) and Jensen (1986) argue for agency effects related to a firm’s leverage choice. Smith (1977), Hull and Fortin (1993/1994) and Hull and Kerchner (1996) posit an impact from issue expenses.

Grinblatt and Hwang (1989) generalize the Leland and Pyle model by allowing for both the mean and the variance of the project’s cash flow to be unknown. A central finding that emerges from their analysis is that inside ownership is not sufficient to signal the project’s expected value. In their model a second signal, the degree of underpricing, is needed. Thus, when explaining market price behavior at the time of an SEO, it could be useful to include an underpricing variable along the lines suggested by Corwin (2003) who provides an analysis of SEO underpricing. DeAngelo et al. (2010) argue that SEOs are undertaken to resolve a near-term liquidity squeeze. In light of this finding, a regression analysis should examine variables capturing a “liquidity” motive. Finally, researchers (Ritter and Loughran 1997; Harvey et al. 2004) suggest a variety of accounting performance variables that can be used to help augment multivariate tests. Besides those mentioned above (like leverage and liquidity), accounting-based variables that can be used include those capturing valuation effects associated with growth, profitability, and Tobin’s Q. It remains to be seen if any of the above-mentioned variables have merit when used with variables representing (i) the Myers and Majluf (1984) school of thought that speaks to the “absolute” inside ownership level and (ii) the Leland and Pyle (1977) way of thinking that argues for the “change” in the inside ownership level.

3 Sample, methodology, and descriptive statistics

3.1 Sample

The following five screening criteria are used to determine inclusion in our SEO sample. First, a firm must announce its SEO in the Investment Dealers’ Digest (IDD) from January 1999 through December 2005. Second, the firm’s shares must have trading data in the Center for Research in Security Prices (CRSP). Third, we must be able to get a prospectus that is filed with an SEO registration statement. This screen eliminates private placements that are not required to file registration statements.

Fourth, the prospectus must provide information to compute the “change in inside ownership” brought about by an SEO and defined as: (Insider Shares after SEO / Shares Outstanding after SEO) minus (Insider Shares before SEO / Shares Outstanding before SEO) where “insider” includes (i) officers and directors as a group and (ii) any other owner who possesses five percent or more of the company’s shares. For all computations, we find only one increase in the inside ownership level. For this reason, we refer to the “change” in insider level as a “decrease” and thus negative mean and median values occur for the variable representing the insider change. This screen tends to eliminate firms with low insider holdings who would have nothing to report in its prospectus. Fifth, the prospectus must indicate a primary purpose for the offering so that we can classify it as either expansionary or non-expansionary. Examples of a non-expansionary classification include offerings done primarily to reduce debt or to allow current holders to sell shares.

After applying these screens, we are left with a working sample of 706 SEOs. This sample is unique in the sense that it is characterized by firms with higher levels of inside ownership. We examined the 1,599 SEOS that did not satisfy all of our criteria but still had CRSP data. Besides including firms with lower inside ownership levels, this sample appears to be biased by containing SEOs with less stock price run-ups. This indicates that SEOs in our sample of 706 observations have more to gain if the stock price run-up represents greater overvaluation.

3.2 Methodology

We utilize the established abnormal return methodology described previously by researchers (Lyon et al. 1999; Li and Zhao 2006; Viswanathan and Wei 2008; Hull et al. forthcoming) where the holding period’s abnormal return equals that period’s compounded raw stock return minus its compounded expected return. Short-run compounded raw stock returns are figured by compounding daily raw stock returns for a designated holding period, while long-run compounded raw stock returns are calculated by compounding monthly stock returns for a selected holding period. In regards to the specific details, we follow Hull et al. (forthcoming) in computing short-run and long-run abnormal returns. This procedure is described below.

For short-run return calculations, the daily expected return is computed using the OLS procedure described by Brown and Warner (1980) where alphas and betas are calculated using an equal-weighted exchange-based index to represent the market where “exchange-based” indicates the exchange on which the stock is traded: NYSE, AMEX or NASDAQ. We utilize a six-year estimation period (from three years before to three year after the announcement date) to calculate alpha and beta parameters. For firms not traded over the full six years, we use whatever trading data are available on CRSP. For long-run expected returns, we use the equal-weighted exchange-based index’s monthly return to represent the expected return. Once short-run (long-run) expected returns are computed for each day (month), we are then able to calculate the holding period’s compounded expected return. Subtracting the compounded expected return from the compounded raw return gives the compounded abnormal return.

Our general findings are robust to other variations used to get expected returns including the weighting scheme used (equal-weights or value-weights), the index used (exchange-based index or NYSE/AMEX/NASDAQ index), the method used (simple index versus the OLS procedure), or the estimation period used for computing OLS parameters. Due to disagreements over the best method to use especially when computing long-run expected return (Ritter and Loughran 1995; Kothari and Warner 1997; Lyon et al. 1999; Panagiotis 2009; Lyandres et al. 2008), we also tested just raw returns. Whatever methodology is used to get a compounded expected return, this return has to be subtracted from the compounded raw return. Thus, it is the raw return that should drive the regression results regardless of the methodology used to compute the expected return. While using raw returns give similar results compared to abnormal returns, it can be pointed out that the use of raw returns yield F values that tended to be smaller for pre-SEO long-run returns and greater for post-SEO long-run returns. However, all F values are highly significant regardless of using compounded raw returns or compounded abnormal returns.

3.3 Descriptive statistics for key variables

Table 1 provides descriptive statistics in two panels. Panel A in Table 1 gives the number of observations for each year followed by its percentage of the sample’s total. The earlier years of 1999–2001 represent 54% of the sample number while forming just 43% of the sample years. These three years of 1999–2001 are the group of years that best characterize SEOs in our sample that capture the internet-technology bubble period. This period covers the years where prices begin escalating sharply during 1999 before deteriorating during 2001 when the bubble bursts.
Table 1

Descriptive statistics for time periods and key variables

Panel A

Years

N (%)

Years

N (%)

1999

140 (19.8%)

2003

75 (10.6%)

2000

143 (20.3%)

2004

95 (13.5%)

2001

101 (14.3%)

2005

70 (9.9%)

2002

82 (11.6%)

1999–2005

706 (100%)

Panel B

Key variables

Means (Medians)

Total Shares Offered: Primary Shares Offered+Secondary Shares Offered

6.83 M (4.40 M)

Secondary Selling Level: Secondary Shares / Total Shares Offered

0.396 (0.250)

Offer Value: (Offer Price) × (Total Shares Offered)

213 M (111 M)

Common Value: (Offer Price) × (Shares Outstanding before SEO)

2.06B (0.66B)

Issue Cost Level: Issue Costs / Common Value

−0.0084 (−0.0053)

Inside Ownership Level Before: (Insider Shares before SEO) / (Shares Outstanding before SEO)

0.491 (0.467)

Inside Ownership Level After: (Insider Shares after SEO) / (Shares Outstanding after SEO)

0.384 (0.345)

Change in Inside Ownership Level: (Inside Ownership Level After) – (Inside Ownership Level Before)

−0.107 (−0.091)

Relative Size of Offering: Total Shares Offered / Shares Outstanding before SEO

0.195 (0.171)

Underpricing: (Offer Price – Estimated Price) / (Estimated Price)

−0.0371 (−0.0300)

Volatility: Daily standard deviation of a firm’s stock return for the two years before its SEO

0.0458 (0.0421)

Tobin’s Q Ratio: (Common Value + Total Asset – Book Value Equity) /Total Assets (n = 699)

7.114 (2.667)

Profitability Ratio: (Operating Income before Depreciation) / Total Assets (n = 682)

0.026 (0.100)

Tangible Assets Ratio: (Net Plant and Equipment) / Total Assets (n = 682)

0.228 (0.137)

Growth Ratio: Capital Expenditures / Total Assets (n = 676)

0.059 (0.037)

Leverage Ratio: Total Liabilities / (Total Liabilities + Common Value) (n = 699)

0.249 (0.161)

This table provides descriptive statistics. Panel A gives the number of observations for each year followed by its percentage of the sample’s total. This total includes the 706 SEO observations that satisfy the selection criteria described in Section 3.1 Panel B reports means and medians for key variables. Offer Value and Common Value are in U.S. dollars with M and B referring to millions and billions, respectively. Inside ownership consists of (i) officers and directors as a group and (ii) any other owner who possesses five percent or more of the company’s shares. The estimated price used in computing values for the Underpricing variable is given by IDD. Underpricing, like issue costs, is considered an outflow to current owners; thus, negative values are computed so that greater underpricing is expressed with more negative values. Values for the last five Compustat variables are computed using available data from the fiscal year ending closest to the announcement date.

