The impact of takeover offer timing on the measurement of Australian bidder gains: 1976 to 1995.Abstract: This paper demonstrates that the measurement of investor reactions to bidder offers is intimately affected by the endogenous nature of bidding. Bidding activity is not random. Offers are made at times that suit bidders. We show that bidder characteristics before an offer differ markedly from those of other firms and that these differences significantly impact on naive inferences of shareholder reactions to an offer. When appropriate adjustments are made to properly control for pre-offer characteristics, bidding shareholders are found to enjoy significant wealth increases. Keywords: TAKEOVERS; ACQUIRING FIRMS; EVENT STUDY, SELF-SELECTION; BIAS. 1. Introduction Two of the most widely accepted procedures in event-study research are the market-adjusted model and its more general form, the market model. The two models are often treated as equivalent forms by researchers, with little justification given for the particular choice taken. The market-adjusted, or (0,1) market model, assesses 'normal' performance relative to standards for a 'typical' market firm. Simplicity and use of a market-wide average are its most appealing features. An obvious limitation is the use of a single benchmark to represent all firms. Regardless of differences in risk or other attributes, the (0,1) market model adopts the same 'normal' return for every firm. By accommodating characteristics of individual firms, the more flexible market model appears better suited to distinguish between changes due to risk and those due capitalised expectations. In a recent paper, Simmonds (2003) finds the measurement of offer reactions for Australian bidding shareholders is highly sensitive to the method by which normal returns are gauged. Measures of expected bidder benefits differ significantly between these two, simple, statistically justifiable and widely accepted models of normal returns. Differences emerge despite controlling for size- and survival-related biases, return measurement methods (buy-and-hold versus cumulated), possible mis-specifications in traditional testing procedures, asymmetry in returns, sensitivity to extreme sample segments, incompatibilities between weighting schemes for sample estimates and benchmarks and the possibility of sample firms contaminating benchmarks. A more significant determinant of estimated bidder gains than any of these factors is the choice between the (0,1) market model and the market model. This paper continues the investigation of Australian bidder gains by exploring further characteristics of the (0,1) market and market models and explaining their differences. It argues that the choice between benchmarking procedures requires careful consideration of the conditions leading up to an event. In the case of Australian bidders, we argue that conditions leading up to an offer are related to the likelihood of being a bidder. This is not a unique condition. Superior returns over intervals leading up to an offer are a consistent feature for bidders in many takeover studies1 and leads Franks and Harris (1989) to conjecture that '[b]idders may time takeovers to coincide with favourable performance by their own stock. The timing of acquisitions in relation to stock-price performance of the bidding (or target) company should be, we believe, a subject for further research.' (2) We demonstrate that the timing of offers by bidders also bears on the appropriate choice of return model. Central to an understanding of which normal-return model more accurately measures abnormal return activity is the pre-offer performance of bidders and how this relates to the timing of offers. These are core features because they affect the two models differently. The (0,1) market model does not recognise prior performance. Only contemporaneous benchmark returns matter to it. In contrast, prior relationships do matter for the market model when earlier relationships fall within its estimation period. A critical issue, therefore, in the choice between these models, is the determination of whether the prior achievements of bidders accurately describe relationships through an event window. Three possible explanations for superior pre-offer performance are considered. The first explanation is that superior pre-offer returns are a consequence of investors in an efficient market bringing forward valuation effects expected to flow from upcoming offers. The capitalisation of these anticipated acquisition benefits may distort the measurement of later reactions during the offer event period. When valuable acquisitions are expected, normal return estimates for the market model could therefore be inflated and ratchet up the offer window response required before significant change is recognised. Furthermore, prior capitalisation of benefits also reduces later reactions in the offer window. Offer reactions are smaller and are assessed against higher standards of normality than is appropriate. Rather than anticipated acquisition benefits, a different possibility is that pre-offer price gains are a regular feature of bidder returns that are unrelated to the pending bid. Superior pre-offer returns then represent an opportunity cost to bidding shareholders, which the (0,1) market model fails to recognise. In this case, the (0,1) market model applies a normal-return measure representing average performance to a sector of the market with above-average standards. If bidders maintain the same superior standards through an offer window, the (0,1) market model can detect a positive investor reaction simply because it applies an inadequate measure of normal returns. A final possibility is that observed pre-offer gains are associated with transitory adjustments capitalising significant news events. Whether related to an upcoming offer or not, one-off gains do not reflect ongoing capital costs and can distort market-model estimates of offer window conditions. Under circumstances where pre-offer conditions do not represent ongoing return conditions, it is the market model which is inappropriate. This last possibility is considered the most compelling because bidders decide when to announce offers. Freedom to choose when an offer is announced means bidders can be expected to do so at times that best suit their purposes. (3) According to Kendig (1998), Australian takeovers4 occur in waves that are related to past economic circumstances. Kendig argues takeover waves are a consequence of greater management hubris and less binding corporate governance during times of prosperity. Increased takeover activity motivated by agency considerations is a consequence of these features. Other authors5 contend that heightened takeover activity is the response of bidders seeking to exploit improved future opportunities. According to this argument, increased share prices reflect improving expectations for future production opportunities, and takeover activity expands after increased share prices because firms attempt to quickly expand productive resources through acquisitions. We also argue that takeover activity correlates with general market conditions. Takeover activity is not randomly scattered through time. Importantly, we show offers often follow periods of considerable cash-flow improvements. Bidder samples contain exceptionally high proportions of firms with superior pre-offer return performance because, we contend, bidders choose when to acquire, and tend to do so after both the market and their own firms experience significant improvements in internally generated cash flows. When the market model is applied, normal returns estimated from intervals preceding both the offer and cash-flow improvements are exaggerated. This leads to a greater representation of negatively biased abnormal returns6 in the bidder sample than can be expected by chance, and reduces the prospect of measuring positive offer-window reactions against standard null distributions. Conditions associated with takeover activity cause naive market-model methods to reject wealth creation too often. A more sophisticated bootstrapping method is introduced to overcome biases associated with bidder endogeneity. In addition to the controls for size and survival regularities in shareholder returns, the expanded-bootstrap method controls for pre-offer risk-adjusted return (RAR) and systematic risk levels. The extra controls ensure market model-bidder CARs are tested against a null distribution that more accurately reflects the exceptional features of bidder returns. Rather than applying complex and costly bootstrapping methods, an alternative remedy is to use a model that is less susceptible to distortions linked with pre-offer return characteristics and sample definition. Although relatively crude, the (0,1) model is argued to be better specified than the market model for measuring bidder abnormal returns. Its inflexibility actually protects against biases created by transitory pre-offer characteristics not maintained during an event. In a sample that defines itself by its own actions, transitory adjustments that would otherwise be quickly capitalised by an efficient market need not be random. In the case of bidders, substantial share price gains are disproportionately more common than in the wider market, before many offers. (7) Over-representation of high pre-offer returns inflates estimates of normal conditions and introduces a negative bias to the sample's market-model abnormal-return estimates. The (0,1) market model is immune from such problems because it does not consider past relationships for individual firms. Only current conditions matter to the (0,1) market model. In the case of bidders, current conditions for the typical firm can provide a better measure of normalcy than prior conditions. The (0,1) market model may be a blunt tool but its robustness to sampling biases can dominate power deficiencies. (8) The next section illustrates the existence of a close relationship between the level of Australian takeover activity and the recent performance of large firms relative to smaller stocks. The better the past performance of large firms over smaller stocks, the greater the average number of offers emerging. Conversely, the larger are recent gains for small stocks, the fewer offers are made. The third section briefly introduces the market and (0,1) market models. Summary measures of pre-offer bidder characteristics are also given. Bidders are shown to possess significantly higher pre-offer RARs and pre-offer systematic risk. The fourth section discusses a general model of investor anticipation and explains the consequences of revised expectations on model parameters. The fifth section moves beyond share prices and incorporates evidence on cash flows. Sudden near-term changes in cash flow growth rates are found to be directly linked to estimated market model risk adjusted returns. This is consistent with predictions from the model of investor anticipation. The sixth section extends the randomised testing procedure of Simmonds (2003) to control for sampling biases created by bidding endogeneity. When the new procedure is applied, both models lead to the conclusion that Australian bidder wealth significantly increases at the announcement of takeover offers. The final section concludes. 2. The Timing of Australian Offers Figure 1 presents takeover offer trends for Australian listed companies between January 1972 and December 2000. Bidding trends are given for two segments of the acquiring population. All offers by any bidder are shown, whether Australian listed or not, as are trends for just Australian listed bidders. (9) In both cases, trends are measured by thirteen month moving averages centred on the current month and deflated by the total number of Australian listed companies, to control for listing variations through time. Bidding trends are measured on the left hand axis. [FIGURE 1 OMITTED] Also presented each month are average market returns over two year intervals ending last month. For instance, 'Avg EW Market [-24,-1]' corresponds to the average equal-weighted market return over months [-24] to [-1] relative to the current month, [0]. 