Repurchase Tender Offers and Earnings Information

Transcription

1 Repurchase Tender Offers and Earnings Information Larry Y. Dann University of Oregon Ronald W. Masulis Vanderbilt University and David Mayers * Ohio State University Abstract: Announcements of stock repurchase tender offers are examined as a source of information to the market on the firm's future earnings prospects and market risk level. We find positive average earnings surprises and equity systematic risk reductions following tender offers but not, in most instances, preceding them. We find positive stock price reactions to tender offer announcements to be positively correlated with earnings surprises over the concurrent and subsequent two years, and negatively correlated with changes in equity and firm market risk. Finally, stock price reactions to quarterly earnings announcements are more strongly correlated with time-series based earnings surprises in the year prior to the tender offer than during the subsequent year, consistent with tender offer announcements conveying earnings information to the market. First Draft: November 1988 Latest Revision: March 8, 1991 * We wish to thank Sanjai Bhagat, Rob Heinkel, Jonathan Karpoff, Vojislav Maksimovic, Wayne Mikkelson, Ed Rice, Katherine Schipper, A.J. Senchack, Laura Starks, Seha Tinic and the finance workshop participants at Colorado (Boulder), Kent State, Indiana, Purdue, Texas A&M, Texas (Austin), Pittsburgh, Toronto, Maryland, The International Finance Conference at Groupe HEC, Jouy-en-Josas, France, and the Pacific Northwest Finance Conference for helpful comments.

2 1. Introduction A central issue in corporate finance and financial accounting is whether changes in a firm's financial structure release information about its prospective economic performance. Theoretical work of Ross [1977], Leland and Pyle [1977], DeAngelo and Masulis [1980], Vermaelen [1984], Miller and Rock [1985], Harris and Raviv [1985], John and Williams [1985], Ofer and Thakor [1987], and Lucas and McDonald [1990], among others, provide linkages between financial structure changes and firms' discounted expected earnings. Predicted earnings changes have been subjected to alternative structure changes in studies by Vermaelen [1981, 1984], Ofer and Natarajan [1987], Healy and Palepu [1988, 1990], Ofer and Siegel [1988], Lys and Sivaramakrishnan [1988], Israel, Ofer and Siegel [1989, 1990], and Hansen and Crutchley [1990]; in addition, Healy and Palepu [1990] explicitly test for changes in firm risk. This paper extends this literature by examining whether announcements of issuer tender offers (stock repurchases) release information about repurchasing firms' current and future earnings and their market risk. Dann [1981] and Vermaelen [1981] conclude that new information associated with announcements of stock repurchases is an important determinant of stock price behavior at that time. Yet evidence of subsequent improvements in firm performance is limited. Vermaelen [1981] reports some initial evidence, but provides no direct tests linking tender offer announcement stock returns with future profitability or changes in the level of market risk, and neglects a comparison with an industry control sample. 1 1 Recent related work by Bartov [1991] examines open-market repurchases while Hertzel and Jain [1991] analyze the effects of repurchase tender offers on analysts' earnings forecasts. 2

3 We subject the hypotheses that stock repurchases precede increases in firm earnings or decreases in market discount rates to a battery of tests using our stock tender offer sample. We find evidence of positive abnormal annual earnings in the years following repurchase using two measures of earnings. We also find a significant reduction, on average, in both asset and equity systematic risk (beta) around the repurchase, but are unable to document completely the source of this reduction. Further, we document a statistically significant cross-sectional relation between stock price reactions to tender offer announcements and both post-repurchase annual earnings realizations (adjusted for the time series trend in prior earnings realizations) and the shift in equity beta estimated before and after the repurchase announcement. We also provide evidence, by analyzing stock price reactions to quarterly earnings announcements, which is consistent with the hypothesis that investors revise their earnings expectations upward after the tender offer announcement. The evidence from our set of tests yields a consistent inference: issuer tender offer announcements typically precede increases in firm earnings (which exceed the level forecasted by past earnings trends) and decreases in market risk, and the stock market appears to capitalize these changes at the time of the tender offer announcement. In exploring the question of changes in market risk, we uncover evidence that, on average, a repurchasing firm's market risk declines both prior and subsequent to the tender offer announcement. Further, the stock repurchase is found to have, on average, a negligible effect on a firm's financial leverage. This suggests that tender offer announcements signal a decline in the market discount rate on firm earnings as well as an expectation for higher future earnings. Lastly, we uncover evidence consistent with an adverse selection effect in stock repurchases as predicted by Lucas and McDonald [1990]; specifically, abnormal stock returns are on average negative in the period 3