Panel B in Table 1 reports mean and median statistics for key variables with medians reported in parentheses. Since the key variables in Panel B are used in regression tests where outliers can pose problems, we checked all variables for outliers to see if winsorization was desirable. This check revealed a need to winsorize the Tobin’s Q Ratio and the Profitability Ratio. We winsorized the extreme values for these two variables as follows. We divided the sample into 40 percentile groups so that each group contains 2.5% of the sample’s observations. We then set the low extreme values (in the first percentile group) equal to the first value in the next percentile group and high extreme values (in the last percentile group) equal to the last value in the next-to-last percentile group.

The “Total Shares Offered” variable has a mean of 6.83 million shares with one-quarter of these shares being typically offered by current owners as reflected in the median value of 0.250 for the variable “Secondary Selling Level.” The mean “Offer Value” is $213 million and the mean “Common Value” is 2.06 billion. The “Issue Cost Level” variable averages −0.0084 with the negative sign indicating a cost from a current shareholder’s viewpoint. Thus, if shareholders are responsible for issue costs, then these costs cause a fall in value of about 84 cents per $100 dollar of common ownership. The computation of −0.0084 does not include the expenses from the secondary shares sold (as these expenses are assumed to be paid by those who sell the secondary shares). The −0.0084 value can be significantly understated because it does not consider noncash expenses like employees’ time, warrants given to underwriters, and underpricing. Although not given in Table 1, the cost to the firm for every dollar raised is about six cents. The “Inside Ownership Level Before” has a mean of 0.491 indicating that, on average, about half of the total shares outstanding before an SEO are owned by insiders.

The remaining ten variables in Panel B in Table 1 will all be found in at least one regression test reported later. The mean “Inside Ownership Level After” is 0.384 indicating insiders own over 38 out of every one hundred shares outstanding after the SEO is completed. The mean “Change in Inside Ownership Level” is −0.107 with a minus sign reflecting the lowering of the inside ownership proportion caused by the SEO where insiders very rarely buy primary shares while often selling their holdings when secondary shares are involved in the offering. The −0.107 reflects the fall in the mean insider ownership level from 0.491 before the SEOs to 0.384 after the SEOs. For our sample, there is only one observation that had a positive inside ownership change and five that showed no change.

The “Relative Size of Offering” has a mean value of 0.195 indicating that about 20 shares are being offered for every 100 outstanding. We compute underpricing following prior research (Hull and Fortin 1993/1994; Hull and Kerchner 1996) as a negative value to represent a negative impact on shareholder wealth from selling a security below its market value. The mean for “Underpricing” is −0.0371 indicating that the offering price is 3.71% below its estimated price. This estimated price used in the underpricing computation is taken from IDD. The negative sign for underpricing emphasizes the lost value to outstanding shareholders. The lost value would be −0.0371 times the dollar value of the offering with the loss directly attributed to selling shares below the current market value. The “Volatility” variable represents the standard deviation of daily stock returns two years before SEOs and has a mean of 0.0458, which is high compared to most reported norms but is in keeping with our sample and years covered. For the 25 observations with a missing return for at least one day, the volatility is computed using whatever returns are available.

The last five variables in Panel B use accounting data taken from Compustat. Due to incomplete Compustat data, the number of observations used in computing means and medians ranges from 676 to 699. These numbers are less than our working sample of 706 observations that satisfy our screening criteria. Accounting numbers used by Compustat are taken from the annual financial statements occurring closest (yet prior) to the SEO announcements. For Tobin’s Q Ratio, the replacement cost of a firm’s total assets is proxied by the book value for total assets. The median for the Tobin’s Q Ratio is 2.667. Like the Volatility variable, it is high compared to most reported norms but is in keeping with our sample and years covered. The median of 0.100 for the Profitability Ratio indicates that a typical firm issuing an SEO has operating income before depreciation that is 10% of its total assets. In regards to the last three ratios in Panel B, the typical firm (as judged by medians) has close to 14% of its total assets in net plant and equipment, has capital expenditures that are near 4% of its total assets, and has about 16% of its firm value in total liabilities.

3.4 Descriptive statistics for compounded abnormal return variables

Table 2 reports means and medians for short-run and long-run abnormal return variables. These return variables will be used as dependent variables in regression tests. Panel A provides short-run compounded abnormal returns (SRARs), while Panel B reports long-run compounded abnormal returns (LRARs). LRARs are given for both a “full” sample and a “partial” sample. The partial sample deletes an observation if at least one monthly return is missing. Because SRARs and LRARs have outliers that can create errant results, we winsorized the extreme values by following the procedure described previously for TBQ and PFT. While the results reported in this paper are similar whether we winsorized or not, we feel the results are more accurate if winsorization is used.
Table 2

Dependent variables used in regression tests

Panel A Four short-run compounded abnormal return (SRAR) variables

Means (Medians)

Eleven-Day SRAR (day −10 to 0)

−0.0325 (−0.0354)

Three-Day SRAR (days −2, −1, 0)

−0.0260 (−0.0233)

Ten-Day SRAR (days +1 to +10)

0.0233 (0.0150)

Twenty-One-Day SRAR (days −10 to +10)

−0.0096 (−0.0200)

Panel B Three long-run compounded abnormal return (LRAR) variables

Means (Medians)

Pre-SEO LRAR (months −24 to −1)

 Full Sample (n = 706)

1.419 (0.593)

 Partial Sample (n = 506)

1.549 (0.610)

Post-SEO LRAR (months +1 to +24)

 Full Sample (n = 706)

−0.201(−0.379)

 Partial Sample (n = 657)

−0.202 (−0.395)

Four-Year LRAR (months −24 to +24)

 Full Sample (n = 706)

0.804 (−0.019)

 Partial Sample (n = 473)

1.168 (0.188)

This table provides short-run compounded abnormal return (SRAR) and long-run compounded abnormal return (LRAR). Panel A gives mean and median SRARs for four holding periods for the 706 SEO observations that satisfy the selection criteria described in Section 3.1. Panel B reports mean and median LRARs for three holding periods. Because a long-run LRAR loses, on average, about one in five observations for pre-SEO and post-SEO tests, we follow Hull et al. (forthcoming) by reporting LRARs for (i) a “full sample” by using whatever monthly returns are available for each observation for a particular holding period, and (ii) a sample where an observation is deleted if at least one monthly return is missing. The Four-Year LRAR includes 49 months: the 24 months before the SEO, the event month, and the 24 months after the SEO.