'Avg VW Market [-24,-1 ]' is a similar metric based on value-weighted market returns. Average market returns are measured on the right hand axis. A final indicator of market returns is 'Avg VW Mkt - EW Mkt [-24,-1]'. It simply records the difference between the value- and equal-weighted averages over [-24,-1] and highlights times of greatest divergence between the VW and EW measures. A strong relationship between bidding trends and prior market performance is evident in figure 1, particularly through the 1970s and 1980s when takeover activity rises and falls closely with past market returns. Takeover activity is more pronounced after periods of sustained market growth and shrinks following periods of market decline. Only during the 1990s is the positive association between prior market performance and subsequent takeover activity obscure. Evidence favouring continuation of the relationship through the 1990s emerges after a more detailed examination of the different market measures. The key determinant of bidding activity (11) is not the past performance of all firms but that of large firms relative to small firms. Equal-weighted (EW) and value-weighted (VW) market returns are useful measures to highlight performance differences between small and large firm market segments. The EW measure is sensitive to the performance of the market's many small firms whereas the VW measure is more heavily influenced by the relatively small number of large stocks. When EW returns are larger than VW returns, the performance of a typical small company exceeds that of the typical large firm, and vice versa. Most of the time, figure 1 shows the two measures track relatively closely to each other, and, consistent with the small firm effect, average EW returns over [-24,-1] are almost always larger than VW returns. The 1990's are exceptional. For brief intervals, the relative values for EW and VW averages are reversed, with VW returns being larger than EW returns. The most striking deviation, however, occurs in the first half of the decade when EW market returns surge past their VW counterparts. EW gains at this time far exceed any in earlier decades (see 'VW Mkt - EW Mkt [-24,-1]') and indicate that, compared to small firms, the returns of many large stocks were depressed through the 1990s. It is this relatively poor performance of large firms which possibly explains the slowing in takeover activity through the 1990s. This is because acquiring targets is primarily an activity of large firms on small firms. When prices for large firms are slower to rise, or faster to fall, than those of small firms, the relative cost of equity capital for large firms increases while the price of small firm equity grows. Reduced bidding activity is consistent with such conditions. Figure 1 shows takeover waves are strongly associated with relative economic conditions for large firms, not just the overall market. The trends presented in figure 1 demonstrate that the level of takeover activity is principally related to the recent relative prosperity or success of large listed firms. Two aspects are important for large firms: their own past performance and how that performance compares to target firms. The more successful large domestic firms had been relative to smaller companies, the greater the number of takeover offers to emerge for Australian listed stocks. Furthermore, the greater the recent gains of both large and small firms, the higher the level of takeover activity. Figure 1 illustrates a relationship between the incidence of takeover offers and the past performance of different market segments. The timing of offers is not haphazard. It follows identifiable periods of relative share price gains for large firms. The results are consistent with bidders becoming more prolific after their own past performance had improved. The next section shows significant bidder gains do indeed precede many takeover offers and are recorded within the bidder sample as exceptional pre-offer risk-adjusted returns. 3. Two Models of Normal Returns This paper extends the analysis of Simmonds (2003) and applies the same procedures described there. Salient features are summarised below. Figure 2 shows the time line defining the pre-offer, post-offer and event window periods applied here]2 (Subsequent date references in this paper are made relative to an offer announcement. Braces are used to indicate monthly intervals. For example, [0] is the offer month, [+1] is the month after.) Shareholder return characteristics are measured by both the market model and the (0,1) market model. [FIGURE 2 OMITTED] The market model is estimated according to a specification proposed by Karafiath and Spencer (1991) as, (1) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.] where: [R.sub.it] = the continuous return experienced by shareholders of company i in month t. [a.sub.i] = a constant intercept term measuring the average risk adjusted return throughout the entire observation period [-34,+31]; [[beta].sub.1] = the sensitivity of company returns over [-34,+31] to changes in benchmark (market) returns; [R.sub.Mt] = the continuous return observed for a benchmark index such as the entire market. Benchmark indices are formed as equal-weighted averages of continuous returns for individual companies. Following earlier arguments and results, value-weights are not pursued. However, benchmarks do include controls for size- and survivor-related biases. [[delta].sub.ie] = the abnormal return estimate for month e within [-10,+7]. [D.sub.et] = 1 when e=t and month t [member] [-10,+7] and is zero elsewhere; [[alpha].sub.i,Post-Offer-Chg] = a constant intercept term measuring the change in average risk adjusted return over [+8,+31]; [D.sub,i,Post-Offer] = 1 when month t [member] [+8,+31] and is zero elsewhere; [[beta].sub.i,Post-Offe-Chg = the change in [[beta].sub.i] over [+8,+31]; and [[sigma].sup.2.sub.i] = the variance of [[epsilon.sub.it].. With this specification, standard regression procedures can properly accommodate the effect on standard errors of dependence between abnormal return and parameter estimates; both [[alpha].sub.i] and [[beta].sub.i] can vary after an offer has been announced; and the estimate for abnormal return variance, [[sigma].sup.2.sub.i], reflects any post-offer parameter changes. The (0,1) market model in equation (2) is estimated over similar intervals, before, during and after an offer. (2) [R.sub.it] = [R.sub.Nt] + [[epsilon].sub.it] (0,1) market model abnormal returns are estimated as [[epsilon].sub.it] = [R.sub.it] - [R.sub.Mt]. Traditional t-ratios are provided to test the significance of outcomes but, given the results of Simmonds (2003), bootstrapping procedures are favoured. These are based on 100 randomly selected size- and survivor-matched look-alike samples which entirely mimic genuine bidder samples. (13) Following Simmonds (2003), the initial collection of look-alike samples are randomly selected with replacement from all firms matching individual bidder companies in terms of the same size and survival qualities near an offer announcement. Bootstrap results are formed across 100 such samples, each of which is the same size as the single bidder sample. Controls for size- and survivor-related biases are also applied directly to each model by varying the 'market' benchmark. Direct controls are applied to isolate the effects of empirical regularities in company returns noted between these characteristics (14) in the finance literature. Size-based controls are tailored to individual companies based on that firm's size-decile ranking across all companies two months before the current month. Except for companies in the largest decile, the size-adjusted return for a current month is that month's return for the size-decile to which a company belonged two months ago. Within the largest decile, benchmark returns are generated from finer size groupings. The largest decile is divided into companies between the 90th and 95th percentile, those between the 95th and 99th percentile and the top 1% by size. This finer classification is applied to ensure sizes are more homogeneous across each stratum, to reduce the impact of small numbers of large companies within any one stratum and to minimise differences between value- and equal-weights when applied to overall average performance. Survivor-based controls are found by applying the same eligibility requirements to all listed companies that are used to accept bidder sample member firms. A surviving company passes two tests: 1) a return is available for the current month; 2) at least twenty returns are available across two twenty four month periods, [-34,-11] and [+8,+31] relative to the current month, with at least ten of these coming from the [-34,-11] period. (15) These intervals correspond to typical pre-offer and post-close periods. Size- and survivor-based controls are also jointly applied. Akin to the individually tailored size-based controls, joint size- and survivor-based controls are formed from a similar range of sub-decile indices comprising just survivor companies. 3.1 Data Sources The company returns data set employed here is predominantly sourced from the Centre for Research in Finance (CRIF), but also includes observations from a Statex data set released by the ASX. The takeover offer data set includes that developed by Dr Steven Bishop for the Centre for Independent Studies (CIS), (which was kindly released for this research), plus additional details for all offers reported in various ASX publications between 1980 and 1995. Every effort was made to record details for all takeover offers between January 1972 and June 1995. (16) Determination of whether or not takeover participants were listed was made by referring to a separate CRIF data set of company names. 3.2 Table Layout for Bootstrap Sample Results All tables presenting bootstrap results adopt the same format. Each contains several panels and three blocks of columns. Each panel is divided into four categories representing different benchmarking policies, such as a single equal-weighted market portfolio, indicated by the heading EW Market, or survivor- and size-adjusted benchmarks, labelled Survivor- & Size-Adjusted EW Market. Within each benchmark category panel two sets of estimates are shown. The first line gives results when no robustness restrictions are imposed. The second line shows estimates when all robustness restrictions are used. (17) These second estimates should be least sensitive to influential sample minorities. Blocks of columns across the tables correspond to particular samples, or their relative positions. The first block of columns, labelled Bidder Samples, presents results for the sample of eligible bidders. Columns under the heading of Medians and Means of 100 Random Samples summarise outcomes for 100 separate samples of 1118 randomly selected look-alike firms. Look-alike firms are initially matched with genuine bidders in size and survival qualities at the time an offer was announced. The same range of variables measuring the performance of the genuine bidder sample are produced for each of the 100 random look-alike samples. Median and mean outcomes across all 100 samples are shown in the tables. The final block of columns is headed Prop'n Random Samples [less than or equal to] Bidder Sample and shows the proportion of bootstrap samples generating smaller outcomes than the genuine bidder sample. These columns provide empirical estimates for the probability that genuine bidder results differ from those for other firms. They answer the question, 'What is the probability that genuine bidders generate greater performance than companies with comparable size- and survival-qualities?' 4. Characteristics of Bidder Returns Before an Offer Table 1 presents estimates of bidder return characteristics in the pre-offer period. The first panel gives average risk adjusted excess returns (RARs). Measured in continuous monthly rates, RARs represent the average excess return experienced by shareholders of an individual bidder after adjusting for changes correlated with benchmark portfolio measures. The RAR of an individual bidder corresponds to [[alpha].sub.i] in the market model of (1). Table 1 reports typical RAR values across all sample companies. Before commenting on the RAR results, it should be noted that performance measures, like RARs, are inextricably linked with the underlying requirements of the performance model. The correct interpretation of an RAR depends on the model's foundations. For instance, if market returns could be observed on the efficient frontier and conditions appropriate for Sharpe-Lintner's Capital Asset Pricing Model (CAPM) applied, individual RARs would all be zero. Finding any RARs differing from zero would signal violations of CAPM's underlying equilibrium requirements. Although the market model closely resembles CAPM, it does not rely on CAPM for justification and can be interpreted as a simple linear relationship between returns for individual firms and a benchmark. When that benchmark is defined to represent all 'comparable' firms, the market model illustrates relative performance within the comparable set. When applied in this manner, RARs measure the relative performance of individual companies within the comparable set, after adjusting for correlated or common effect price movements, [[beta].sub.i] then measures the sensitivity of individual stocks to common benchmark effects, rather than sensitivity to common market-wide factors, as is appropriate for CAPM. The remainder of table 1 reports typical [[beta].sub.i] characteristics, in the second panel and typical cumulative abnormal returns, as measured by the (0,1) market model, in the third panel. The first panel shows that, on average, bidder returns over [-34,-11] significantly exceed those of benchmark firms. Typical (median) bidder pre-offer RARs range between 0.6% and 0.9% per month, depending upon the benchmark. Every bidder t-statistic is highly significant. That significance is confirmed by results for the bootstrap samples which report no sample of look-alike firms ever generates a higher median pre-offer RAR than that of the genuine bidder sample. Mean bootstrap RARs are also always lower than those of genuine bidders when less stable estimates are removed. According to these results, bidders enjoy significantly higher returns before an offer than are experienced by a 'typical' benchmark firm. The gains emerge after controlling for differences related to size, survival, size and survival qualities jointly, skewness and outliers. Table 1 also reveals that offers tend to follow intervals when prices across the market grew. Bidder pre-offer RARs exceed those of look-alike firms, but look-alike firm RARs are also consistently positive through [-34,-11], ranging between 0.5% and 0.6%, as benchmarks change, with highly significant t-statistics. If look-alike RAR estimates are unbiased and testing procedures properly specified, these positive results imply firms matching bidders in terms of size and survival qualities, at the time an offer is announced, outperform standards achieved by a typical benchmark firm. In other words, firms