4 preceding stock repurchases, and further, tender offer announcement abnormal returns are more positive for those stocks which are preceded by abnormal negative performance. This is consistent with the hypothesis that managements tend to repurchase stock when the stock is undervalued relative to their superior private information. Other explanations for the positive stock price reactions to repurchase announcements exist; such as, the gain in firm value from reducing management's discretionary spending or free cash flow (see Jensen [1986]), wealth redistributions from debt to equity induced by the rise in leverage (see, for example, Galai-Masulis [1976]), or the increase in stock price due to the shift in the stock's aggregate supply function in the face of a downward sloping aggregate demand function (see Gurel-Harris [1986], Shleifer [1986], and Bagwell [1989]), which are not contradicted by the evidence in our study. However, we document a systematic relation between stock price behavior at the repurchase announcement and post-repurchase earnings performance and changes in market risk, which is an unlikely event if other factors were the dominant explanation for the valuation effects of stock repurchases. In the next section we identify our sampling procedures and provide descriptive statistics of the firms that comprise our sample. Section 3 describes the earnings forecast models and earnings measures we use. The empirical tests and results are reported in Section 4. Our summary and conclusions are in Section Data Description and Abnormal Stock Return Evidence Sampling criteria. The sample we employ in this study consists of 122 repurchase tender offers by 101 firms during the period, and is derived from a 4

5 comprehensive set of issuer tender offers by NYSE and AMEX listed firms used in earlier studies by Dann and Masulis. 2 This sample period has the advantage that it ends prior to when firms were known to employ tender offers to deter hostile takeovers. 3 The preliminary sample was compiled from information in the Wall Street Journal Index, the Corporate Financing Directory, various Moody's Manual News Reports and the Capital Changes Reporter. Tender offer announcement dates and descriptions of offers were obtained from the above sources as well as individual firm "offers to purchase" and press releases. Daily common stock price and returns data were obtained from the CRSP NYSE/AMEX daily master and return files. All tender offers in the preliminary sample are successfully completed, cash only offers, which are not restricted to odd lot shareholders, where the common stock was listed on the NYSE or AMEX from the tender offer announcement through the offer's expiration date. To be included in this study, earnings data for the repurchasing firm must be available for either: (i) a minimum of 6 years before through 3 years following the tender offer announcement from the Compustat Annual Industrial or Research Files; or (ii) a minimum of 15 quarters before through 4 quarters following the tender offer announcement from Compustat's Quarterly Industrial File. Firms with fiscal year changes that cause a gap in the time series of earnings in (i) or (ii) above are excluded from the sample. The resulting final sample consists of 122 offers of which 78 observations meet both criteria (i) and (ii) above, 110 satisfy (i), and 90 meet (ii). 2 For further description of these data characteristics, refer to Masulis [1980] and Dann [1981]. 3 Our sample offers appear to be clean of this potentially confounding motivation. 5

6 Earnings announcement dates were obtained from the Compustat Price-Dividend-Earnings (PDE) file and verified with the Wall Street Journal Index. Table 1 presents a time profile by year and month of the sample of 122 issuer tender offers. Monthly and annual totals are reported in the last column and row, respectively. The table shows that the sample is concentrated in the years and and in the last four months of the year. We also explore whether there are any seasonalities in our sample of tender offer announcements. We find announcements to be equally frequent in the first and second halves of a given month and are slightly more frequent on Wednesdays and Thursdays than other days of the week. Table 2 contains descriptive statistics for our sample on prior equity capitalization, tender offer induced percent changes in common stock outstanding, the dollar size of the tender offers, and the tender offer premium as a percent of the pre-announcement stock price. The mean and median figures on stock capitalization prior to the offer announcement are $53 million and $29.7 million respectively. Our decile figures, when compared to those in Keim's (1983) analysis of CRSP-listed firms, indicate that our sample tends to be drawn from smaller firms in the overall population of NYSE- and AMEX-listed stocks. The positive skewness reflects IBM's presence in the sample. The mean and median percent changes in common stock outstanding due to the offer are 16% and 13% respectively. These are slightly higher than those found in the earlier studies of Dann, Masulis, and Vermaelen. The mean and median offer sizes are $26.2 million and $6.1 million respectively. The tender offer premium as a percent of stock price two days prior to the Wall Street Journal announcement date have mean and median values of 24% and 21% respectively. These are again slightly higher than the figures reported in the earlier studies. 6

7 Evidence for abnormal stock returns. The magnitudes of abnormal stock returns (prediction errors) for firm tender offer announcements are estimated by standard event study methods. For each tender offer, we calculate a one factor market model, using continuously compounded stock and market returns, estimated over the 350 trading days ending 2 trading days before the initial tender offer announcement. 4 The CRSP daily equal-weighted returns series is our measure of the market return. The announcement period is defined as the two-day period encompassing the Wall Street Journal announcement date and the prior trading day, since this is the shortest interval that will capture the market's reaction to a firm announcement. Panel A of Table 3 reports that the announcement stock prediction errors have a sample mean of 17.68%, with a z-statistic of which indicates a highly significant increase in stockholder wealth. The median prediction error is 15.32%, and a remarkable 120 of 122 prediction errors are positive. Panel B presents the frequency distribution of the stock return prediction errors for the tender offer announcements. The panel uncovers considerable dispersion in the size of the prediction errors, while at the same time documenting a striking uniformity in their sign. This evidence is clearly consistent withtender offers releasing positive information about the firm's future earnings prospects but is far from conclusive. For example, one alternative explanation for the positive repurchase announcement effect which has no earnings implications, is that the stock price rise represents the effect of short term arbitrageurs seeking to profit from the firm's tender offer, which can be viewed as the distribution of a short lived in-the-money European put option to stockholders. 4 For five offers, market model parameters are estimated over the 350 trading days beginning 2 trading days after expiration of the offer because the CRSP daily return file contains an insufficient number of returns prior to the offer. 7