Panel A in Table 2 reports that that the mean Three-Day SRAR (days −2, −1, and 0) is −0.0260. This is similar to prior SEO research that typically finds around a −0.03 (or −3%) announcement day response. The mean Twenty-One-Day SRAR (days −10 to +10) of −0.0096 is noticeably less negative than the mean Three-Day SRAR of −0.0260. The more neutral Twenty-One-Day SRAR of −0.0096 reflects the negative mean Eleven-Day SRAR (days −10 to 0) of −0.0325 and the positive mean Ten-Day SRAR (days +1 to +10) of 0.0233. Thus, the ten days after the offering can largely offset whatever negativity is occurring up to the day of the announcement. The Twenty-One-Day SRAR of −0.0096 is similar to the negative impact from issue costs as given in Table 1 by the Issue Cost Level mean of −0.0084. Thus, the price negativity for these 21 days can be primarily explained by issue costs.

Consistent with prior research, Panel B reveals that Pre-SEO LRARs (months −24 to −1) are extremely positive. To illustrate, the Pre-SEO LRAR means are 1.419 for the full sample (n = 706) and 1.549 for the sample without any missing monthly returns (n = 506). These means indicate that the stock price increases about 150% regardless of which sample is used. While the Pre-SEO LRAR medians for each sample are virtually the same (0.593 versus 0.610), they are noticeably smaller than the means indicating that some large positive returns are driving the larger means. The negative Post-SEO LRARs (month +1 to +24) are like prior research. The Post-SEO LRARs means and medians for both the full and partial samples are almost identical (−0.201 versus −0.202 and −0.379 versus −0.395). The likeness in means and medians occur because there are fewer lost observations for Post-SEO LRARs for the partial sample. The Four-Year LRAR (months −24 to +24) for each observation is computed using CRSP return data for the 49 months consisting of (i) the 24 months before the event month, (ii) the event month, and (iii) the 24 months after the event month. From the Four-Year LRAR means of 0.804 and 1.168 for the full and partial samples, respectively, we see that an investor could roughly double their money during this period even after we adjust for expected returns.

4 Regression model and variables

4.1 General regression model

To determine a regression model from independent variables suggested by researchers (and described in Section 2.3), we begin by conducting Pearson and Spearman correlation tests. If there is no significant correlation between a dependent variable with an independent variable, then a meaningful explanatory relation should be ruled out. For briefness’s sake, we do not report details for all of the independent variables tested but only those that pass the following two tests. First, the independent variable had to be significantly correlated with a dependent SRAR or LRAR variable given in Table 2. Second, it had to be significant within a regression framework when entered with other independent variables. Table 3 defines the fifteen independent variables that passed these two tests. Except for the dummy variables in Table 3, the descriptive statistics for the independent variables in Table 3 were given in Panel B of Table 1.
Table 3

Fifteen independent variables used in regression tests

Variable

Definition of variable

ILA

Inside Ownership Level After: (Insider Shares after SEO) / (Shares Outstanding after SEO)

CIL

Change in Inside Ownership Level: (Inside Ownership Level After) – (Inside Ownership Level Before)

TIM

TIM = 1 if SEO announced after 12/31/01; else TIM = 0

RSD

Volatility: Daily standard deviation of a firm’s stock return for the two years before its SEO

UND

Underpricing: (Offer Price – Estimated Price) / (Estimated Price)

EXC

EXC = 1 if NYSE/AMEX; else EXC = 0

SIZ

Logarithm of Common Value (described in Table 1 and expressed in millions before the log is taken)

RSZ

Relative Size of the Offering: Total Shares Offered / Shares Outstanding

SEC

SEC = 1 if secondary selling greater than one-third of total offering; else SEC = 0

EXP

EXP = 1 if major purpose is expansion-related; else EXP = 0

TBQ

Tobin’s Q Ratio: (Common Value+Total Asset – Book Value Equity) /Total Assets

PFT

Profitability Ratio: (Operating Income before Depreciation) / Total Assets

PTA

Tangible Assets Ratio: (Net Plant and Equipment) / Total Assets

GRO

Growth Ratio: Capital Expenditures / Total Assets

LEV

Leverage Ratio: Total Liabilities / (Total Liabilities + Common Value)

This table reports definitions for the fifteen independent variables used in regression tests. Each independent variable shares two common characteristics. First, the variable had to be significantly correlated with a dependent return variable. Second, it had to be significant when regressed against this same dependent variable when other independent variables were included in the test. UND values are negative so that less underpricing represents a less negative value indicating a smaller cash outflow for current equity owners.

In determining our independent variables, we occasionally found that an independent variable could be highly significant in correlation analysis with a dependent variable but not significant in regression tests with that same dependent variable. To illustrate, consider a liquidity variable (LIQ) such as computed by adding cash plus short-term investments and then dividing by the market value of its common stock. Bates et al. (2009) suggest that LIQ should be important given the gradual increase in liquidity since 1980 that enables firms to be viewed as less risky. However, LIQ never entered one of our regression tests as significant because it is highly correlated with our profitability variable, PFT, and our risk variable, RSD. These two variables (PFT and RSD) apparently better capture a valuation effect that might otherwise be associated with LIQ.

Our general regression model includes four of the fifteen independent variables in all regression tests. These four variables are included based on following two considerations. First, because this study aims at extending the univariate insider research, we include two insider variables in all tests. These variables are ILA (which captures the inside ownership level after the SEO) and CIL (which represents the change in the inside ownership level caused by the SEO). Second, we include two additional variables because they are deemed important for our study. The first of these two variables is the time period dummy variable (TIM) that is judged important because our study covers the bubble period and the post-bubble period. The second variable (RSD) is considered important because SEOs are characterized by great stock price volatility. For example, our mean standard deviation of daily returns for the two years prior to SEOs (as given in Table 1 as 0.0458) is over 70% greater than that reported by Brown and Warner (1984). In addition, the average daily stock price volatility of 0.0530 during the bubble period is much greater than the post-bubble period average of 0.0371.

With the above in mind, our general regression model is:
$$ Return = {\beta_0} + {\beta_1}ILA + {\beta_2}CIL + {\beta_3}TIM + {\beta_4}RSD + {\beta_5}V5 + { }{\beta_6}V6 + {\beta_7}V7 + {\beta_8}V8 + \xi . $$
  • Return is one of the SRAR and LRAR dependent variables described in Table 2.

  • ILA is the “Inside Ownership Level After”: (Insider Shares after SEO) / (Shares Outstanding after SEO).

  • CIL is the “Change in Inside Ownership Level”: (Inside Ownership Level After) – (Inside Ownership Level Before).

  • TIM = 1 if SEO announced after 12/31/01; else TIM = 0.

  • RSD is the daily standard deviation of a firm’s stock return for the two years before its SEO.

  • V5 through V8 denotes that no more than four other independent variables enter any one of our regression tests.

  • ξ is the error term.

Each regression test is allowed to add independent variables (beyond ILA, CIL, TIM, and RSD) to produce a set of independent variables that best explain the dependent variable being used. As will be seen later when presenting regression results, no more than four other independent variables besides ILA, CIL, TIM and RSD will enter a regression test. This is because no more than four additional variables are significant for any one test.

When conducting our regression tests, we attempt to adjust for collinearity if two independent variables are highly correlated and the regression results deviate from what is indicated from correlation analysis. For example, consider the leverage variable (LEV) which has Pearson and Spearman correlation coefficients of −0.48 and −0.59 with RSD. Suppose LEV and RSD both enter a regression test and LEV’s regression coefficient is inconsistent with its correlation coefficient with the dependent variable. In order to determine if this inconsistency is caused by collinearity, we get the residuals formed from regressing LEV against RSD. These residuals are nearly perfectly correlated with RSD but not correlated with LEV. Using this residual for RSD may weaken the significant effect of RSD but this can be more than offset if LEV is enabled to better represent its true effect.