with similar size and survival characteristics to bidders at offer announcements, grew more quickly through [34,-11] than an average firm. Bidders come from market segments where recent share price gains have been significant and those of bidders most significant. Such results are consistent with the patterns displayed in figure 1 between bidding trends and previous returns for the entire market. Although the previous interpretation is consistent with figure 1, a different explanation for widespread gains in bidder market segments is the possibility that RAR estimates are biased and testing procedure ill-specified, (18) even after controlling or size and survival qualities. Bidder 'gains' may evaporate if biases exist and can be removed. One advantage of bootstrap sampling is that it permits underlying biases (19) to be estimated and controlled. By drawing random selections from an available population, bootstrap methods reveal typical population characteristics for the procedure employed. When underlying biases are present, unbiased bidder values can be obtained by subtracting the typical bootstrap result. In table 1, bias-adjusted or net bidder gains range from 0.1% (=0.6%-0.5%) to 0.4% (=0.9% - 0.5%) for median RARs. Thus, even after adjusting for possible underlying biases, the conclusion of exceptional pre-offer bidder price gains is unchanged. The model controlling the most variables, the Survivor & Size Adjusted EW Market benchmark with all robustness restrictions applied, produces net gains of 0.1% and 0.4% (median and mean RARs). No bootstrap sample exceeds gains recorded for bidders by this measure. The second panel reports the sensitivity of bidder returns to common factors among benchmark firms. It shows bidder firms are significantly more sensitive to common influences than bootstrap look-alikes. Bidder beta estimates exceed those of bootstrap samples by approximately 0.08, with differences ranging between 0.06 and 0.12. Bidder beta values are higher than those in almost every bootstrap sample. Bidders enjoy higher average excess returns before an offer and they also gain more than comparable firms when the market increases. As figure 1 shows, market increases are common in the months leading up to heightened bidding activity. The second panel also indicates underlying biases are not severe for the market model. When market model parameters are estimated without bias, when the benchmark population is stable and when characteristics defining look-alikes are unrelated to RARs and betas, the mean outcome across all benchmark firms is unity for beta and zero for RAR. Consistent with insignificant bias, mean betas for bootstrap samples are all close to one. For example, the mean beta estimate across all bootstrap samples is 1.00 for the Survivor & Size Adjusted EW Market benchmark when all estimates are considered. (The overall estimate falls slightly to 0.98 when more volatile betas are removed.). Potential look-alike firms are defined here by their size and survival characteristics. Those same qualities define the Survivor & Size Adjusted EW Market benchmark. Selected look-alike firms are therefore just random selections at different times from firms used to construct the Survivor & Size Adjusted EW Market benchmark. Hence a mean beta of one for this benchmark confirms bootstrap selections are random and that beta estimation is unbiased. (20) An apparent difficulty with the previous argument is that exactly the same reasoning can be applied to pre-offer RAR estimates. Under the same conditions, the mean RAR of randomly selected bootstrap firms should be zero for the Survivor & Size Adjusted EW Market benchmark. However the mean bootstrap RAR is not zero. It is 0.5%; with both t-tests indicating the difference is significant. Influential extremes do not explain the difference because the same finding obtains when all robustness restrictions are applied. The explanation for positive mean bootstrap RARs is revealed by their proportional spread, % < [C.sup.[dagger]]. Look-alike pre-offer RARs are not evenly split between positive and negative. Commonly, only 40% are negative. It is the strong representation of positive pre-offer RAR firms that causes the positive overall result. Too many firms experience positive pre-offer RARs for the zero outcome expected when estimation is unbiased and bootstrap selection is unrelated to RAR values. Nevertheless, the previous results for look-alike betas indicate modelling procedures are not the source of any bias. Overall risk measures do conform to expectations for random samples. It will be shown below that the fundamental problem is an uncontrolled implicit relationship between pre-offer RARs and the selection of these bootstrap samples. As was noted above, takeover activity increases after periods of overall market gain. Table l shows these times were also when bidders achieved outstanding performance in their own share prices. Furthermore, they were also times when firms matching bidders in terms of size- and survival-qualities at the time of an offer, look-alike firms, often experienced recent share price performances exceeding benchmark standards. This is not a coincidence. Nor is it a consequence of estimation bias. (21) Instead, it shows market segments resembling bidders at the time of takeover offers are the principal beneficiaries of market gains. Remember, look-alike firms are selected to match bidders at particular points in time, near takeover offers. A large look-alike firm two years before an offer is not matched with a large bidder now. Eligible look-alikes are all large near the offer. It follows therefore that firms similar to bidders at the time of an offer announcement, outperformed other firms in the months leading up to an offer. The fact that bidder pre-offer RARs are greater than those of any look-alike sample confirms bidders as the largest beneficiaries in their own market segments. The final panel of table 1 reports abnormal returns estimated by the (0,1) market model and cumulated through [-34,-11] (corresponding to [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.]. for an individual company). It shows bidder returns before an offer typically exceed benchmark levels by between 0.7% and 1.0% per month. Similar measures for look-alike samples also exceed benchmark values but by only 0.3% to 0.6% per month. As with RAR measures, it is typical for 59% of look-alike CARs to be positive and for even higher proportions within the bidder sample. No bootstrap sample ever produces greater proportions of positive CARs, or higher levels of CARs than the genuine bidder sample. The market model and (0,1) market models both indicate outstanding pre-offer achievements for bidders. Since the market model assesses subsequent abnormal performance relative to observed pre-offer standards but the (0,1) market model focuses only on contemporaneous benchmark values, a fundamental issue to resolve is whether or not that earlier performance properly describes normal behaviour. The next section takes up this issue. It expands the justification of Malatesta and Thompson (1985) for the market model to include unexpected announcements and then considers the impact of such announcements on measures of normal returns. 5. Anticipation and Market Model Parameters Malatesta and Thompson (1985) describe market model parameters in an efficient market context as adjustments for partial anticipation of future acquisition events. In their model, pre-offer RARs are proportional to the expected adjustment required when an anticipated event, such as a takeover offer announcement, does not occur. An adjustment is required because efficient market investors anticipate an expected benefit before it happens. Before a period starts, the anticipated benefit is equal to the expected economic benefit B discounted by the probability of the event occurring, q. At the end of a period, if that event has not occurred, pre-event expectations must be unwound and prices adjusted to reflect the news of no news. In Malatesta and Thompson's framework, RARs reflect this constant unwinding and are proportional to -qB, the opposite of pre-event anticipation. RARs are also important when an anticipated event does occur because share price change is then proportional to just the unanticipated component, (1-q).B. The full economic consequences of the event, B, are only revealed after adding back pre-event anticipation, that is, after adding qB or subtracting -qB, the RAR. Malatesta and Thompson's model provides a theoretical foundation for the market model. It shows that pre-offer RAR values measure the regular adjustment required to unwind a previously anticipated, non-occurring event. When events are infrequent and non-occurrence is the norm, pre-offer RARs describe a regular or normal component of returns and provide the adjustment needed to distinguish between investor perceptions of economic benefits following an event and their updated anticipation. The market model correctly makes such an adjustment when determining abnormal returns. In contrast, the (0,1) market model makes no such adjustment and so potentially under-estimates the economic impact of valuable takeover offers. In Malatesta and Thompson's model the most significant capitalisation of acquisition benefits occurs at some time preceding model estimation. It is at this time that expected future acquisition benefits, (qB/R - F), are first capitalised. (22) This is when the benefits expected each period are first impounded into share prices and hence anticipated. The initial capitalisation does not feature in their model except to be constantly undone or updated, as events unfold. Incorporating the initial price capitalisation, or updating earlier predictions of economic benefits after later news, extends the model to include the effects of unanticipated announcements. Unanticipated announcements can occur at any time and may impact severely on share prices. For example, news that current cash flows are to be substantially higher than previously predicted may lead to upward revisions of many later period estimates. Whether due to modified predictions or the disclosure of a completely new investment opportunity, unanticipated announcements in an efficient market can be expected to cause irregular capitalised price adjustments, like ([q.sub.u][B.sub.u]/R - [F.sub.u]). Such price adjustments may be large relative to those otherwise considered as 'normal'. When unanticipated announcements are also recognised, RARs reflect both the regular unwinding required for a non-occurring anticipated event and the capitalisation of any unanticipated news. A more complete description of anticipation effects on pre-offer RARs includes adjustments for unanticipated announcements. It also accounts for the consequences of news on other anticipated events than acquisitions. Thus, each period, RARs actually comprise (1) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.] where: [V.sub.t-1] is the market value of the firm in period t-1 and [D.sub.X,t] are dummy variables that take the value 1 in period t when X occurs and are zero elsewhere. The first summation on the right represents the total impact of all unanticipated announcements during period t, [U.sub.t]; the second summation corresponds to anticipated events that did not occur and are unwound, [A.sup.N.sub.t]; and the final summation is the extra contribution from anticipated events that do occur, [A.sup.Y.sub.t]. [U.sub.t] and [A.sub.t] define the set of unanticipated and anticipated events before period t. A time subscript is required because unanticipated events last period become anticipated events in the next. The first summation involves the immediate consequences of today's unanticipated announcements plus the capitalised value of similar expected future events. RA[R.sub.t] summarises numerous announcement effects as a single measure. It captures more than just -qB. Of particular concern is the contribution of [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.] Although the likelihood of such events may be low, when they do occur their impact can be large and produce biases in RAR estimates as measures of -qB. 5.1 Factors Mitigating Abnormal Return Biases The issue of whether or not bias in estimated RARs is empirically significant hinges on the systematic properties of the bias and its size. Systematic patterns are important because systematic bias is not eliminated by sample diversity. Nonsystematic, or random patterns, are eliminated. For example, announcements concerning other investments that are randomly scattered over intervals leading up to a takeover offer, can be argued to have negligible effect on abnormal return estimates for an entire bidder sample because idiosyncratic effects average out across the sample. Takeover related estimates for B may be too high for some bidders and too low for others but overall, an unbiased measure for B is gained for the entire sample. The smaller the bidder sample, the less plausible this argument is. The diversification argument also fails if a particular class of announcement tends to precede takeover offers. Pre-offer RARs are then systematically affected and mis-represent the necessary adjustments required to properly measure B. In terms of bias size, the largest influence on RAR measures can be expected from responses to announcements outside the event window, particularly announcements to unexpected (23) investments or unexpected changes for previously predicted benefits. Although the impact of such changes is potentially large, the diversification argument still applies and can alleviate the effect of randomly scattered surprises. Once again, it is less reasonable when surprises are not random. Before a takeover offer, non-random surprises are perhaps most likely as updated predictions of acquisition benefits for the upcoming offer, or as revised forecasts for existing investment opportunities. For example, shareholders may learn that takeover offers tend to follow news of improved circumstances in a bidder's existing operations. News of increased cash flows could then lead to updated predictions of possible future acquisition benefits. Alternatively, news of surprising cash flow improvements may have marginal effect on future acquisition plans and produce little change in investors' acquisition beliefs. It may be the surprising cash flow news that sustains positive bidder price revisions before many takeover offers. If so, pre-offer RARs will tend to be positively biased across the bidder sample as measures of -qB. This would lead to a negative bias in the difference between an offer period reaction and the pre-offer RAR and underestimation of the genuine economic value of the acquisition. An accurate assessment of the economic value of acquisitions requires isolation and removal of other news effects on RAR estimates before abnormal returns are calculated. When other news effects are present, the market model can produce biased measures of anticipated acquisition benefits. On the other hand, by failing to adjust for pre-offer anticipated acquisition benefits, the (0,1) market model can also lead to biased assessments when anticipation effects are significant. 