8 Earlier investigations by Dann, Masulis and Vermaelen document large (14%-16%) and statistically significant positive average stock price reactions at the announcement of issuer tender offers. Comparing those results with Table 3 indicates that the abnormal stock returns found here are similar though larger and more uniform across the sample than reported in earlier studies. This similarity provides some assurance that our earnings availability requirement does not render our subsample of tender offers unrepresentative of the offers investigated in earlier studies. We also examine stock return patterns in periods around the tender offer announcement and find evidence consistent with predictions of the Lucas and McDonald [1990] adverse selection model. Lucas and McDonald argue their model of the stock issuance process can be applied to share repurchases: they assume that firms periodically find it optimal to repurchase shares (to distribute cash in excess of current investment opportunities, for example), and argue that managers of firms with overvalued shares have incentives to delay a repurchase until their negative private information becomes public. In contrast, managers with undervalued shares have incentives to repurchase immediately. It follows that stock returns are predicted to be on average negative prior to a repurchase announcement. An ancillary prediction is that firms experiencing prior price rundowns will realize larger repurchase announcement gains. To test these predictions, we calculate cumulative stock returns (excess of the market) for the 120 trading days preceding the repurchase announcement and average them across our sample. The results are a significant -4.7% (p-value = 0.021) average cumulative excess return for the full sample (median = -5.7%, p-value = 0.002). Thus, stock prices fall prior to the repurchase announcement as predicted by Lucas-McDonald. 8

9 To test the second prediction we split the sample by whether the prior 120 day cumulative excess return is negative or non-negative. For the stocks experiencing a negative excess return over this prior period, the repurchase announcement average excess return is 19.7% (median=15.6%) and represents 77 of the 122 observations. For the shares having positive prior excess returns, the repurchase announcement average excess return is 14.3% (median = 14.3%). The difference in the means is significant (p-value = 0.025) and the median test (Wilcoxon) indicates marginal insignificance (p-value = 0.11). Thus, stocks with negative prior abnormal performance have larger positive announcement effects consistent with the adverse selection model of Lucas and McDonald [1990]. 3. Earnings Forecast Models and Earnings Definitions 3.1 Earnings Forecast Models The principal thrust of our investigation involves an analysis of unexpected annual earnings for several years around issuer tender offers. We also investigate quarterly earnings for the year before and after the tender offers in tests more suited to quarterly data. We estimate the market's expectations of each firm's earnings based upon time-series models of past earnings realizations. To determine appropriate univariate time series models for earnings realizations, we rely on the extant literature which has explored the time-series properties of earnings figures for NYSE and AMEX listed stocks. For any given earnings forecast model, we define an earnings forecast error, fe(e t ), as the difference between the actual earnings realization, E t (after any correction by the firm) and the corresponding earnings forecast, f(e t ); i.e., fe(e t ) = E t - f(e t ). 9

10 The major evidence on the statistical properties of annual earnings figures is reported in Ball-Watts [1972], Lookabill [1976], Albrecht-Lookabill-McKeown [1977] and Watts-Leftwich [1977] who find that an effective and parsimonious representation of the time-series process can be obtained from a martingale model, i.e., (1) f (E t ) = E t-1 or a similar expression with a drift term (a submartingale model). We use both the martingale and submartingale models to estimate the market's expectations of each firm's annual earnings. For quarterly earnings data, studies of ARIMA and other time series models by Foster [1977], Griffin [1977], Watts [1978], Brown-Rozeff [1979], Bathke-Lorek [1984] and Brown-Hagerman-Griffin-Zmijewski [1987a, 1987b] find that several simple autoregressive processes capture both the seasonal and other major time-series properties of the data. Among these, Foster's model of quarterly firm earnings expectations is especially appealing because of its computational simplicity. Specifically, this model can be represented as (2) f (E t ) = E t-4 + θ (E t-1 -E t-5 )+ δ. This model is estimated by fourth differencing the time series of quarterly earnings, then regressing the fourth-differenced earnings at time t on lagged fourth-differenced earnings. The model's parameters and forecast errors are sequentially estimated using a constant number of lagged earnings figures which are updated for each subsequent forecast period. In addition to the Foster model, we study quarterly earnings forecasts derived from firm earnings for the same quarter of the prior year, both with and without a drift term (which we call the quarterly submartingale and martingale models), and the instrumental variables approach suggested by Brown-Hagerman-Griffin-Zmijewski [1987b]. Evidence from estimating these 10