4.2 Predictions for independent variables when regressed against dependent variables

Each independent variable and its predicted coefficient sign are given in Table 4. Except for the constant term, one-tailed t-tests are used since each variable has a definite prediction as to its coefficient sign. Below is a description of each independent variable and an explanation for its expected coefficient sign.
Table 4

Predicted coefficient signs of independent variables for the seven dependent variables

Variables

Eleven-Day SRAR

Three-Day SRAR

Ten-Day SRAR

Twenty-One-Day SRAR

Pre-SEO LRAR

Post-SEO LRAR

Four-Year LRAR

ILA

+

CIL

+

+

+

+

+

+

TIM

+

+

+

+

+

+

RSD

+

UND

+

+

+

+

+

+

EXC

+

+

+

+

+

+

SIZ

+

+

+

+

+

+

RSZ

+

SEC

+

EXP

+

TBQ

+

PFT

+

+

+

+

+

+

+

PTA

+

+

+

+

+

+

+

GRO

+

+

+

+

+

+

+

LEV

+

+

+

+

+

+

This table gives the predicted coefficient signs for the fifteen independent variables (described in Table 3) when regressed against each of the seven dependent return variables (described in Table 2). Even though all independent variables do not enter all regression tests, we still give all predicted signs because generally speaking the same predicted sign for each independent variable occurs for most (if not all) tests.

Myers and Majluf (1984) predict that Inside Ownership Level After (ILA) will have a negative coefficient since a greater ownership level will signal more negative news at the time of the announcement (as insiders have more to gain by issuing overvalued equity when their ownership levels are greater). The greater negative news prediction should carry over in the post-SEO LRAR. The negative signaling about overvaluation suggests a positive coefficient for the Pre-SEO LRAR if greater pre-SEO price run-ups signal more overvaluation. We predict a negative coefficient for the Four-Year LRAR as firms with more inside ownership that undergo SEOs would likely be signaling greater overvaluation and thus underperforming for a long-run period that includes post-SEO returns. Although they do not test long-run returns, Hull and Mazachek (2001) find that a variable resembling ILA is the key variable in their short-run regression tests and is significantly negative when regressed with three other independent variables that are included in our tests: the relative size of the offering (RSZ), firm size (SIZ), and secondary selling (SEC).

Leland and Pyle (1977) predict a positive coefficient for the Change in Inside Ownership Level (CIL) when used with the four SRAR variables and a post-SEO LRAR variable. This is because a sample of SEOs that almost exclusively reveal net decreases in inside ownership should have more negative SRARs when the decreases in inside ownership is greater. The greater negative signaling for greater negative values for CIL should be fully realized through the Post-SEO LRAR value, thus suggesting a positive coefficient. The notion of insiders unloading greater shares indicates greater overvaluation from greater price run-ups. If so, this indicates a negative coefficient for the Pre-SEO LRAR. Relatedly, Fahlenbrach and Stulz (2009) suggest that greater decreases in inside ownership occur when firms have been performing well. We predict a positive coefficient for the Four-Year LRAR as firms with more inside ownership decreases would likely be underperforming for long-run periods surrounding SEOs.

For TIM, we predict positive coefficients for SRAR variables. This is because SEOs that occur after the end of the year 2001 (when TIM = 1) take place after the internet-technology bubble period has ended and thus overvaluation is less of a concern due to relatively smaller stock price run-ups. This translates into a less negative market response immediately surrounding an SEO announcement. For Pre-SEO LRARs, we expect a negative coefficient because SEOs that occur in later years (when TIM = 1) will have less positive price performance compared to SEOs that occur during a bubble period where there is much greater market-adjusted price run-ups for firms having SEOs. Because SEOs after the bubble period avoid the dramatic drop-offs, we predict positive coefficients for Post-SEO LRARs and Four-Year LRARs with the positive coefficient for the latter partially resulting from the fact some post-bubble SEOs will have also shared in part of the bubble period’s price run-ups.

Greater values for the daily standard deviation of a firm’s stock return for the two years before its SEO (RSD) indicate more uncertainty and greater negative asymmetric information for a negative event like an SEO. Thus, we predict negative coefficients for RSD when used with SRAR variables and the Post-SEO LRAR. We predict a positive coefficient when used with Pre-SEO LRAR as greater price volatility before SEOs should be associated with greater positive prices given the fact that SEOs follow on the heels of price run-ups. Greater uncertainty should imply inferior returns for longer time periods surrounding SEOs, so we predict a negative coefficient for the Four-Year LRAR.

UND refers to underpricing. Greater underpricing (which is represented by greater negative numbers because it creates greater costs to current shareholders) should cause more negative SRAR values. Thus, we predict positive coefficients for SRAR tests. Greater underpricing (greater negative values for UND) can reflect greater positive stock price run-ups so we predict a negative coefficient for Pre-SEO LRAR tests. Because greater underpricing can indicate negative news about overvaluation, we hypothesize a positive coefficient for Post-SEO LRARs and longer periods as captured by the Four-Year LRARs.

The exchange listing and size of the firm variables (EXC and SIZ) should behave in a similar fashion because EXC = 0 represents NASDAQ firms with smaller values for SIZ. As suggested by Bhushan (1989), we predict positive coefficients for SRARs due to greater negative news associated with more differential information for smaller firms that tend to be traded on NASDAQ. Hull and Mazachek (2001) find a positive coefficient for SIZ in their announcement period tests. The negative announcement period news should be realized over time causing a positive coefficient for the Post-SEO LRAR and the Four-Year LRAR tests. We predict a negative coefficient for the Pre-SEO LRAR test as smaller firms should do better during periods of stock price run-ups where uncertainty and risk tend to be underestimated due to the well-documented euphoria and irrationality associated with a bubble period.

Greater values for the relative size of the offering (RSZ) should signal greater negative news due to fears that owners (including its selling shareholders) are taking greater advantage of an overvalued stock price situation. The rationale for the coefficient predictions for RSZ are like those stated previously for CIL. However, the predicted signs of the coefficients are opposite because values for RSZ are positive and values for CIL are negative. The predictions for our secondary selling variable (SEC = 1 if secondary selling greater than one-third of total offering) are (like RSZ) opposite of CIL since SEC = 1 signifies greater selling by current owners and we estimate that half of these current owners can be insiders. Hull and Mazachek (2001) test a variable similar to SEC and report short-run results consistent with our predictions.

According to Myers and Majluf (1984), SEOs undertaken for expansion-related purposes should signal negative news because good projects should typically be financed with debt. Thus, we predict a negative coefficient for EXP (EXP = 1) for SRAR tests. Miller and Rock (1985) and Brennan and Kraus (1987) also suggest a negative coefficient as they argue that the use of proceeds for debt reduction mitigates any negative market response. Thus, the lack of a debt purpose indicates greater negative news. This negative response should carry over for the Post-SEO LRAR and the Four-Year LRAR tests. A positive coefficient is predicted for Pre-SEO LRAR tests as the greater negative news about overvaluation is likely associated with greater pre-SEO price run-ups.

Greater values for Tobin’s Q (TBQ) indicate the possibility of overvaluation, thus exacerbating risk and causing negative coefficient for all tests except Pre-SEO LRAR tests where increases in stock prices translate into rising TBQ values. Greater profitability should mitigate any negative signaling and also magnify any pre-SEO run-ups, thus we predict positive coefficient for all PFT tests. Likewise, greater levels of tangible assets should also mitigate risk in its own fashion and thus cause all coefficients for PTA to be positive. Growth (as captured by relatively greater capital expenditures for the past year) indicates a greater likelihood that firms have good projects and so an SEO can be greeted with less negativity. Thus, like PFT and PTA, positive coefficients for GRO are expected for SRAR and LRAR tests. Agency theories (Jensen and Meckling 1976; Jensen 1986) predict that greater amounts of debt prevent wasteful spending by managers. If so, returns surrounding SEOs should be less negative. Thus, we predict positive coefficients for LEV for SRAR and LRAR tests. An exception is Pre-SEO LRAR tests where more positive price run-ups should lead to lower leverage ratios. Thus, we predict a negative coefficient for LEV for this test.