6. Anticipated Acquisition Effects Our modified version of Malatesta and Thompson's (1985) model indicates two possible sources of systematic influence on pre-offer RAR estimates for bidder sample members. The first of these stems from increased anticipation of the upcoming offer before its announcement. The second originates from consistent news of better than expected cash flows before an offer. (24) The next section investigates cash flow improvements as a source of bias. This section examines evidence on pre-offer anticipation bias. Pre-offer anticipation bias is avoided when expectations of an impending bid do not increase before the offer. An intuitive method traditionally adopted to minimise pre-offer anticipation bias and improve normal return estimation is to observe pre-offer returns well before an offer. For instance, our pre-offer estimation period (see fig. 2) ends four months before acquirers typically first disclose target share ownership reaching 5%. Not all bidders submit substantial shareholder notices before an offer. The first indication of target ownership for many bidders comes with the takeover offer itself. For others, a substantial holding is not achieved, even at the end of the offer. Nevertheless, the pre-offer estimation interval we use always ends eleven months before the offer announcement, three months before the typical substantial shareholder announcement date. Significant prior updates of acquisition benefits could be considered unlikely over two year intervals ending almost a year before the offer. 6.1 More Direct Measures of Anticipation A comparison of model characteristics for bidders making the most surprising offers with those that are least surprising provides a more direct assessment of updated acquisition benefit expectations. If future acquisition benefits are significantly anticipated (25) and capitalised before an offer is announced, two possibilities can be forecast according to the predictability of offers. Both possibilities are consistent with the efficient markets hypothesis. Assuming investors are better able to obtain or predict valuable news before impending offers by probable bidders, one possibility is that price revisions for updated expectations are greatest for more probable bidders. Bidders making the most anticipated offers will then possess higher (26) pre-offer RARs than more unexpected bidders because their acquisition gains are recognised sooner. Rather than occurring immediately before the current offer, a different possibility is that investors have already capitalised acquisition benefits for the most probable bidders. When investors look beyond a current offer and consider the likely acquisition benefits a bidder obtains from the first and all subsequent acquisitions, price revisions before later offers will be reduced by the earlier capitalisation. Pre-offer RARs for the current offer could then be relatively small for bidders making the most anticipated offers because their benefits were largely capitalised earlier in their takeover careers. This is the situation considered by Malatesta and Thompson (1985), where pre-offer RARs are unaffected by earlier capitalised benefits, except as those benefits are unwound after news of no offer. The first possibility predicts higher pre-offer RARs for bidders engaged in more predictable takeovers while the second forecasts lower pre-offer RARs because benefits are capitalised earlier in acquisition careers. The empirical importance of anticipated acquisition benefits, as an explanation of positive pre-offer RARs, is weakened if no significant differences are found between bidder types. Various forms of anticipation have been examined in the economics literature on takeovers (see for example Pound and Zeckhauser (1990), Aitken and Czernkowski (1992) on pre-offer media rumours, Schipper and Thompson (1983) on declared acquisition agenda, Castagna & Matolcsy (1985) and Palepu (1986) on predicting takeover targets). Perhaps the strongest signals of impending acquisitions come from actions taken by managers, like the declaration of an acquisition agenda. Actually announcing an offer is a clear signal. Obviously so, for the present target, but importantly, also for others, further into the future. The more a company is involved in acquisitions, the less surprising each subsequent offer could be expected to be. Indeed, companies frequently develop reputations for acquisition skills. Terms such as 'raiders' and 'greenmailers' have been applied to frequent bidders (e.g. Dechow (1987) or Casey, Dodd & Dolan 1987). The pre-offer RARs of 'raiders' could be expected to reflect their acquisition abilities. In order to investigate anticipated acquisition benefits more directly, we explore differences between pre-offer RARs as bidders become increasingly experienced at conducting takeovers. For this purpose, all Australian takeover offers were classified according to the number of offers previously announced by a bidder for any identifiable ASX listed target. (27) It is the order in which an offer appears, its place within an acquisition agenda that is important for our purposes, not the total number of offers eventually announced by a bidder at the end of the sample period. Accordingly, the bidder sample was divided into three similarly sized classifications (28) as: 1. First-time or novice bidders, representing first-time offers by a bidder for any target. 2. Apprentice bidders, who have announced offers for one or two other companies prior to this target. 3. Experienced bidders, who have sought more than two other targets before this offer. Offers by first-time bidders are considered least likely of the three categories to be anticipated. Those by Experienced bidder are predicted to be most strongly affected by anticipation effects. Table 2 presents measures of pre-offer returns for each category, after applying all robustness restrictions to estimates from the Survivor & Size Adjusted EW Market benchmarks. Measures are given for both the market and (0,1) market models. Also given are measures of significance for differences between parameter estimates. For instance, lines labeled 'First-time--Experienced' show the proportion of random bootstrap samples with smaller differences than those between genuine First-time and Experienced bidders. It is the significance of these differences that are important for testing the relevance of anticipated acquisition benefits as a source of pre-offer RAR bias. Some results in table 2 do appear to support a relationship between pre-offer RARs and anticipated future acquisition benefits. As predicted by the first hypothesis and the prospect of positive acquisition opportunities, pre-offer RARs increase for bidders displaying a greater aptitude or willingness to become involved in takeovers; RAR estimates trend upwards as prior bidder activity increases (from 0.5%/0.7% (median/means) for First-time bidders, through 0.6%/0.9% to 0.9%/1.1% for Experienced bidders). However, the estimated trend is not sufficiently strong to infer significant differences between activity classes because a similar trend also exists within the random bootstrap samples. Controlling for the trend in bootstrap samples produces a bias-adjusted 'trend' for genuine bidders of just 0.1%/0.4%, 0.2%/0.4% and 0.3%/0.6% (differences in medians/means). Bias-adjusted RARs still increase with bidder activity but not significantly so. Comparisons between bidder and bootstrap samples show First-time bidder pre-offer RARs are no further from those for Experienced bidders than the same difference between corresponding look-alikes. In other words, no significant differences are found between the activity classes. The median difference between First-time and Experienced bidders, for example, exceeds the same difference for just 49% of look-alike samples. Contrary to predictions for anticipated benefits, pre-offer RARs of Experienced bidders are not significantly different from those making First-time offers. Despite the appearance of a rising trend, the anticipated acquisition benefits hypothesis is not supported. Anticipated acquisition benefits do not explain the significant pre-offer features of RARs generally observed for bidders. Interestingly, a conclusion of no relationship is not uniformly supported across table 2. When measured by the (0,1) market model, significant differences do exist between pre-offer abnormal returns for Experienced bidders and other bidder categories. However, before concluding anticipated acquisition benefits are an empirically important component of (0,1) market model pre-offer CARs, (29) another aspect of takeover activity, the timing of offers, must also be controlled. In the last panel of table 2, Experienced bidder prices are shown to 'be more sensitive to factors affecting overall benchmark returns than other bidder types. Given this greater sensitivity to broader market conditions, the fact that offer timing is a matter for bidders to decide, and the earlier evidence that offer activity increases after overall market gains, the larger pre-offer (0,1) market model CARs of Experienced bidders are better explained by a talent for announcing offers after their own circumstance improve. It is this timing of offers and greater sensitivity to overall market conditions that better explains the superior pre-offer gains of Experienced bidders. Overall, therefore, anticipated future acquisition benefits are not found to be a significant factor explaining positive pre-offer abnormal returns. This is to be expected when estimation intervals sufficiently precede offer windows. The next section explores links between news on cash flow improvements and pre-offer RARs. 7. Cash Flows and Market Model Measures This section investigates cash flow evidence to distinguish between different explanations for pre-offer RAR bidder gains. It addresses the issue of whether pre-offer RARs only describe normal returns for bidders, as assumed by the market model, or are instead significantly affected by systematic factors, such as news concerning existing cash flows or anticipation of upcoming acquisition prospects. As before, the Australian equity market is assumed to be efficient, where expectations are unbiased and are updated as investors become more certain an event will happen, rather than when it actually occurs. When these conditions hold, realised cash flows and investor expectations will be related. Realised cash flows may deviate from expectations but, on average, realised cash flows and expectations will agree. Occasions when cash flows suddenly change are therefore likely to be times of significant revisions for expectations and, furthermore, will tend to trail those revisions. Occasions of actual cash flow change are important indicators of expectation revisions and can help to explain consequent RAR outcomes. If unexpected news on existing operations explains the positive pre-offer RARs of many bidders, that news will emerge during or slightly after the interval in which pre-offer RARs are estimated. News before the RAR estimation interval will be already capitalised in an efficient market. Exactly how long after an estimation interval cash flow changes should emerge is unclear because the influence of future events on investor valuations is an empirical unknown. Anticipated cash flow changes could materialize after an acquisition, before the offer, or not at all, if a longer investor horizon applied than was sampled. The delay between updated expectations and realised cash flows may be indeterminate, but evidence of significant change in different intervals can still be revealing. Significant cash flow change before an offer renders anticipation of upcoming acquisition benefits an implausible explanation of pre-offer RAR gains because pre-offer cash flows are not influenced by yet to be assimilated target operations. Instead, such evidence supports unexpected news on existing operations as the reason for pre-offer bidder price rises. Significant change after an offer is less informative, as it does not distinguish between existing and assimilated target operations. Evidence of no significant change is also inconclusive because it fails to dismiss the possibility of change further into the future being anticipated now. If pre-offer RAR levels are largely determined by news on existing operations, evidence of altered cash flow expectations may be found before the offer. In contrast, observed pre-offer cash flows and RARs will be unrelated if anticipated acquisition benefits drive pre-offer RAR gains. Significant correlations after the offer, when target firm operations can influence bidder firm cash flows, are more consistent with anticipated acquisition benefits driving pre-offer RARs. Finally, support for pre-offer RARs representing normal bidder returns is strengthened if cash flow changes are unrelated to RAR levels, either before or after an offer. 