11 models on our sample of firm earnings figures indicates that these specifications characterize the data reasonably well Earnings Definitions There are several possible measures of earnings or cash flows that might be appropriate for evaluating the predictions of the previously noted corporate finance theories. We employ annual earnings before interest and taxes (EBIT) and annual and quarterly earnings per share (EPS) before extraordinary items and discontinued operations because of (1) constraints imposed by data availability, (2) the ability to compare our results with previous research on the topic, (3) the existence of evidence on the time series properties of these alternative cash flow and earnings figures, and (4) the desire to minimize the direct effects of the tender offer and its financing sources on measured earnings. Earnings measured by either EBIT or EPS can be affected by how the tender offer is financed, both in the accounting period encompassing the tender offer and for some cases in subsequent periods. Specifically, if the tender offer is debt-financed, current EBIT is unchanged, but current and future EPS increase if the percentage decline in aftertax earnings is exceeded by the percentage reduction in shares outstanding; and an increase is likely, since the firm's after tax interest rate on debt is lower than its after tax expected rate of return on investment. If the tender offer is financed by a reduction in 5 For example, our stock prediction error and scaled earnings forecast error correlations are comparable to those reported in earlier studies. An alternative specification which adjusts for market-wide effects is to subtract out the average earnings of the firms making up the S&P 400 industrial index, dfe(e t ) = E t E mt - [f(e t ) - f(e mt )] where E mt are market earnings in period t (based on S&P 400) and f(e mt ) is the forecast of market earnings in period t based on the above forecasting models. We have employed this alternative estimation method without notable differences in our results, which is also the experience of Vermaelen [1981]. 11

12 cash and near-cash assets, current and future EBIT are expected to be reduced, whereas current and future EPS again will rise if the percentage reduction in before tax earnings is exceeded by the percent reduction in shares; and this is likely, if the liquidated assets have relatively low returns. Thus, forecasts of EBIT based on univariate time-series models will be biased against the hypothesis of positive abnormal earnings, since repurchases financed by cash imply that current and future EBIT will be reduced while debt financed repurchases imply no immediate change. As a consequence, EBIT-based evidence will be emphasized. As the above analysis indicates, EPS forecasts following tender offers (from univariate time-series models) can be downward biased, relative to the firm's actual situation, because of the reduction in shares outstanding. The problem of bias is most relevant for the analysis of quarterly earnings because only EPS figures are consistently available from Compustat's Quarterly Industrial File. However, our earnings forecasts based on the Foster model are updated throughout the test period using the most recent set of prior quarterly EPS figures. This implies some self correction of any tender offer induced bias in EPS figures for quarters 2, 3, and 4 following the tender offer because of the moving average property of the forecast. 6 To counteract any remaining downward bias we also report results using restated quarterly EPS figures where shares outstanding are augmented by the shares repurchased in the 6 For example, the forecast for quarter +2 is f (E +2 ) = E- 3 + θ (E +1 - E -4 ) + δ. The first element in the forecast, E- 3, is too low assuming some upward shift in the EPS series, beginning in quarter +1 (which is caused by tender offer induced fall in shares outstanding), whereas the second term in the RHS of the equation, the difference between EPS figures in quarters +1 and -4, is too large. This assumes the coefficient, θ, is positive. There is no self correction in the forecast for quarter +1 because all of the EPS figures used in f(e +1 ) are prior to the tender offer. Note that our timing convention for this analysis involves no quarter 0. The quarterly EPS announcement just preceding the tender offer is denoted by E-1, while the first quarterly EPS announcement following the tender offer is denoted by E+1. 12

13 tender offer. This adjustment will overcorrect for the above identified direct effects of the tender offer. Since our quarterly earnings tests are sensitive to the level of reported earnings, comparison of results using restated and as-reported earnings per share provides evidence on the importance of the bias problem. A potential problem for any time-series study of firm earnings figures is the distortionary effect of mergers, acquisitions and divestitures. Compustat restates quarterly earnings to an "as-if-combined" basis for the year prior to a merger and restates prior annual earnings for "as many years as possible." Thus, if material year to year changes in the extent of acquisition/merger activity occur, the earnings figures can be distorted. To assess the importance of this problem for our sample, we investigated the frequency of such events for years surrounding our sample of tender offers. We find the frequency distribution over time to be relatively flat across the pre- and post-tender offer intervals. 7 Moreover, the relatively high frequency of these events in our sample and our requirement of a time series of adjacent earnings figures precludes eliminating earnings figures when such acquisition/merger activity occurs. The uniformity of the distribution of acquisition announcements in the years surrounding the tender offer provides some assurance that this will not impart a serious bias on our results. 7 The distribution of total net acquisitions for years surrounding the tender offer year: year net acquisitions less U.S. Ind The first row indicates year relative to the tender offer year, the second (net acquisitions) is the frequency distribution for the total sample of events, the third row is the frequency distribution with one company (representing four repurchase events and a net of 48 acquisitions in tender offer years), U.S. Industries, omitted. These data are collected from Moody's Manuals for all sample firms for the seven year period centered on the repurchase announcement year. 13