As capsulized in Table 4, we always predict that the Ten-Day SRAR will render the same predicted sign as for the three other SRAR tests. However, this may not be the case if there is overreaction on Day 0 such that the Ten-Day SRAR will have an opposite market response to what occurs immediately prior to an SEO. Similarly, buying by underwriters to smooth the market or trading strategies by large institutional investors (like hedge funds) may cause prediction problems. Table 4 reveals that the predicted sign for Post-SEO LRAR tests resemble the Four-Year LRAR tests. This prediction may not always hold if an independent variable is associated with an extremely large positive Pre-SEO LRAR that can dominate the return for the longer four year horizon. While only 25% of the signs will later be shown to disagree with their predicted signs, those that do disagree tend to be associated with the Ten-Day SRAR and Four-Year LRAR tests.

5 Regression results for SRAR variables

Table 5 reports short-run regression results for the four SRAR dependent variables. A SRAR test can lose anywhere from 26 to 31 observations due to missing Compustat data. Variance inflation factors (VIFs) are not reported in Table 5 because they are well below any generally accepted cutoff (ranging from 4.0 to 10.0) for indicating multicollinearity concerns. For example, the maximum VIF is only 1.38. Regardless, as described toward the end of Section 4.1, collinearity tests were conducted when warranted. The more noteworthy of these tests are incorporated into the results in Table 5. Because R2 values are similar to adjusted R2 values, we only report adjusted R2 values. The greatest adjusted R2 value of 0.20 and F value of 28.9 occur for the Three-Day SRAR (days −2, −1, and 0) test indicating this short-run test does the best job of accounting for the market response. The test that does the poorest job is the Ten-Day SRAR test (+1 to +10), which has adjusted R2 and F values of 0.04 and 4.36, respectively.
Table 5

Short-run regression results for four dependent variables

Dependent Variable: Eleven-Day SRAR for days −10 to 0 (n = 682)

Const

ILA

CIL

TIM(R)

RSD(R)

UND(R)

PFT(R)

EXP

SIZ

AdjR2/F

−0.059

−0.079

−0.050

0.019

−0.552

0.610

0.056

0.026

0.007

0.08

−1.92*

−3.16***

−0.68

1.68**

−2.15**

5.78***

2.32***

2.15**

1.64**

8.36***

Dependent Variable: Three-Day SRAR for days −2, −1, and 0 (n = 682)

Const

ILA

CIL

TIM(R)

RSD(R)

UND(R)

PFT(R)

  

AdjR2/F

−0.010

−0.035

0.027

0.020

−0.609

0.617

0.036

  

0.20

−1.49

−2.66***

0.72

3.25***

−4.52***

11.1***

2.83***

  

28.9***

Dependent Variable: Ten-Day SRAR for days +1 to +10 (n = 675)

Const

ILA

CIL

TIM(R)

RSD(R)

UND(R)

RSZ(R)

GRO

 

AdjR2/F

0.036

−0.022

0.046

−0.026

0.262

0.282

0.086

0.138

 

0.04

3.24***

−1.02

0.66

−2.56***

1.15

3.04***

2.09**

2.12**

 

4.36***

Dependent Variable: Twenty-One-Day SRAR for days −10 to +10 (n = 682)

Const

ILA

CIL

TIM(R)

RSD(R)

UND(R)

PFT(R)

EXP

 

AdjR2/F

0.013

−0.092

−0.041

−0.016

−0.526

0.880

0.088

0.034

 

0.08

0.78

−2.76***

−0.43

−1.06

−1.51*

6.17***

2.69***

2.06**

 

9.38***

TIM(R): A residual nearly perfectly correlated with TIM, but not correlated with RSD, was used in the test.

Pearson and Spearman rhos between TIM and RSD are −0.38 and −0.40.

RSD(R): A residual nearly perfectly correlated with RSD, but not correlated with ILA, was used in the test.

Pearson and Spearman rhos between RSD and ILA are 0.21 and 0.16.

UND(R): A residual nearly perfectly correlated with UND, but not correlated with RSD, was used in the test.

Pearson and Spearman rhos between ILA and RSD are −0.22 and −0.22.

PFT(R): A residual nearly perfectly correlated with PFT, but not correlated with RSD, was used in the test.

Pearson and Spearman rhos between PFT and RSD are −0.42 and −0.41.

RSZ(R): A residual nearly perfectly correlated with RSZ, but not correlated with ILA, was used in the test.

Pearson and Spearman rhos between RSZ and ILA are −0.13 and −0.11.

This table gives statistical results for short-run regression tests. Our general regression model for these tests is described in Section 4.1 and allows each dependent SRAR variable to be used with four independent variables of concern (ILA, CIL, TIM, and RSD) and any other independent variable described in Table 3 that is significant for the SRAR being tested. The first column gives the constant’s coefficient with its t statistic below. Subsequent columns for independent variables provide coefficients with the corresponding t statistics below. Except for the constant variable, the t-test is one-tailed because each independent variable has a definite predicted sign for its coefficient for each test (see Table 4 for each predicted sign). The last column gives the adjusted R2 value with the F value below. The symbols *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively.

5.1 Results for Eleven-Day SRARs and Three-Day SRARs

Table 5 reveals that the Eleven-Day SRAR (days −10 to 0) and Three-Day SRAR tests give similar results except EXP and SIZ only enter the Eleven-Day SRAR test as significant. Three independent variables stand out for both tests: ILA, UND, and PFT. The negative coefficient for ILA and the positive coefficients for UND and PFT are significant at the 1% level and have their predicted signs as given in Table 4 with the results for ILA consistent with Myers and Majluf (1984) negative signaling about overvaluation. While less significant, Table 5 discloses that TIM and RSD are two other factors of importance. TIM has its predicted positive sign and is significant at the 1% level for the Three-Day SRAR test and at the 5% level for the Eleven-Day SRAR test. RSD has its predicted negative sign and shares in the same significant levels as TIM. We conclude that the market will react in a less negative fashion (immediately before and at the time of the SEO announcement) if the firm has lower inside ownership, less underpricing, more profitability for the prior year, SEO occurring after a bubble period, and less pre-SEO price volatility.

There are two other independent variables in Table 5 for Eleven-Day SRAR tests that are significant: EXP and SIZ. First, EXP has a positive coefficient that is significant at the 5% level. However, its positive sign is opposite of that predicted by signaling theory indicating that an expansion-related purpose (EXP = 1) is not perceived as more negative compared to other purposes such as debt reduction or the needs for insiders to sell shares. However, the correlation between EXP and the Three-Day SRAR is negative indicating that for days −10 to −3 firms may be doing things differently when the purpose of the offering is for expansion. Whatever is being done serves to offset the negative response that occurs for days −2, −1, and 0. Thus, we cannot necessarily find evidence against signaling theory (Myers and Majluf 1984; Miller and Rock 1985; Brennan and Kraus 1987) that suggests more negative SRARs for expansion purposes. Second, the coefficient for SIZ for the Eleven-Day SRAR test has its predicted positive sign and is significant at the 5% level. Consistent with Bhushan (1989), this indicates that the negative market response can be lessened by firms with greater sizes.

Finally, the results for CIL are not significant. For the Eleven-Day SRAR test, CIL has a negative sign that is opposite of that predicted by Leland and Pyle (1977). For the Three-Day SRAR test, CIL has its predicted positive coefficient but remains insignificant. Overall, the results for CIL do not support the Leland and Pyle signaling theory as there is no evidence that the change in inside ownership exercises a significant influence on short-run returns. In light of the results for ILA and CIL, we conclude that the short-run response is determined more by the absolute level of inside ownership than the change in this absolute level.