7.1 Annual Report Data Set and Cash Flow Measures This analysis relies on net income, earnings before interest and taxes, residual cash flow and revenue information sourced from a data set released in 1995 by the Statex Department of the Australian Stock Exchange (ASX). Originally provided in binary form, this data set was unusable until first converted to text. (30) After considerable effort, I subsequently developed a series of C programmes to assemble the 7668 different item classifications into a consistent report structure. These C programmes contain more than 8,000 lines of code. The Statex Annual Report data set contains information for 2545 different companies. Balance sheet dates range between 1962 and 1994. (31) The longest span for any company is 30 years. Only 8 companies contain such data. The shortest span is just one year. Twenty six companies have only a single year of data. The average span for all companies is 8.5 years. Despite the extensive translation programmes, not all of the data is usable for every company. Unusable data was identified within the translation process by means of consistency checks. Only data passing these checks are used in tables that follow. Among the 1118 bidder sample firms, 1015 contain entries in the final Statex Annual Report file. The remaining bidders are unknown to the data set and are excluded from this part of the analysis. For each of the identified companies, balance sheet dates were extracted and compared with offer dates. Balance sheet dates within six months of the offer date were classified as year 0 reports (henceforth curly braces {.} are used to identify years relative to an offer announcement and square braces [.] retained for months). All available information on sales, net income (32) (NI), earnings before interest and taxes (33) (EBIT), operating cash flow (34) (OCF) and residual cash flow (35) (RCF) were collected from years {-5} through to {+5}. Much of the analysis focuses on residual cash flows (RCF = EBIT - Taxes - Interest - Preference + Depreciation + Amortisations) per adjusted fully paid share. (36) This is because the economics literature argues investors set share prices according to residual cash flow expectations. (37) However, the analysis is not limited to residual cash flows. Also considered is evidence on the operating efficiency of each firm. For this purpose operating cash flows per sales are investigated. Operating cash flows per sales (OCF/Sales) indicate the firm's profitability per dollar of revenues, without distinguishing between obligations to capital suppliers or taxing authorities. Like RCF per adjusted share, OCF/Sales is not affected by biases due to the successful bidder's choice between the purchase and pooling of interests reporting methods. OCF/Sales also implicitly controls for price inflation, (38) since both OCF and Sales can be expected to grow with inflation. 7.2 Cash Flows and RARs Relationships between pre-offer RARs and investor cash flow expectations are investigated below. Several different samples are used. Each tests the strength of associations between estimated RAR levels and observed cash flows in years surrounding an offer. Associations between RARs and cash flows are tested by comparing typical cash flows within each of three different sample segments. RAR values are not contrasted with cash flow levels directly. This is to reduce the possibility of volatile RAR estimates spuriously correlating with cash flows and leading to mistaken conclusions on the existence of relationships. Each sample segment is defined by the RAR level of its members, according to relative RAR levels across all firms, (39) not the distribution of RARs within a sample. Firms with RAR estimates among the largest 30% of all firms are classified as High-range RAR firms. Those with RAR estimates in the bottom 30% of all firms are labeled Low-range RAR firms. Remaining firms, in the middle 40% of all firms, are called Mid-range RAR firms. Population based rankings illustrate characteristics of the sample. It is possible for an entire sample to be classified as High-range, say, if all sample RAR estimates exceed population boundaries for the top 30%. In fact, 39.2% of the bidder sample is ranked High-range, 39.6% is rated Mid-range and just 18.1% are Low-range, which indicates the bidder sample is not representative of the wider population. The bidder sample contains a much larger proportion of High-range and a far smaller proportion of Low-range RAR firms than chance alone permits. (40) As the results of table 1 indicated, bidders possess exceptionally positive RARs during the pre-offer period. Typical cash flow outcomes for High-range firms are compared with those for Mid- and Low-range firms to test associations between pre-offer RARs and cash flows. A positive association, say, is implied when cash flows of High-range firms are significantly larger than those of Low-range firms. Table 3 reports typical operating cash flow measures for each RAR segment of the bidder sample. Table 4 presents typical residual cash flow measures. Both tables contain several panels. The top half of each table provides estimates from firms whose annual reports from Statex are available every year between {-4} and {+4} and satisfy all consistency checks. Such firms are labeled Complete-Data Firms. The bottom half of each table contain results from all available annual reports satisfying consistency checks in just that year. No requirement is imposed for data outside the current year. These are the All-Available-Data results. Three smaller panels appear within each table half. These correspond to bidder sample segments for High-, Mid- and Low-range RAR firms. Average and Median cash flows of all bidders within each classification are given. Two lines appear next to each Average and Median label. The first line gives the level of each measure while the second line corresponds to its value relative to offer year levels. (41) Obs refers to the number of firms contributing to the level measures. Tables 3 and 4 report results after removing cash flow outcomes which were three or more standard deviations from mean values for that sample segment. That is, cash flow measures are trimmed to exclude the most extreme outcomes. (42) Only a small number of firms are removed by trimming. Trimming explains why Obs for Complete-Data firms is not identical between {-4} and {+4}. Results are examined for both the Complete-Data and All-Available-Data sets because it is not clear which is likely to be the more reliable. Positive survivor biases can be predicted from various sources for Complete-Data results while results for All-Available-Data lack suitable benchmarks and are subject to biases linked with sample composition. A positive bias for the Complete-Data firms can be expected because firms which are known to exist between {-4} and {+4} do not expire in that time. Financially distressed companies, who cease to provide annual reports, are therefore excluded. By definition, contributing companies perform more strongly than non-contributing firms, on average. (43) The longer the duration of required existence, the greater is the likelihood of positive survivor bias. (44) Results for the All-Available-Data set also involve difficulties. Two of these are the absence of suitable benchmarks and the lack of control over sample composition. There is no reason to assume that all bidders, or their shareholders, achieve, or expect to achieve, the same level of cash flows in comparison to other companies. An operating margin of 10%, say, may be high in one industry and low in another. Similarly, residual cash flow of 30 cents per share may be high for a firm developing its products and low for an established firm. Without some mechanism to measure performance relative to expected or required levels, (45) we cannot determine whether an individual bidder's accounting based performance is exceptionally good, bad or indifferent. Given that cash flow differences between companies may not accurately highlight performance differences, any factor influencing which companies contribute to a sample measure can cause biases. Cross-company statistics summarise results only for companies contributing to the measure at that time. When companies fall in and out of a sample for reasons outside our control, high/low cross-company outcomes in one year may indicate good/bad performance, or may just be caused by changes in the mix of contributing companies. For instance, evidence of cash flow increases in {-1} may be caused by greater representation in the sample of large cash flow firms, rather than being a typical bidder characteristic. Comparison of cross-company results between years also becomes problematic. A distinct advantage of the Complete-Data sample is its stability through time. Even though it is not representative of all firms and cash flow estimates for any particular year may be positively biased, the Complete-Data sample can provide reliable estimates of change for survivors. Results from All-Available-Data samples may be more representative, but suffer from a lack of control over alternative explanations. Results for both samples are presented below. Table 3 presents bidder operating margins per sales and table 4 records residual cash flows per adjusted share. Figures 3 and 4 illustrate median operating margins and residual cash flows from tables 3 and 4. Attention focuses on medians because both performance measures reveal sensitivity to influential results in the upper tail of values, particularly for High-range bidders. Table 4, for example, shows the average residual cash flow is $1.177 in {+5} for High-range bidders with Complete-Data when the median value is just $0.624. (46) Skewness is present in residual cash flows across most years, although it is less serious as bidder RAR rankings drop. Operating margins in table 3 are also affected by skewness, with High-range bidders again the worst affected. Figures 3 and 4 present median outcomes for both the Complete- and All-Available-Data sets. As expected, measures for the Complete-Data set are generally larger than those for All-Available firms, especially in years following an offer, when implicit sample restrictions on continued survival are least stringent. (47) Figures 3 and 4 display evidence consistent with positive survivor biases. [FIGURES 3-4 OMITTED] Despite understandable differences between measures using Complete- and All-Available-Data, their similarities are more revealing. The same trends appear in figure 3 for operating margins from both data sets. Each shows operating margins for High-, Mid- and Low-range bidders are reasonably alike five years before an offer. Each shows differences soon emerge between bidder types. In the case of High-range bidders, margins steadily grow between {-5} and {0} and thereafter decline. Five years after an offer, operating margins for High-range bidders are still higher than before the offer but the rapid gains between {-2} and {0} are lost. The trend for Mid-range bidders is different. Their margins are reasonably constant, being only marginally higher near the offer year. Five years after an offer, their margins are similar to those reported five years before it. The pattern for Low-range bidders also differs. Low-range bidders maintain margins at much the same level between {-5} and {0}. Then margins drop. Five years after the offer, Low-range bidder margins are only 75% of the offer year value. The proposition that investors can and do distinguish between the future prospects of bidders is supported by the patterns in figure 3. Bidders with improving efficiency experience significant share price gains before announcing an offer. Bidders with declining efficiency suffer share price losses and negative pre-offer RARs. Bidders with negligible margin changes display negligible price movements. The patterns are all consistent with investors anticipating subsequent bidder performance. Figure 4 adds further support. It reports median cash flow entitlements per adjusted share for each bidder type. Entitlements are shown to increase between {-5} and {+5} for all bidder types, but, importantly, not at the same rate. High-range bidder shareholders enjoy the greatest increases, particularly from {-1} onwards. The rate of increase is reasonably steady for Mid-range bidders. In contrast, Low-range bidders provide the smallest growth, slowing noticeably between {-3} and {-1}. Figure 4 also explains an anomalous aspect of the relationship between operating margins and capitalised investor anticipation. Figure 3 shows pre-offer movements in bidder share prices are consistent with investor anticipation of reported pre-offer operating margins. Nevertheless, Figure 3's results are not uniformly consistent with anticipation arguments. Of particular concern is the general deterioration in margins after an offer