14 4. Empirical Evidence 4.1 Earnings Behavior in Periods Surrounding Tender Offers If corporate finance theories which predict that tender offers are associated with subsequent improvements in earnings are to be supported by the data, then earnings forecast errors derived from a univariate time series forecast model of past earnings should be on average positive subsequent to the tender offer. Vermaelen [1981] reports some initial evidence on this issue. He uses annual EPS data from Compustat to proxy for net cash flows per share. Vermaelen calculates standardized forecast errors and then averages these variables across firms for each event year surrounding the tender offer. 8 His evidence is consistent with the unconditional earnings forecast errors in the post offer period being insignificantly different to the earnings forecast errors conditional on the tender offer announcement. Vermaelen finds significant positive standardized EPS forecast errors in years 0, 1, 3 and 5. While Vermaelen concludes that firm annual earnings increase, on average, over the five years subsequent to the tender offer, there are several unresolved questions that indicate a re-examination of the evidence is warranted before we move on to the central focus of our analysis, which is an examination of the link between stock price behavior at repurchase announcement and subsequent earnings and risk changes. For example, Vermaelen uses the Compustat Industrial and PDE databases, which have since been shown to impart positive survivorship biases as discussed in Banz-Breen [1986]. Ignoring the survivorship effect can cause average earnings derived from these files to be upward biased over time. We minimize this bias by merging annual earnings data from both the 8 Vermaelen standardizes forecast errors for each firm by the estimated standard deviation of all forecast errors available for the firm over the period from year -5 through

15 Industrial and Research files and by augmenting the Compustat databases, where feasible, by collecting missing earnings data from the Moody's Industrial manuals to increase sample size. 9 A second question stems from Vermaelen's use of EPS as his measure of earnings. As previously indicated, the reduction in shares outstanding following the repurchase can cause a downward bias in subsequent EPS forecasts and consequently upward biased forecast errors, especially in year 0 and 1. Our study also analyzes the behavior of an alternative earnings measure (EBIT) that is not biased in favor of the hypothesis that earnings increase following repurchases. Further, although Vermaelen controls for market-wide earnings behavior, this may not be adequate if industry earnings trends differ from the overall economy. Our approach is to use a matched pairing of firms of similar size from the same industry to examine earnings behavior. Finally, Vermaelen's study does not indicate whether multiple tender offers by individual firms in nearby periods are included in his earnings analysis. We investigate the potentially confounding effects on reported earnings of multiple tender offers in adjacent years by examining whether these offers (34 offers by 15 companies) are critical to any conclusions to be drawn from this portion of our analysis. Table 4 presents mean and median standardized forecast errors, associated t values, and Wilcoxon Z's for annual EBIT and EPS earnings measures. Forecast errors are calculated as the difference between a firm's realized annual earnings in year t and an associated forecast of these earnings derived from a submartingale or a martingale model. The submartingale model is defined as: K (3) f (E t ) = E t-1 + 1/K Σ (E t-k - E t-k-1 ), k=1 9 When Compustat decides to discontinue collecting data on a company in its Industrial file (say, because the company is going bankrupt), it deletes the company and its earlier data from the Industrial file. These companies that have been deleted from the Industrial file comprise the Research file. 15

16 where the second term on the right-hand-side, (excluded in the martingale model) captures the trend in earnings and K represents the number of prior earnings differences available. The variances of firms' earnings forecast errors tend to be positively correlated with firm size. Standardizing the earnings forecast errors reduces heteroscedasticity in the cross-sectional data and makes these errors more economically comparable across firms. Thus, we employ two procedures for scaling forecast errors. In one, each forecast error is scaled by a measure of the firm's prior earnings variability; in the other, we scale by the stock's closing market value (equity capitalization) taken from two trading days prior to the Wall Street Journal announcement of the tender offer. 10 The results, reported in Table 4, use the submartingale model for forecasting where we standardize with earnings variability and the martingale model where we standardize with equity capitalization. The submartingale results replicate Vermaelen's procedure and the martingale results are more consistent with our regression analysis which follows. The evidence in Table 4 shows large positive average earnings forecast errors for most of the five year period following the tender offer year. Specifically, we find that for both earnings measures, average standardized forecast errors are significantly positive for year 0, and year 3 when we scale by earnings variability. The EPS measure is significantly positive for year 5 as well, but the EBIT measure is not Our measure of earnings variability is the estimated standard deviation of forecast errors. For the preoffer period, forecast errors are divided by the standard deviation of forecast errors estimated over event years -5 through -1. For each post-offer year, forecast errors are scaled by the standard deviation of forecast errors estimated over all years prior to the current earnings announcement. This standardization procedure annually updates the post-tender offer earnings forecasts for changes in earnings variability, but otherwise replicates the standardization procedure of Vermaelen. 11 Consistent with this evidence, we also find that Value Line one year and 3 to 5 year EPS forecasts are on average revised upward significantly around the tender offer announcement period. More detailed evidence can be found in our earlier working paper. 16