From the Eleven-Day SRAR and Three-Day SRAR results in Table 5, we offer the following conclusion. Firms undergoing SEOs will experience less negativity at the time of its SEO announcement and for up to ten days prior to that if these five main conditions are present: (i) the SEO has less underpricing; (ii) the firm has less inside ownership; (iii) the firm has been more profitable during the fiscal year prior to the SEO; (iv) the firm’s stock has experienced less price volatility for the two years prior to its SEO; and, (v) the SEO occurs after a bubble period (where stock price run-ups can be less than what is experienced during the bubble period).

5.2 Results for Ten-Day SRARs and Twenty-One-Day SRARs

As seen in Table 5, the results for the Ten-Day SRAR test (days +1 to +10) differ from the Eleven-Day SRAR and Three-Day SRAR tests. The one major exception is for UND, which once again has its predicted coefficient and is significant at the 1% level. The two insider variables (ILA and CIL) have their predicted signs but are insignificant. Both TIM and RSD change their signs for the Ten-Day SRAR test but only TIM is significant. Neither TIM nor RSD has its predicted sign. Besides UND, and TIM, two other independent variables have significant coefficients: RSZ and GRO. While GRO has its predicted positive sign indicating that growth firms can be received more positively immediately after an SEO announcement, the positive sign for RSZ is not like that predicted in Table 4. Inconsistent with signaling theory, we conclude that the relative greater offering sizes (that include both primary and secondary shares) do not signal greater negative news as judged by the Ten-Day SRAR test.

The Twenty-One-Day SRARs cover a period of about a calendar month surrounding SEOs. Table 5 reveals that the results for ILA, RSD, UND, PFT, and EXP are alike for both Three-Day SRAR and Twenty-One-Day SRAR tests in that all variables have coefficients that are not only of the same signs for both short-run tests but are also statistically significant. The insignificant coefficient sign for TIM for the Twenty-One-Day SRAR tests is like its sign for the Ten-Day SRAR tests. We conclude the firms undergoing SEOs will perform better for the 21 days surrounding SEOs if they undergo less underpricing, have less inside ownership, exhibit more pre-SEO profitability, are more likely to use proceeds for expansionary purposes, and have less pre-SEO price variability.

When one looks at the change in the inside ownership level (CIL) results in Table 5, we find that our multivariate results for ILA and CIL re-enforces the univariate analysis of Hull et al. (forthcoming). In brief, like their investigation, we find that (i) greater levels of insider holdings are associated with more negative short-run returns and (ii) the degree of the decrease in the inside ownership has no obvious impact on short-run returns that accompany SEOs. While insignificant in short-run tests, we will see in the next section that the change in the inside ownership level (CIL) is a dominant variable for long-run tests that looks at post-SEO price behavior.

6 Regression results for LRAR variables

Table 6 provides long-run regression results for three LRAR dependent variables for both the “full” sample and the “partial” sample. Like a SRAR test, a LRAR test can lose up to 31 observations due to insufficient Compustat data. As noted in Table 2, the partial sample also loses observations without full monthly return data. Table 6 reveals that similar LRAR results occur for both the full and partial samples with greater likeness in results occurring for the post-SEO LRAR test. This is the test where fewer observations are lost for the partial sample. As was true for SRAR tests, VIFs are very low for LRAR tests and so they are not reported (the maximum VIF is only 1.78). Nevertheless, as described toward the end of Section 4.1, collinearity tests were conducted when warranted. The more noteworthy of these tests are incorporated into our results in Table 6. The last column in Table 6 reports adjusted R2 and F values for each LRAR test. We only report adjusted R2 values for LRAR tests because R2 values are once again similar to adjusted R2 values. The greatest adjusted R2 and F values of 0.28 and 24.8, respectively, occur for the Pre-SEO LRAR (months −24 to −1) for the partial sample test. The Four-Year LRARs (months −24 to +24) generate the lowest adjusted R2 of 0.06 for the full sample test and the lowest F value of 7.15 for the partial sample test.
Table 6

Long-run regression results

Dependent Variable: Pre-SEO LRAR (months −24 to −1): Full sample (n = 681) followed by partial sample (n = 500)

Const

ILA

CIL

TIM(R)

RSD(R)

LEV(R1)

PFT(R)

TBQ(R)

SEC(R)

AdjR2/F

1.045

−1.680

−0.557

−0.326

25.962

−1.771

1.619

0.063

−0.720

0.14

9.99***

−4.35***

−0.51

−1.85**

5.67***

−4.80***

4.07***

7.60***

−4.05***

15.1***

Const

ILA

CIL

TIM

RSD(R)

LEV(R1)

PFT(R)

TBQ(R)

SEC(R)

AdjR2/F

2.034

−0.630

−0.855

−0.483

69.026

−3.098

2.279

0.166

−0.465

0.28

9.79***

−1.36*

−0.68

−2.47***

9.80***

−7.40***

4.77***

10.59***

−2.21***

24.8***

Dependent Variable: Post-SEO LRAR (months +1 to +24): Full sample (n = 675) followed by partial sample (n = 631)

Const

ILA

CIL

TIM(R)

RSD(R)

LEV(R2)

EXC(R)

EXP

PTA(R)

AdjR2/F

−0.257

−0.176

−1.232

0.172

−8.452

0.506

0.187

−0.083

0.314

0.14

−4.02***

−1.38*

−3.55***

2.93***

−5.29***

4.29***

2.72***

−1.30

2.71***

14.4***

Const

ILA

CIL

TIM(R)

RSD(R)

LEV(R2)

EXC(R)

EXP

PTA(R)

AdjR2/F

−0.264

−0.185

−1.322

0.173

−8.233

0.559

0.181

−0.085

0.332

0.14

−3.98***

−1.41*

−3.66***

2.86***

−4.96***

4.55***

2.54***

−1.35*

2.79***

14.1***

Dependent Variable: Four-Year LRAR (months −24 to +24): Full sample (n = 676) followed by partial sample (n = 456)

Const

ILA

CIL

TIM(R)

RSD(R)

RSZ

PFT(R)

  

AdjR2/F

0.659

−1.227

−2.580

0.379

−13.753

1.316

1.388

  

0.06

2.81***

−2.86***

−1.96**

1.84**

−2.79***

1.74*

3.49***

  

8.51***

Const

ILA

CIL

TIM(R)

RSD(R)

RSZ

PFT(R)

TBQ

 

AdjR2/F

0.204

−0.829

−2.744

0.344

−10.267

2.073

2.912

0.081

 

0.09

0.59

−1.38*

−1.50*

1.11

−1.13

1.98**

4.82***

4.26***

 

7.15***

TIM(R): A residual nearly perfectly correlated with TIM, but not correlated with RSD, was used in the test.

Pearson and Spearman rhos between TIM and RSD are −0.38 and −0.40.

RSD(R)): A residual nearly perfectly correlated with RSD, but not correlated with LEV, was used in the test.

Pearson and Spearman rhos between RSD and LEV are −0.48 and −0.58.

LEV(R1): A residual nearly perfectly correlated with LEV, but not correlated with TBQ, was used in the test.

Pearson and Spearman rhos between LEV and TBQ are −0.41 and −0.89.

PFT(R): A residual nearly perfectly correlated with PFT, but not correlated with RSD, was used in the test.

Pearson and Spearman rhos between PFT and RSD are −0.42 and −0.41.

TBQ(R): A residual nearly perfectly correlated with TBQ, but not correlated with RSD, was used in the test.

Pearson and Spearman rhos between TBQ and RSD are 0.50 and 0.58.

SEC(R): A residual nearly perfectly correlated with SEC, but not correlated with ILA, was used in the test.