announcement, especially for High-range bidders, because investors do not react at the offer by reducing bidder share prices. In fact, offer window reactions are generally positive, according to the (0,1) market model (see Simmonds 2003). One explanation for offer period expectations not falling is that investors may not be surprised by an offer and its consequences for operating margins. The fact that investors still enjoy higher margins five years after an offer than were available five years before, even after the performance slide, tends to support continuing confidence by investors. However, this explanation is inadequate. Investors are not just confident, they seem positively surprised by an offer. If they were not surprised, bidder prices would not rise near an offer announcement. Investors appear to increase valuations of future prospects, even though bidder operating efficiency soon falls sharply. The explanation figure 4 supports is that cash flow entitlements do not fall with operating margins. Margins deteriorate from {0} levels but High-range bidders continue to enjoy sustained cash flow growth. Even though each dollar of sales generates smaller profits than immediately before the offer, the distribution flowing through to High-range bidder shareholders actually continues to rise. A plausible explanation for a rise in bidder prices at an offer announcement is that bidder shareholders subsequently capture real cash flow benefits. (48) Residual cash flows for many bidder shareholders continue to grow after an offer in figure 4. Those of High-range bidders enjoy the fastest post-offer growth but many other bidders also improve later cash flows. Only 18.1% of the bidder sample shows little ability to expand cash flows, the Low-range bidders. Since the bidder sample contains disproportionately many firms producing rapid cash flow growth and relatively few, by market standards, making negligible cash flow improvements, a positive offer period response is not surprising for a 'typical' bidder. Figure 4 also suggests pre-offer RARs are affected by anticipation of subsequent cash flow entitlements, perhaps including and beyond an offer announcement. High-range bidders do produce the fastest growing cash flows. Low-range bidders do report the slowest cash flow growth. Mid-range bidders do generate cash flow growth between that of High- and Low-range firms. All these results are as expected when pre-offer RAR reflect capitalised expectations of subsequent cash flow entitlements. 7.2.1 Pre-Offer Bidder RARs and Unexpected Residual Cash Flow Changes A positive correlation between pre-offer RAR measures and residual cash flow growth is illustrated in Figure 4. Whilst this is consistent with efficient market predictions and reassuring for the particular cash flow measure adopted here, the issue remains as to whether pre-offer RAR estimates usefully describe 'normal' returns for bidders. Of particular concern is the possibility that pre-offer RARs are influenced by previously unexpected cash flow changes. Rather than measuring just -qB, pre-offer RARs could be influenced by [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. Figure 4 sheds light oil this question too, however, in order to understand the relationship between pre-offer RAR estimates and news in cash flow changes, let us first identify exactly which annual reports present cash flow news to investors when pre-offer RARs are measured. Pre-offer RARs are measured here for every bidder from returns in months [34,-11] (see fig. 2). Available cash flow information comes from annual reports whose balance sheet dates vary from one firm to another. Cash flow information is aligned for each bidder with offer dates by setting the report closest to an offer month to be year {0}. The nearest balance sheet date can vary from bidder to bidder. It may be five months before an offer for one bidder and five months after an offer for another. On average, though, the nearest balance sheet date across the bidder sample falls in the offer month. (49) On average, therefore, year {0} corresponds to months [-11, 0], year {-1} represents months [ 23,-12] and year {-2} matches months [-35,-24]. Comparing these ranges with [-34,-11] shows pre-offer RARs are typically estimated over intervals matching reports for years {-2} and {-1}. It is news associated with {-2} and {-1} that most likely influences pre-offer RARs. (50) In an efficient market all information sources are monitored when assessing a company's future prospects, including reports by other firms in the same industry. The enormous range of available information sources complicates the task of defining exactly when cash flow news is capitalized by the market. It is arguable though that an efficient market is well prepared before an earnings announcement. (51) Considering these other sources, news associated with results for {-1} is likely to be at least partially capitalised within [-34,-11]. The same reasoning implies price responses to news from {-3} results may often be negligible within [-34,-11] because that information is largely capitalised before [-34]. According to equation (1), share prices react to unexpected change, not just for its immediate effects but also for its previously unanticipated future consequences. It is unexpected change that significantly influences RARs and it is unexpected change within [-34,-11] that most likely explains observed RAR levels. In particular, unexpected news from reports for {-2} and {-1} is most likely to influence RAR estimates. Unexpected news from earlier reports is not likely to affect estimated pre-offer RARs because that news is already incorporated into share prices before [-34,-11]. Before unexpected change can be identified in the latest news release, existing expectations must first be assessed. For this purpose, we assume existing predictions are shaped by previous reports and that the latest report guides current expectations. In other words, exceptional change since the last report will tend to be associated with unexpected news on cash flow growth. Firms with no appreciable change in cash flows are likely to comply with investors' previous growth rate forecasts. When investors look only to the future, reports from years {-5}, {4} and {-3} will bear little relationship with RAR estimates from [-34,-11] because any associated price effects are already capitalised. Figure 4 confirms this. It shows RAR types are not related to cash flow growth before [-34,-11]. The pace of residual cash flow growth in these years is equivalent across RAR types. News from {-2} and {-1} should be more influential. If a previous report guides predictions for the next, firms with exceptionally large cash flow growth will tend to produce superior RAR measures. Similarly, inferior RAR estimates will be linked with firms with exceptionally poor cash flow growth. Firms with no appreciable change in cash flow growth rates can be expected to produce unexceptional RAR values. These predictions are all confirmed by the growth rate patterns in figure 4. Growth rates jump sharply between {-2} and {-1} for cash flows of High-range bidders. Growth rates for Low-range bidders are noticeably lower after {-3} and remain low through to {-1}. Finally, growth rates for Midrange bidders do not change appreciably near {-2} or {-1}. All these results indicate pre-offer RARs are strongly related to cash flow news reported at {-2} and {-1}. It should be noted that cash flow news in reports at {-2} and {-1} is disclosed well before a takeover offer is announced. That news concerns the performance of existing bidder assets, before any acquisition plan is implemented. High-range bidders are those whose pre-offer performance has been outstanding just before an offer, Low-range bidders are firms with inferior performance leading up to the offer and Mid-range firms maintain performance standards at existing levels. Rather than representing a 'normal' component of bidder returns, pre-offer RARs are influenced by unexpected news on the performance of existing bidder assets. When contrasted with the insignificant outcomes obtained already for tests of anticipated acquisition benefits, these cash flow results imply market model measures of 'normal' returns are more significantly affected by conditions immediately preceding a takeover offer announcement, than the potential takeover itself. In the next section the analysis is extended to firms with similar attributes to bidders at the time of an offer. The influence of pre-offer cash flow announcements on RAR estimates is explored in this broader context and the significance of the relationship is measured. 7.2.2 RARs for Non-Takeover Firms and Unexpected Cash Flow News The significance of the previous bidder results is examined in this section after first considering the extent to which RARs for other firms are influenced by cash flow news in an estimation interval. For this purpose, the same analysis is conducted on each of ten random samples of non-takeover look-alike firms. Summary measures are presented for the ten separate samples. Non-takeover firms are those not involved in their own takeovers, either as a bidder or target, during the interval defined in figure 2. Attention focuses on non-takeover firms to isolate takeover related effects as an explanation for PAR levels. (52) Except for its non-takeover activity, a selected firm matches its bidder in terms of size and survival characteristics at the time of the bidder's offer. One look-alike firm is randomly selected for each bidder to form a look-alike sample. Ten such look-alike samples are chosen to increase the accuracy of overall estimates and more clearly reveal underlying relationships between sudden growth rate changes and pre-offer RARs. Ten control samples also permit estimates of significance for the bidder sample. As with the bidder sample, look-alike RAR groupings are determined against population boundaries at each offer month. For instance, High-range RAR look-alikes are firms with RAR estimates from [-34,-11] exceeding the upper 30% boundary for the population at that time. Figure 5 presents the median across the ten trimmed samples of median relative cash flows for Complete-Data firms within each RAR grouping. Cash flows are shown relative to reported year {0} values, to highlight growth rate changes at the level of an individual firm. [FIGURE 5 OMITTED] The main results of figure 4 are all repeated in figure 5. As with the bidder sample, cash flow growth rates before {-2} are equivalent across RAR types, indicating estimated RARs are unrelated to earlier news on cash flow growth. Low-range firms report the largest relative cash flows and High-range firms the smallest, but the rate of change is similar across all three RAR-types. Although it might seem counter intuitive for High-range firms to report the smallest cash flows relative to {0}, this is actually a consequence of RARs reflecting changes first reported at {-1}. It is the sharp rise in growth rates between {-2} and {-1} for High-range firms, the noticeable reduction for Low-range firms and the absence of significant change for Mid-range firms that explains the earlier rankings relative to {0}. High-range firms grow more quickly up to {0} than Low-range firms, so their cash flows are a substantially smaller fraction of {0} values before that growth starts. This is why High-range firms record the lowest relative cash flows before {0}. Figure 5 also indicates growth rates after {-1} are largely unrelated to RAR rankings. Between {-1} and {0), growth rates for different RAR types are indistinguishable from each other. Between {0} and {+1}, Mid-range firms appear to grow slightly faster than other RAR types. Over the longer term, rates do not appear to be significantly different between RAR types. Figure 5 clearly demonstrates a strong relationship between pre-offer RAR rankings and growth rate changes in the most recently reported residual cash flows. Also revealed is the relatively short life of the relationship. RARs and growth rate changes are unrelated before {-2} and become unrelated again after {-1}. It is price adjustments at {-1} that RARs predominantly reflect. This is when expectations are updated to surprising cash flow news. Longer term opportunities may support current price levels but do not explain recent price changes and are not the source of differences in pre-offer RARs. The next table presents median annual growth rates from figure 5 for the ten look-alike samples. It reports significance levels for rate comparisons across RAR types. Also given are corresponding bidder growth rates and their position within the distribution of look-alike sample outcomes. As indicated by figure 5, relative growth rates for look-alike firms are equivalent across RAR types virtually everywhere in table 5 except between {-2} and {-1}. P-values for the Kruskal-Wallis test of equivalence indicate growth is the same across RAR types through {-4} to {-2} and {+1} to {+4}. The most significant differences occur between {-2} and {-1}. This is when annual growth for High-range RAR firms is 45.8%, up sharply from 22.7% in the previous period, growth for Low-range firms is 0.4%, down from 20.3% and growth for Mid-range firms remains unchanged near 21.3%. All these results are consistent with RAR estimates reflecting sudden changes in cash flow growth between {-2} and {-1}, long before events at {0} and beyond. They show RARs measure the capitalized consequences of updated predictions following near-term news. Longer-term relationships are too weak to be noticed. 