17 The non-monotone relation described by our data is not easily interpreted in terms of known corporate finance theories. 12 Yet this evidence is qualitatively similar to that reported by Vermaelen [1981]. Our conclusions would be even more consistent with Vermaelen's if we were to make his assumption that each scaled forecast error is distributed as a (0,1) standardized normal variate. For example, multiplying the EPS mean estimate in year 1 (0.382) by the square root of the sample size produces a 4.01 t-value. Thus, following Vermaelen's testing procedure, we find significantly positive earnings forecast errors in years 0, 1, 3, and 5. Additionally, Bartov [1991], in his examination of changes in various earnings measures around the announcement of open market repurchase programs finds a non-monotone earnings pattern: specifically, he finds positive, negative, and positive average earnings changes in year 0, 1, and 2, respectively. Thus, a nonmonotone relation is not specific to our sample and appears to be associated with stock repurchases whether by tender offer or in the open-market. When we scale EBIT by pre-offer equity value, the evidence is similar with an additional significant positive result for year 2. The median evidence, which is insensitive to outliers, appears to be generally consistent with the mean results where we scale by earnings forecast error variability, and indicates positive performance almost everywhere when we scale by equity value. The qualitative nature of these results are unchanged if offers by firms that make multiple offers within a five year span are deleted. To provide an earnings benchmark for our repurchasing firms, we collect a matched-pair sample of comparison firms for the 110 offers comprising the annual earnings sample. To control for potential industry or economy-wide influences on earnings, for each 12 In worrying about this problem, we insure our EPS numbers are stated on a common (1981) units basis (i.e., adjusted for any subsequent stock splits or stock dividends). 17

18 repurchasing firm we attempt to choose a comparison firm that (i) is in the same industry, (ii) has earnings data available over the same time span, (iii) has the same fiscal year end, and (iv) is closest in size as measured by the book value of assets at the year-end before the repurchase. These criteria are not all fully met from the set of available Compustat-listed firms, but the match is close. 13 Table 5 presents mean and median paired differences of standardized earnings forecast errors, associated t values, and Wilcoxon Z's for annual EBIT and EPS earnings measures in the same format as Table 4. When we scale by earnings variability the EPS measure is significantly positive for years 0 and 3 and the EBIT measure is significantly positive for year 2. When we scale EBIT by pre-offer equity value, the evidence for year 2 holds and there is positive evidence (one-tailed) from the year 3 median as well. The year-by-year cross-sectional distributions (whose means and medians are reported in Table 5) contain large breaks in the data for extreme values when we scale by equity value. We have also calculated Table 5 results with symmetrically trimmed data, where observations in the tails (.025 trim) are deleted from each sample distribution. 14 The major modification this produces, for Table 5, is to enhance the mean results for years 2, 3, and 4 for EBIT scaled by pre-offer equity value. The trimmed means (and t- statistics) are 0.05 (2.21), 0.06 (2.06), and 0.06 (1.76) for years 2, 3, and 4, respectively. These years produce the largest estimated means using the trimmed data. The next largest 13 Other details of the comparison sample are that (1) 13 repurchasing firms from the Compustat Research file are matched with comparison firms from the same file to invoke the same degree of survivorship bias in the comparison sample; (2) in 105 cases the matching firm has the same 4-digit SIC code as the repurchasing firm, and the remaining five cases are matched by 2-digit code; (3) comparison firm earnings data are available for 5 years prior to the repurchase year in all cases, and available over the entire period for which the corresponding repurchasing firm's earnings data is available in 92 cases; and (4) in 102 cases the comparison firm's fiscal year-end is within two months of the corresponding repurchasing firm's fiscal year-end. 14 See Foster [1986] for a discussion of 'trimming' and other methods for handling outlier observations in accounting data. For a more general discussion of robust estimation techniques see, for example, Amemiya [1985] 18

19 estimate occurs in year -4 where the mean and t-value are 0.03 and The estimated means for all other years are insignificant with point estimates no larger than Thus, we find strong evidence, in the trimmed data, of positive abnormal earnings performance in the post repurchase years, after controlling for industry effects. This evidence supports the hypothesis that issuer tender offers contain information about increases in expected future earnings. 4.2 Market Risk in Periods Surrounding Tender Offers Market model regressions are estimated for five separate years around (three before and two after) the repurchase tender offer. Each year is defined to contain 250 trading days. Table 6 contains mean and median statistics for the common stock beta (market model slope parameter estimates), the year-to-year changes in common stock beta, the residual standard error, and the year-to-year changes in the residual standard error for the issuer tender offer sample (Panel A) and for the comparison sample (Panel B). The evidence in Panel A describes a decline in the mean and median beta prior to the repurchase announcement with a further decrease in year plus one and a slight increase in year plus two. The decrease from year minus one to plus one is strongly significant (pvalue < 0.01); whereas, the other year-to-year mean and median changes are at best of only marginal significance. The market model residual variance appears to be constant across the five year observation period, although we provide no formal test. This indicates that on average stocks' idiosyncratic risk is relatively stable over the observation period. This evidence of a change in the level of equity market risk for the issuer tender offer firms 19