Pearson and Spearman rhos between SEC and ILA are 0.13 and 0.14.

LEV(R2): A residual nearly perfectly correlated with LEV, but not correlated with TIM, was used in the test.

Pearson and Spearman rhos between LEV and TIM are 0.31 and 0.36.

EXC(R): A residual nearly perfectly correlated with EXC, but not correlated with LEV, was used in the test.

Pearson and Spearman rhos between EXC and LEV are 0.44 and 0.47.

PTA(R): A residual nearly perfectly correlated with PTA, but not correlated with EXC, was used in the test.

Pearson and Spearman rhos between PTA and EXC are 0.27 and 0.23.

This table gives statistical results for long-run regression tests. Our general regression model for these tests is described in Section 4.1 and allows each dependent LRAR variable to be used with four independent variables of concern (ILA, CIL, TIM, and RSD) and any other independent variable described in Table 3 that is significant for the LRAR being tested. The first set of results (for each dependent variable) is for the full sample and the second set of results are for the partial sample. The first column gives the constant’s coefficient with its t statistic below. Subsequent columns for independent variables provide coefficients with the corresponding t statistics below. Except for the constant variable, the t-test is one-tailed because each independent variable has a definite predicted sign for its coefficient for each test (see Table 4 for each predicted sign). The last column gives the adjusted R2 value with the F value below. The symbols *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively.

6.1 Results for Pre-SEO LRARs

Table 6 begins by reporting results for Pre-SEO LRARs. Despite the fact that 181 observations (26.6% of the full sample) are lost for the partial sample test, the independent variables that enter this test are identical to the full sample test. Except for ILA and SEC, all coefficients have their predicted coefficient signs as given in Table 4. The significant negative coefficients for ILA do not bear out our prediction that SEO firms with greater inside ownership take advantage of greater price run-ups but indicate they perform worse prior to their SEO announcements. The significant negative coefficient for SEC fails to find that greater secondary selling through an SEO follows a greater pre-SEO stock price run-up. Except for CIL, all independent variables have coefficients that are significant at the 1% level for at least one of the Pre-SEO LRAR tests. Consistent with the univariate analysis of Hull et al. (forthcoming), the change in the inside ownership (captured by CIL) has no significant relation with prior stock price run-ups. Thus, greater decreases in inside ownership (attained through SEOs) are unrelated to greater stock price run-ups. We infer that inside ownership behavior at the time of an SEO is not associated with (and thus cannot be predicted from) the value of the Pre-SEO LRAR.

Given the significant coefficients for all independent variables except CIL, we offer the following conclusion about pre-SEO stock price performance. Firms have superior pre-SEO stock price run-ups when they manifest the following seven conditions: higher Tobin’s Q ratios prior to the SEO (where greater price-run-ups help cause higher Tobin’s Q ratios); experience more variability in stock prices for the two years before an SEO (where the higher variability can be associated with greater stock price run-ups); have lower leverage ratios (which can be related to the greater pre-SEO stock price run-up); exhibit greater profitability (from a book standpoint) for the fiscal year before the SEO; have less secondary selling by current owners that accompanies the SEO; have lower inside ownership; and, occur during a bubble period.

6.2 Results for Post-SEO LRARs

Table 6 next gives results for Post-SEO LRARs (months +1 to +24). Once again, the results for both the full and partial samples mirror one another. The even greater similarity between the full and partial samples (for Post-SEO LRARs relative to Pre-SEO LRARs) can be attributed to the fact that the partial sample for the Post-SEO LRAR test only loses 44 observations (6.5% of the full sample). Except for the negative coefficient for CIL (which is significant at the 1% level), all coefficients for the Post-SEO LRAR tests have their predicted signs as given in Table 4. The positive prediction for CIL was based on Leland and Pyle (1977) who hypothesize that greater decreases in inside ownership should signal inferior future performance. Clearly, our results are inconsistent with this prediction. We conclude that decreases in inside ownership are done for other reasons that dominate signaling. These reasons can include (i) diversification needs for those insiders who are selling; (ii) the exercise of expiring options by insiders where their gains are greater than any loss that might be expected to occur from their selling; and (iii) the insider unfounded fear that share prices will fall in the long-run due to poor prospects.

Besides CIL, five other variables are significant at the 1% level for both tests: TIM, RSD, LEV, EXC, and PTA. Thus, once again the time period of occurrence (TIM), risk as captured by pre-SEO stock price variability (RSD), and leverage (LEV) are important determinants of long-run returns. However, the coefficient signs for all three variables are opposite of their pre-SEO LRAR signs. This indicates that the influence of the time period, pre-SEO price variability, and leverage all undergo a dramatic change in their influence on stock prices. EXC and PTA enter a regression test for the first time as significant variables. Thus, the Post-SEO LRAR test is the only test of importance that involves (i) the listing of the firm’s stock (NASDAQ versus NYSE/AMEX) and (ii) growth as represented by the relative amount of tangible assets (as measured by net plant and equipment to total assets). There are two variables for the Post-SEO LRAR tests that significant: ILA and EXP. The negative coefficients for ILA indicate that firms with higher inside ownership perform worse after an SEO. The very marginal results for EXP indicate some weak support for the notion that firms will have poorer post-SEO performance if the purpose of the SEO is related to expansion.

From our regression analysis of long-run post-SEO stock price behavior, we offer the following conclusion. Firms will have superior post-SEO stock price returns when they exhibit the following six conditions: less pre-SEO stock price volatility; more leverage at the time of their SEOs; greater decreases in inside ownership; occurrence of SEO after a bubble period; greater profitability for the year prior to their SEOs; and, NYSE/AMEX listing.

6.3 Results for Four-Year LRARs

The last set of results in Table 6 is for Four-Year LRARs (months −24 to +24). Despite being overall similar, the “full” and “partial” results for the two Four-Year LRAR tests differ more than the pre-SEO and post-SEO LRAR tests. This can be explained by a greater number of missing observations for the longer time period of four years.

The insignificant coefficient for TIM for the partial sample test is not surprising given the loss of 220 observations and the fact that it had significant negative coefficients for the Pre-SEO LRAR tests and significant positive coefficients for the Post-SEO LRAR tests. Similar conclusions could be voiced for the variables RSD and LEV (where the latter is not shown in Table 6 because it does not have a significant coefficient for the Four-Year LRAR tests).

Unlike pre-SEO and post-SEO results, coefficient predictions for the two Four-Year LRAR tests are more likely to be wrong as seen from comparing the signs for CIL, RSZ, and TBQ in Table 6 with their predicted signs in Table 4. The results for ILA, CIL, RSZ, and PFT all render similar significant levels for both the full and partial sample tests. However, RSD and TBQ present a noteworthy difference because each is only significant for one of the two tests.

From our investigation of long-run price behavior surrounding SEOs, we offer the following conclusion. In order of statistical significance, firms will perform better for the four years around SEOs when the following four conditions are present: greater profitability before an SEO; lower inside ownership levels; relatively greater offering sizes; and, greater decreases in inside ownership brought about by their SEOs. It also appears there is some support that superior long-run returns can be associated with a larger Tobin’s Q, less pre-SEO price variability, and occurrence during a post-bubble period.

7 Other tests and future research

Of the other tests that we performed, eight of the most noteworthy are summarized below. First, variations of the fifteen independent variables given in Table 3 were computed. The results did not change significantly from those presented in this paper. Second, we repeated our tests by deleting observations typically omitted from tests such as utilities, financials, and repeat observations. The similar results (with or without deletions) are expected given these observations are small in number. Third, White (1980) t statistics were also computed and were found to be similar to OLS t statistics indicating heteroscedasticity is not a problem.