7.2.3 Cash Flows in the Market and the Timing of Takeover Offers Another important outcome to note from Table 5 is the consistent evidence of lower cash flow growth after {0}, across all RAR types. Before {-2}, cash flows grow by around 21% per year for each RAR type. After {+1}, growth rates drop to just 10%. None of the ten samples records a rate rise after {+1}, so the 11% drop is a statistically significant outcome. An important consequence of declining longer term growth is the perspective it gives on conditions in the broader market when bidders announce offers. Remember, these are results for samples of non-takeover firms matching bidders in terms of size and survival qualities just before an offer is announced. Sampled firms are aligned with bidders in time but not takeover activity. Despite their detachment as takeover participants, sampled firms experience substantial deterioration in economic prosperity around times of bidding activity. Considerable prosperity is displayed before offers from their matched bidders but evaporates soon afterwards. Such patterns are important for what they reveal about bidders. In particular, they suggest offers often occur in harmony with cycles of economic prosperity across the market. Offers frequently emerge after periods of increased economic prosperity and before periods of wider economic slowdown. Timing offers in harmony with economic cycles is not the only interpretation of slower growth for non-takeover firms following offers by bidders. Rather than being a feature of the market in general, a different possibility is that detachment from acquisitions by non-takeover firms causes the subsequent decline. For instance, lower cash flow growth for non-acquirers may be a consequence of increased market shares for firms who do acquire. Non-takeover firms may also suffer because an opportunity to grow is lost when bidder peers announce offers. Alternatively, a demonstrated inability to grow via acquisitions may affect sales and share prices. All these are plausible alternative explanations. They all lead to the same prediction for bidder growth rates after an offer when compared with those of look-alikes. If not acquiring reduces growth, all these alternatives imply bidder firms will report higher post-offer growth than non-takeover firms. Table 5 sheds light on this possibility too. The performance of bidder firms relative to non-takeover peers is given in the second panel of table 5. Before {-2}, bidder cash flow growth across RAR types is higher than that of most non-takeover look-alikes, although not significantly so. Between {-2} and {-1}, bidder growth exceeds that of almost all look-alikes. Median relative growth of 80.8% for High-range RAR bidders is larger than that achieved by any look-alike sample, while the relative growth of Mid- and Low-range bidders is larger than that for 90% of look-alikes. However, the superior pre-offer gains of bidders are soon eliminated following a bid, so that between {+1} and {+4} bidder relative growth is exceeded by most look-alike samples. Bidders move from being average to exceptionally strong performers before an offer to become relatively poor performers afterwards. (53) Contrary to predictions of cash flow growth being sustained by acquisition policies, bidder growth is not significantly higher than that of non-takeover look-alikes, except possibly near the offer announcement. (54) Just before an offer announcement bidders report superior cash flow growth but cannot maintain that growth beyond {+1}. (55) The relatively poor post-offer growth rates of bidders imply the slowdown by non-takeover look-alikes is not caused by their non-participation in acquisitions. Rather, the rise and subsequent fall is a feature of broader market conditions and the harmonisation of takeover offers within that setting. Links between bidding activity and cash flow growth in the broader market are not the first instance of an association between bidder activity and wider economic conditions. This same characteristic was demonstrated earlier in terms of waves of takeover activity and recent market conditions (see fig. 1) where takeovers were found to be more common after general improvements in market conditions. That relationship is now demonstrated in terms of recent cash flows. Bidders are more numerous, or more likely to emerge, after recent improvements in their own performance. This is when monitors (Boards of Directors, majority shareholders, lenders) of bidder managers can be expected to favour expansion plans from their 'obviously' skilful managers. It is when conditions are most likely favourable for initiating takeovers. (56) It is also when management's confidence in its own abilities is most likely greatest (Roll 1986). 8. Controlling for Self-Selection Biases The timing of takeover offers is not random. It is an endogenous feature of the market for corporate control which event studies have hitherto largely ignored. Papers from the conditional methods literature, such as Prabhala (1997), deal with the measurement of private information in event studies when updated expectations are conditioned on insider information and motives. They do not focus on the wider timing phenomenon. Kendig (1998) does draw attention to these timing aspects of Australian takeovers, arguing that takeover waves occur because corporate governance mechanisms are less severe during economic prosperity, when management's self-confidence also grows. Less binding constraints and increased self-confidence lead to more agency motivated takeovers. But Kendig does not address the issue of how this clustering affects event studies and the measurement of acquisition benefits for society. In this paper, pre-offer returns are argued to reflect features of existing bidder investments which event studies must properly control before event related changes can be isolated from other systematic characteristics of the sample. Significant systematic differences are demonstrated in table 1, where bidder pre-offer RARs and betas are both found to be significantly larger than corresponding measures for size and survivor matched portfolios. Significant differences are not eliminated by varying the benchmark index adopted, or by selecting control samples to match on other empirical return regularities. Nor are the differences due to extreme outcomes or skewness in the bidder sample. Malatesta and Thompson's (1985) model explains differences in pre-offer RARs between bidders and other firms as a consequence of anticipated acquisition benefits. Although their model predicts negative pre-offer RARs for bidders, an extended version can forecast positive RARs when the initial capitalisation of acquisition benefits is included. Section 6 tested for evidence of significant anticipated acquisition benefits and found little support. More compelling was the evidence in section 7 linking pre-offer RAR levels with unexpectedly good news on cash flow surges between {-2} and {-1}. Cash flow growth for bidders was found to be among the fastest at that time. After an offer, bidder cash flow growth slowed to levels comparable with that of non-takeover look-alikes. Bidders and their non-takeover look-alikes present evidence of rapid cash flow growth before an offer and subsequent declines afterwards but the pre-offer experience of bidders tends to be exceptional, both in comparison to look-alike behaviour and in comparison to their own post-offer performance. The important consequences of these features are, firstly, bidders do display different attributes to other firms in the market, even when size and survival qualities are controlled. Controlling empirical size and survival based return regularities through suitably constructed benchmarks, or by testing significance against portfolios matching bidders on these same characteristics, does not remove the systematic differences for bidders. Compared with null hypothesis distributions of standard testing procedures, the bidder sample is over-represented with high preoffer RAR and high beta firms. Secondly, these systematic features do not both describe return characteristics that can be considered 'normal' for bidders. Pre-offer RAR estimates reflect positive price responses to news released considerably before an offer announcement. They capture the capitalised consequences of unexpectedly good news. There is no evidence linking pre-offer RAR estimates with unexpected good news on cash flows after the pre-offer period. This is to be expected in an efficient market where future outcomes are unbiasedly anticipated. However, it means market model estimates of 'normal' returns through the offer period are inflated by the capitalised responses to pre-offer good news because pre-offer conditions do not accurately represent offer window conditions. The counter-factual return estimated by pre-offer conditions is biased. Thirdly, bias in 'normal' return estimates exists because systematic differences in the bidder sample are a consequence of bidders choosing when to announce takeover offers. The timing of offers is not haphazard. Offers frequently follow periods of economic improvement across the market, when the performance of the bidder has been among the best. Offers also tend to precede periods of broader economic decline. Greater bidding activity after periods of prosperity is not a chance outcome but one predicted when corporate governance mechanisms bind less tightly and management self-confidence grows. Finally, since the chance of sampling high and low pre-offer RAR estimates is not independent of being identified as a bidder, (57) and pre-offer RARs do not accurately describe return characteristics for the offer window, traditional applications of the market model produce a biased overall estimate of bidding shareholder response at an offer announcement. Diversification across the sample of 'normal' returns that are too high for some bidders and too low for others, does not protect against these individual biases. Traditional applications of the market model are biased because that model is systematically influenced by characteristics associated with being a bidder. The systematic features must be controlled to remove that bias. The next section describes appropriate control procedures. 8.1 Controlling for Pre-Offer Characteristics of Bidders Bootstrapping methods can be applied to control systematic features of bidder sample returns and produce unbiased measures of offer window responses from the market model. The procedure described in section 3 is modified to also identify and select eligible controls only from firms with similar pre-offer return attributes to bidders. That is, control firms should not only match bidders in terms of size and survival qualities, but also in terms of pre-offer RAR and beta characteristics. Applying such controls restores comparability for sample selection probabilities on these attributes between bidder and bootstrap samples. It ensures bootstrap samples are taken from that part of the market with similar attributes to genuine bidders. Three different procedures are implemented here. The first controls jointly for pre-offer RAR and beta. The second controls only for pre-offer RAR and the final procedure controls only for pre-offer beta. Appendix 2 contains exact matching procedure details. The methods are all straightforward. The only complication comes from ensuring several eligible control firms are available for look-alike selection because the definition of finer look-alike categories reduces the number of eligible alternatives within each category. In the case of joint controls for pre-offer RAR and beta, the exact matching procedure is: (a) Identify the relative size of each firm by the market capitalisation of its fully paid shares two months before the current month. Classify all firms into size quintiles. Size deciles were used previously but are combined here to accommodate the additional restriction that look-alike firms share similar pre-offer RAR and systematic risk characteristics with bidders. (b) At each month between January 1976 and June 1995 find all firms with the same survival qualities required of bidder sample members. That is, find all firms with at least ten valid monthly returns during [-34,-11] relative to the current month, plus at least five valid returns in [+8,+31], plus at least twenty returns during [-34,-11] and [+8,+31] combined, plus a valid return in the current month, [0]. (c) Exclude any firm who received or announced a takeover offer near the current month. Firms whose pre- or post-offer estimation periods ([-34,-11] and [+8,+31]) overlap with those for the current period are excluded. Such firms are not excluded from selection at other times. This filter ensures any differences found between sampled bidders and look-alikes are not due to comparisons with other bidders or targets at different stages in their own takeovers. (d) Estimate pre-offer RARs and beta for each firm by the simple market model using a single equal-weighted market index. Determine decile rankings across the market for each estimate in models where market related risk exceeds 1% and then assign the following classification. Identify the same codes for bidder firms.
Joint Classification of Pre-Offer RAR and Beta ([double dagger])
Pre-offer Beta Decile
4 [less than
or equal to]
Beta Decile RAR decile Beta Decile
[less than [less than or [greater than
Pre-offer RAR Decile equal to] 3 equal to] 7 equal to] 8
RAR decile [less than or 1 2 3
equal to] 3
4 [less than or equal 4 5 6
to] RAR decile [less
than or equal to] 7
RAR decile [greater than 7 8 9
or equal to] 8
Note: ([double dagger]) Beta refers here to the slope estimate in a
market model based on a single market index of equal-weighted returns.