20 contrasts with that for the comparison sample in Panel B. That evidence indicates that, for two years on either side of the matching issuer tender offer dates, the comparison sample mean and median betas are essentially constant. The only significant shift in the mean and median betas is a positive one which is between year minus three and minus two. Thus, there appears to be no secular trend which would explain the observed beta decline for the stock repurchasing firms. The observed decrease in equity beta for the repurchasing firms could be a result of a financial leverage change or of a more fundamental change in the underlying assets of the firm. The leverage change hypothesis seems unlikely since a share repurchase is mechanically a financial leverage increasing event. However, our calculations indicate that the sample's mean and median leverage increase is very slight, about two percent, over the period of the repurchase offer. This combination of equity beta decline with slight leverage increase suggests that the tender offer announcement is signalling a decline in the market discount rate on the assets generating the firm's earnings. We calculate firm asset betas before and after the stock repurchase using Hamada's [1972) model which assumes that the firm's debt is riskless. This model implies β E = β (V/E), where β E represents the equity beta, β is firm beta, and V and E represent the market values of the firm and its equity. We use the sum of the book value of debt and equity capitalization (from prior to the repurchase announcement) to estimate the market value of the firm. As might be expected the mean and median asset beta changes are very close to those reported in Table 6 for the common stock betas around the time of the repurchase period: the estimated mean and median firm asset beta changes are and respectively (p-values < 0.01). Thus, the evidence is consistent with the 20

21 hypothesis that issuer tender offers contain information about the riskiness of firm assets which implies a reduction in the market discount rate on expected earnings. We further pursue this reduced earnings volatility hypothesis in a manner similar to the analysis by Healy and Palepu [1990]. They investigate earnings volatility for firms announcing primary equity offerings by examining cross-sectional variances and interquartile ranges of the distributions of standardized annual earnings changes for years surrounding the offerings. Healy and Palepu [1990, fn. 26] recognize the limitations of analyzing cross-sectional data, stating that their procedure, "assumes that the distribution of standardized earnings changes is the same for all firms, and does not discriminate between changes in firm-specific and market-wide risk changes." In spite of these limitations, they interpret their evidence as indicating that volatility measures increase subsequent to equity offerings for their sample firms, but not for their comparison firms, which is consistent with their hypothesis of a market risk increase. We follow the Healy and Palepu [1990] approach in analyzing our data. Table 7 contains estimated cross-sectional variances and interquartile ranges of the sample distributions of annual EPS forecast errors (annual earnings changes) standardized by the stock price two trading days prior to the repurchase tender offer announcement. One difference between our analysis and that by Healy and Palepu is that our calculations use symmetrically 'trimmed' data, where observations in the tails (.025 trim) are deleted from each sample distribution to eliminate the large sample variances produced by outliers. 15 Examination of Table 7 indicates that, contrary to evidence on equity betas, the earnings data indicates greater cross-sectional variability in years following the repurchase tender offer. 15 Before trimming the estimated variances are largely uninterpretable; for example, for the comparison firms the smallest estimated variances, and are in years -5 and +5; whereas, the largest estimated variances, and are in year -2 and 0. See footnote 13 above for references on trimming and other robust techniques. 21

22 For example, using the estimated variance in year -1 as a benchmark (as in Healy and Palepu [1990]), the variances in years 2, 3, 4, and 5 are significantly larger as indicated by F tests. The same pattern is evidenced by the interquartile range measure, which is largely unaffected by trimming, and appears in the comparison sample as well. In the comparison sample, years 3, 4, and 5 have significantly larger variances relative to year -1 using an F test. Healy and Palepu indicate earnings volatility does not increase in their comparison sample when measured by the earnings change variance, but they apparently discount the contradictory interquartile range evidence in arriving at this conclusion. In fact, their comparison sample interquartile range pattern is similar to ours. Also, our observation period, 1969 to 1978, is spanned by Healy and Palepu's 1963 to 1981 observation period. Thus, one interpretation of these results is that over our observation period (and Healy and Palepu's) there has been a secular increase in cross-sectional earnings variability, which appears unrelated to a change in the earnings generating process associated with stock offerings or repurchases. An alternative interpretation is that, as Healy and Palepu acknowledge, cross-sectional variability measures are not reliable estimates of changes in earnings variability across time. 4.3 Relation of Tender Offer Announcement Stock Returns to Earnings Forecast Errors and Changes in Market Risk If issuer tender offers signal information to the market concerning firms' subsequent earnings prospects and the riskiness of its assets, then the size of the signal, proxied by the stock return prediction error for the tender offer announcement period, should be positively related to the size of the firm's contemporaneous and future earnings forecast errors and to the change in equity market risk around the repurchase. To test these propositions, we estimate regressions under two alternative specifications to allow earnings adjustments to 22