Fourth, we investigated other insider variables. For example, we replaced the “Inside Ownership Level After” (ILA) with the “Inside Ownership Level Before” and repeated tests, but the results were similar. This is because those firms with large holdings before generally have large holdings afterwards as reflected in their correlation coefficient 0.95 for both the Pearson and Spearman tests. We also examined if our insider results for our two insider variables (Change in Inside Ownership Level and Inside Ownership Level After) could be better explained by the “director and officer group” by itself or the “other five percent or more ownership group” by itself. When we tested each separately, they generally gave similar results albeit weaker when tested individually versus tested together. This suggests that using both groups together offers a more powerful test. While these tests indicate our general conclusions hold, we still recommend that future research build on this study’s findings by better cataloging the effects of the “director and officer group” and the “other five percent or more ownership group.”

Fifth, we chose to report results using winsorized variables when extreme outliers were found. However, our main conclusions and general results are robust regardless of whether winsorizing is used. Sixth, one might argue that for a regression analysis, we want to know how the independent variables respond to raw returns. This is because any abnormal return methodology contains the cumulative raw returns. Whatever compounded expected return is subtracted out may be irrelevant even if one used a methodology that is more biased (albeit within the literature one can find strong claims about biases for any methodology). In conducting tests with just raw compounded returns, we found that results were similar with one exception: the R2 values increased for long-run tests when using just raw returns. For the post-SEO long-run test, the adjusted R2 jumped from 0.14 to 0.23 for both the full and partial sample tests (compared to jumps of only about 0.02 for other long-run tests).

Seventh, long-run regression tests for different periods were performed other than two years before SEOs, two years after SEOs, and four years around SEOs. While our tests for these years confirmed many of the general results presented in this paper, a more detailed analysis would be needed to explore if other regression models could be formed to better account for periods where one year or three years replace the two years before and after that we used. Eighth, for short-run abnormal return tests, we also conducted tests after abnormal returns were adjusted for flotation costs following the methodology used by Hull and Fortin (1993/1994) and Hull and Kerchner (1996). The results were similar. We surmise that adjusting for issue costs tends to have little impact compared to the underlying compounded returns. Thus, the failure of the adjustment for issue costs to influence results is parallel to the failure of the expected compounded returns methodology to affect results. In conclusion, the independent variables are influenced by the underlying compounded raw returns.

We now offer five suggestions for future research. First, given all of the attacks on long-run methodology, future research needs to continue to explore how different methodologies might affect the results reported in long-run return studies. Second, future research can go beyond our research by including the role of an industry. Third, our findings suggest that insider selling at the time of an SEO can be done for multiple reasons that trump any effect tied to Leland and Pyle (1977) signaling theory. Future research might explore the strength of other reasons including: diversification of personal assets, exercise of expiring options, or selling overvalued shares as might be caused by pre-SEO stock price run-ups. Fourth, by reporting both short-run pre-SEO and short-run post-SEO returns, we reveal an opposite price movement for the ten days before and after SEOs. Future research might try to explain these short-run findings. For example, is there short-term manipulation of prices by large institutional investors such as hedge funds? Fifth, our sample is unique in that it is characterized by observations with large inside holdings. Future research can test some of our conclusions for more diverse samples of inside ownership. For example, one might find support for Leland and Pyle if one tests a sample of SEOs that contain not only large holdings by insiders (which is a sample like what this paper analyzes) but also contains SEOs that would have small (or no) insider holdings. Thus, our lack of multivariate support for CIL can only be said to hold for a sample like what this paper analyzes and may not be found for a sample that contains a wider diversity of inside ownership levels. Getting such a sample may be difficult since prospectuses accompanying SEOs tend to only report inside holdings when there is something substantial to report.

8 Summary

Univariate seasoned equity offering (SEO) research investigates the valuation impact of inside ownership. We extend this research by investigating if the inside ownership level and its decrease are significant factors within a multivariate analysis. We find that the level of inside ownership is a consistent factor in accounting for both short-run and long-run returns surrounding SEOs. The change in the inside ownership level is not a factor influencing short-run returns at the time of an SEO, but is a major factor associated with long-run returns, especially those that occur after SEOs. While we find these factors to often be statistically significant and influential in determining stock price behavior around SEOs, there are other variables that can be just as influential. Below we describe the impact of such variables in accounting for both short-run and long-run stock returns.

For short-run regression tests, we find a number of factors that can influence (in varying degree) stock returns for a variety of short-run periods. Of particular relevance are two short-run periods: (i) the three-day announcement period, and (ii) the twenty-one-day period (about a calendar month) surrounding SEOs. The four major factors associated with superior stock returns for these two short-run periods are: (i) lower underpricing (degree that the offer price is set below the estimated offer price); (ii) greater profitability for the fiscal year prior to SEOs; (iii) smaller inside ownership levels; and, (iv) less pre-SEO price variability (as measured by the daily standard deviation of stock returns for two years before SEOs). Not included in this list is the occurrence of the SEO during or after the internet-technology bubble period. However, it can have a very significant impact if we look at either a short-run period before and including day 0 or a period that includes the ten days after an SEO. Similarly, the pre-SEO price variability variable has a more pronounced effect on stock prices if we look at either its pre-SEO or post-SEO impact on short-run returns.

For long-run tests for both the full and partial samples, we find four major independent variables linked to superior returns for the four years surrounding an SEO. These four variables are: (i) greater profitability prior to SEO; (ii) lower inside ownership level; (iii) greater relative size of the offering; and, (iv) greater decrease in inside ownership level. Three other possible factors contributing to superior long-run returns surrounding SEOs are less pre-SEO price variability prior to SEOs, the occurrence of an SEO after the internet-technology bubble has ended, and greater Tobin’s Q values,.

While the regression tests we offered are all highly significant, they can still differ in R2 and F values. In general, the three-day test for short-run returns and the two-year pre-SEO long run test for the partial sample do the best job of accounting for stock price behavior associated with SEO announcements. However, for the two-year pre-SEO tests, the independent variables cannot be viewed as having predictive power about the future as their values do not occur prior to the pre-SEO period.

Our multivariate results for our two insider variables are consistent with prior univariate research. First, we verify univariate findings that the inside ownership level does have a negative impact on short-run returns as predicted by Myers and Majluf (1984). We extend this short-run finding by showing that this negative effect also occurs for long-run returns. One might think that insiders are encouraging SEOs because they want to capitalize on prices that are temporarily elevated and that this could be somewhat destructive in the long-run and may even lead to some post-SEO negativity. However, this argument is countered by valid reasons for why insider might want to encourage an SEO. These reasons include (i) diversification needs for those insiders who are selling; (ii) the exercise of expiring options by insiders where their gains are greater than any loss that might be expected to occur from their selling; and (iii) the insider unfounded fear that share prices will fall in the long-run due to poor prospects. Together such reasons can collectively account for why insiders would be encouraging many of the SEOs. Of noteworthy importance, if shares are overvalued, then insider behavior that encourages SEOs is not destructive but wealth-enhancing. This is because insiders can profit when new owners buy overvalued primary shares that are offered, while insiders sell their own overvalued shares through the secondary component of the offering.

Second, we authenticate the univariate findings that the change in inside ownership has no noticeable impact on short-run returns, but the change can influence long-run returns in ways not consistent with Leland and Pyle (1977) signaling theory. The inconsistency occurs when firms with greater decreases in inside ownership do relatively better for the two years after an SEO. However, given the fact that the average stock performance after an SEO is negative, one cannot definitively say that insiders (who use an SEO to lower their levels of ownership) are worse off on the average. We are left to infer that the motivation for insider selling at the time of an SEO is complicated and involves a number of reasons (diversification, exercising expiring options, and fear of poor future prospects) that could all be self-serving.

Copyright information

© Springer Science+Business Media, LLC 2010