23 occur across a single year as well as more gradually. In the first formulation we use the cumulative forecast errors across the contemporaneous and four following years as a single explanatory variable and in the second specification we use each year's earnings forecast error as a separate explanatory variable. Since the martingale model is used to estimate the annual earnings forecast errors, cumulating the earnings forecast errors subsequent to the repurchase is equivalent to using the earnings for year 4 minus the earnings for year -1 as the forecast error. The earnings forecast errors are scaled by equity market value as of two days prior to the tender offer announcement. 16 For the annual EBIT based tests, the following regression models are estimated: 4 (4) spe j = α + β Σ (fe jt /P j *n j ) + θ beta j + υ j t=0 4 (5) spe j = α + Σ β t (fe jt /P j *n j ) + θ beta j + υ j t=0 Here, spe j is the stock's tender offer announcement prediction error, P j is the stock price two trading days prior to that announcement, and n j is the number of shares outstanding as of the announcement. 17 These equations assume the earnings forecast errors, fe jt, are EBIT based. The regression models employing annual EPS data have the same specification as equations (4) and (5) except that the scaling factor is the prior (day -2) stock price. The final explanatory variable 16 Vermaelen [1981] indicates (see footnote 15) that he finds no significant cross-sectional relation between tender offer announcement date abnormal stock returns and post-announcement earnings forecast errors standardized by an earnings variability measure. Forecast errors standardized by variability might be interpreted as a measure of the strength of an information signal relative to an estimate of its noise. In contrast, regressing abnormal stock returns on forecast errors scaled by equity capitalization in essence posits a relation between the change in capitalized value and abnormal earnings (cash flow) per period (Christie [1987] argues in favor of this scaling method). 17 We also considered an alternative procedure that controls for a market-wide earnings effect by subtracting out the market earnings forecast error from the firm's earnings forecast error. We find that our results are essentially invariant to this adjustment for market earnings. 23

24 in the equations, beta j, is the difference between the equity beta estimated in the year after the stock repurchase and in the year prior. We report results estimated from an ordinary least squares formulation. To reduce any possible heteroscedasticity induced by differing return variances across our sample of stocks, we have also employed a weighted least squares regression formulation, where the weighting factor is the inverse of the residual standard deviation from the market model estimation. Since the qualitative results of the two statistical models are indistinguishable, only ordinary least square estimates are reported. For each model we specify four measures of the common stock's valuation effect of the tender offer announcement. One measure is the market model stock prediction error described above (spe j ). We have also estimated regressions using two-day announcement period raw returns, what we call estimated prediction errors (EPE j ), and a variable we call information (Info j ) derived by Rosenfeld [1981]. These variables are all correlated, with estimated Pearson correlation coefficients ranging from 0.70 (for Info and the two-day raw return) to 0.98 (for the two-day raw return and the stock prediction error). Because of the high correlation for the twoday raw returns and prediction errors, we omit reporting the similar results found for the raw returns. The estimated prediction error (EPE j ) is a value we estimate from a cross-sectional regression of the offer announcement period stock return prediction errors on selected signalling variables that are initially observable by investors at the time of the tender offer announcement. If characteristics of the offer or the firm serve as signalling variables, then they should enable investors to discern cross-sectional differences in the extent to which earnings expectations and risk assessments will be revised. Under the information signalling hypothesis (and assuming that 24

25 we have selected informative proxies of these signals), we expect to observe a positive relation, in a regression, between EPE and subsequent earnings forecast errors and a negative relation with the change in equity market risk. If the hypothesis is valid, then this regression model demonstrates a direct linkage between the observable signals and our performance measures. The signalling variables that we have analyzed include financial leverage defined as the ratio of long term plus short term debt to total value (Total value is equal to debt plus preferred plus common stock, where debt is measured by its book value, preferred stock by its liquidation value and common stock by its market value); the tender offer premium (PREM), defined as the ratio of the tender price to the stock price two days before the tender offer minus 1.0; the fraction of outstanding shares sought in the tender offer (F s ); and the percentage of outstanding shares held by management. Since management's holdings and leverage are not important variables in our regressions, we use the following regression model to determine the estimated prediction errors, EPE: Model: spe j = α + β 1 PREM j + β 2 F sj + υ j where α β 1 β 2 R 2 F estimates: t-statistics: (-2.37) (17.80) (3.26) and EPE j = αˆ + βˆ 1 PREM j + βˆ 2 F sj The information (Info j ) variable is used because the two-day announcement return may be less than fully informative. For example, if the supply curve for shares for a particular company were sharply upward sloping (because of a capital gains tax lock-in effect, for 25