(e) For each sampled bidder, select randomly with replacement one survivor/non-target/ non-bidder/equal market capitalised firm with the same pre-offer RAP/Beta coding as the bidder at the offer announcement date. If at least five alternative look-alike firms are not available in the same joint RAP/Beta segment combine across beta categories until five or more alternatives exist. (Beta categories were collapsed for one hundred and sixty one bidders.) Repeat look-alike selection until 100 samples of look-alikes are obtained. 8.2 Offer Window Reactions when Pre-offer Bidder Characteristics are Controlled Table 7 presents results for the market model from each of the different bootstrap selection criteria. Results are reported for the Survivor & Size Adjusted EW Market benchmark. (Other index benchmarks produce similar results and are not reported here.) The first panel is copied from table 6 of Simmonds (2003). It gives CAR estimates for bidders over [-3,+3] when size, survival and takeover participation characteristics are controlled during bootstrap sample selection but pre-offer return attributes are not. It is labeled No Extra Controls. It shows bidder CARs for [-3,+3] are typically negative; median and mean measures are -1.7% and -1.6% when all robustness restrictions are applied. Furthermore, 52.2% of the bidder sample generates a negative offer window CAR. Traditional t--statistics indicate the negative bidder CAR is not significant. Such results are all consistent with many previous studies in the takeover literature. Insignificance is also confirmed by results for the bootstrap samples where 63%/71% (median/mean) of control samples report more negative offer window CARs than bidders. The typical look-alike sample CAR is -1.9%/-2.3% (median/mean). According to this procedure, bidder shareholders are not surprised by an offer announcement. The same conclusion obtains whether benchmarks are size-, survivor- or both size- and survivor-adjusted. The second panel, labeled Pre-offer RAR and Beta, reports results for bootstrap samples selected by the procedure outlined in the previous section. Size/survival status/takeover participation/pre-offer RAR ranking and pre-offer beta ranking are all defined for eligible look-alike candidates. One hundred look-alike samples are selected. Each one matches the pre-offer characteristics of genuine bidders more closely than those selected for No Extra Controls because random selection is now constrained to occur within each of these characteristics, rather than across some. A source of variability is therefore removed from comparisons. Bidder results are as before because changes only affect the definition of eligible bootstrap samples. Constraining pre-offer RAR and beta to match bidder characteristics more closely produces bootstrap sample CARs that are more negative than when look-alike selection ignores these variables. Typical look-alike sample results are now--4.6%/-5.2% (median/mean). 56.0% of bootstrap samples produce negative CARs and no look-alike sample ever achieves the same offer window CAR as the genuine bidder sample. According to these results bidder CARs for [-3,+3] are significantly higher than those of comparable controls by 2.9%/3.6%. Investors are surprised by an offer announcement and typically increase their valuations of the bidder's future prospects. Once again the conclusion is robust to benchmarks that are size-, survivor- or both size- and survivor-adjusted. Panels Pre-offer RAR and Pre-offer Beta show controlling for pre-offer RAR is more influential than controlling for pre-offer beta. (58) In Pre-offer RAR, look-alike samples are aligned with bidders on pre-offer RARs while beta is allowed to vary stochastically. Although only 30 look-alike samples are selected, the conclusion is unmistakably significant; bidder CARs strongly exceed look-alike alternatives. For instance, bidder net gains are now 4.5%/6.1% when size- and survival-adjusted benchmarks are applied and all robustness restrictions imposed. On the other hand, controlling pre-offer betas and not controlling pre-offer RARs produces look-alike CARs that are not significantly different from those of bidders (see Pre-offer Beta). Again, only 30 look-alike samples are used but the conclusion is unlikely to change when more samples are selected. Too many look-alike samples obtain higher and too many obtain lower CARs than bidders to conclude significant differences exist. 8.3 A Simpler Approach Table 7 shows the dominant distortion in market model CARs comes from pre-offer RARs. When bidder pre-offer CARs are not carefully matched with look-alike samples, investor responses to bidder offer announcements are found to be insignificant. When the pre-offer RAR experiences of bidders and look-alikes agree, investor responses significantly exceed returns available to shareholders in the control firms. Pre-offer beta characteristics do not influence CAR measurement sufficiently to alter the significance of CAR measures. In this paper, pre-offer RARs are argued to predominantly reflect capitalised reactions to unexpectedly good news on surges in cash flow growth, commonly reported by bidders before an offer is announced. A close association between pre-offer RAR rankings and pre-offer cash flow surges has been demonstrated. It has also been shown that surges beyond the pre-offer period are not a regular feature for bidder shareholders and are not related to pre-offer RARs. The capitalised price response to the pre-offer surge produces a boost to pre-offer RAR estimates that does not typify 'normal' returns for bidders. Rather than describing 'normal' returns, the pre-offer RAR experience of many bidders is abnormal as a measure of expected offer window conditions. It exaggerates that portion of returns that could be considered 'normal'. When abnormally large pre-offer RAR estimates are applied to predict normal offer window conditions, negative CARs are the result. This is confirmed by look-alike samples constrained to match the pre-offer experiences of bidders. Negative CARS are again produced, even more negative than for bidders. If pre-offer RAR estimates tend to exaggerate normal offer window conditions, it follows that less biased measures of abnormal return activity near an offer may be obtained when no RAR adjustment is applied. This is despite the possibility that a smaller 'normal' return does actually exist, as in the sense of Malatesta and Thompson (1985). Not applying any RAR adjustment and relying instead on overall population characteristics, could produce less biased CAR measures than abnormal pre-offer RARs. Since CARs are not significantly influenced by pre-offer betas, an adjustment for systematic covariation is also not required. These characteristics describe the (0,1) market model. They imply the (0,1) market model, with its population-wide attributes, could provide a less biased measure of offer window reactions than a market model distorted by widespread pre-offer gains common within the bidder sample. By not applying individual adjustments, estimated during periods of unusual share price activity, the (0,1) market model avoids systematic distortions affecting the market model. Consequently, it can provide less biased estimates of normal returns for the offer window and so generate less biased measures of investor responses to an offer announcement than the market model. A suitably chosen contemporaneous benchmark can provide a better measure of ongoing 'normal' returns for bidders than their own recent experiences. Table 8 addresses this issue. Table 8 reports a similar range of results as those appearing in table 7. Attention focuses again on the most flexible benchmark, the Survivor & Size Adjusted EW Market benchmark, but similar results are found for the others. The panel labeled No Extra Controls comes from table 8 of Simmonds (2003). It reports CARs for the (0,1) market model over [-3,+3] when pre-offer RAR and beta are not controlled in look-alike selection. Size and survival characteristics are controlled, as is takeover participation. The benchmark index is also matched on size and survival qualities. 100 look-alike samples are examined. Bidder CARs over [-3,+3] are 3.4%/3.5% (median/mean), when all robustness measures are applied. T-statistics indicate the bidder gains are significant. So too do the look-alike results, which are typically smaller than bidder values by 1.4%/1.8%. Only 93% of look-alike samples produce higher median CARs than genuine bidders and even fewer report higher mean CARs. In addition, fewer bidders now report negative CARs (44.7%). Ensuring look-alike candidates simultaneously share similar pre-offer RAR and beta characteristics with bidders has little effect on any of the (0,1) market model results (see Pre-offer RAR and Beta). Typical outcomes for bootstrap samples are slightly smaller and empirical significance levels slightly higher but the main conclusions are unaffected. Controlling for pre-offer beta attributes also has little effect (see Pre-offer Beta). The greatest influence on (0,1) market model CARs occurs when pre-offer RAR values are matched between bidders and look-alikes. Bidder gains then exceed typical look-alike values by 2.2%/2.7% (median/mean) and are always larger than look-alike sample results (see Pre-offer eAR). This lack of sensitivity by the (0,1) market model to extra controls on pre-offer RAR and beta characteristics is as expected when the pre-offer attributes are unrelated to normal returns through an offer window. It shows contemporaneous measures, based on population-wide attributes, are more accurate indicators of ongoing normal returns than the past characteristics of bidders. Such an outcome is as expected when the market is efficient and pre-offer attributes reflect the consequences of rapid capitalisation of infrequent significant news. It is the overrepresentation of such attributes in the bidder sample that ultimately leads to biased market model results across the sample. The (0,1) market model unaffected by the prior excellence of bidders and so is immune to their over-representation of this characteristic relative to the wider population. Whether or not the (0,1) market model controls for pre-offer RAR and beta, it concludes that bidding shareholder wealth increases significantly during the offer window. Gains are small but surpass those recorded in almost every comparable portfolio of size/survivor/time matched non-takeover firms. A comparison of results between the market and (0,1) market models finds similar outcomes and conclusions for each model when pre-offer RAR and beta are controlled. Estimated net CAR gains for the market model are near 2.9%/3.6% while those for the (0,1) market model are near 1.6%/2.1%. Net gains are significant under both models. When pre-offer RAR and beta are not controlled, market model net gains drop to just 0.2%/0.7%, an insignificant result. This is because look-alike samples do not then contain the same proportion of exaggerated normal return measures as the bidder sample. In contrast, (0,1) market model net CARs, which are unaffected by the distorting influence of pre-offer performance, remain significant at 1.4%/1.8%. Results for the (0,1) market model are robust to pre-offer return attribute controls but those for the market model are sensitive to these same characteristics. 9. Conclusion Ideally, event-window abnormal returns are determined after identifying and removing characteristics representing 'normal' or expected returns. A 'normal' return is that which is expected to occur in the absence of an event; it is the counterfactual return. In the case of bidders, normal returns estimated by the market model are significantly different from those estimated by the (0,1) market model and lead to different conclusions on investors' perceived acquisition benefits. That the two models should disagree may not seem surprising because the (0,1) market model is a constrained version of the market model and always applies the same normal return estimate to all firms, without regard to their individual return characteristics. In contrast, the market model adapts to the particular characteristics of an individual firm. This greater flexibility could be considered an advantage for the market model and lead it to produce superior estimates of normal returns through an offer window. However, this paper showed that intuition is mistaken. Despite its inflexibility, the (0,1) market model was found to be less sensitive to factors occurring systematically throughout a bidder sample. These factors tend to exaggerate market-model normal-return estimates and negatively bias its ability to record significant shareholder gains at an offer announcement. The explanation for this surprising outcome consists of three parts. The first part rests on the property of an efficient market to capitalise the expected future consequences of news affecting cash flows. This leads to a tendency for estimated pre-offer market-model parameters to describe more than just ongoing return characteristics. The final step recognises that the timing of takeover offers is endogenous for bidders. The combination of these features leads to exceptionally positive estimates of 'normal' returns by the market model for much of the bidder sample. As a result, market-model CARs tend to be negatively biased for bidders. The principal underlying cause for negative bias under the market model is the failure of pre-offer parameter estimates, for much of the bidder sample, to accurately describe event-window return characteristics. The failure of traditional testing procedures to recognise this feature means commonly applied tests are misspecified. The bidder sample contains a significantly higher proportion of firms with exceptionally large pre-offer PARs than a random selection from the market. The large pre-offer RARs of bidders are not a spurious characteristic but correspond to evidence on their own outstanding recent achievements in growing shareholder cash flows. This is no accident. It shows the timing of an offer announcement is not haphazard but frequently follows periods when the bidder's own cash flow performance has markedly improved. It is a feature that will emerge in any randomly selected sample of bidders. Event-study methods must recognise and accommodate these endogenous features if investor reactions to an offer are to be accurately measured. Failing to control market-model measures for pre-offer characteristics of the bidder sample, which are correlated with the decision to bid, leads traditional testing procedures to conclude investors do not approve of the offer proposal. This is because traditional testing procedures, even robust, non-parametric procedures, apply null hypothesis distributions that do not recognise the inherent properties of the bidder sample. When these properties are specifically matched in the null distribution, investors are found to value acquisition proposals. The influence of pre-offer characteristics on event-window estimates for normal returns also explains why the (0,1) market model is found to provide superior measures of investor reactions to an offer. Unlike the market model, the (0,1) market model is unaffected by the unusual pre-offer behaviour of returns. It applies the same, population-wide standard to each firm. An immunity from pre-offer distortions means a population-wide standard can contain less bias, particularly across the entire sample, than one that is sensitive to these influences. In the case of bidders, the ability to announce an offer at a time of their own choosing, and a shared preference for doing so after their own firms report exceptional improvements in cash flow growth, lead to distortions in market model parameters that invalidate traditional testing procedures for that model. Under these conditions, the cruder (0,1) market model provides superior measures of investor reactions to an offer announcement. When testing procedures are adjusted to recognise the particular characteristics of a bidder sample, investors are found to value takeover offers. The (0,1) market model leads to this same conclusion without the need for more sophisticated testing procedures. Both models find bidding shareholder wealth significantly increases when a takeover offer is announced. (Date of receipt of final transcript: November, 2002. Accepted by Garry Twite, Special Issue Editor.) Appendix 1 Adjustments to Reported Cash Flows per Share Adjustments applied to residual cash flows are described in this appendix. These are necessary to consistently measure the cash flow entitlements of an individual share through time. Throughout this analysis, residual cash flows are standardised by the number of adjusted fully paid shares Adjustments to fully paid share counts are described in detail below. The task of standardising by the number of reported fully paid shares introduced its own difficulties. These are discussed first. Independent checking of issued share counts identified errors in the Statex Annual Report file. The absence of consistency checks in that data set also made it impossible to confirm share counts were actually recorded in the stated units, thousands or millions, say. Nor could changes be tested between years. There was no method to reliably confirm available data. Consequently, the number of listed shares was gathered from the Share Price and Price Relatives data file (SPPR), as at the end of the balance sheet date month. The SPPR records monthly counts for fully paid ordinary shares quickly and consistently. All new shares issued from splits, rights and bonus issues are included from the effective date of the issue. Share changes due to consolidations are similarly implemented on the effective date of the change. Share counts are fluid and fluctuate over time as companies adjust their capital needs by raising or returning funds to shareholders. When share numbers change, so too can the reported cash flow entitlements of an individual share. As explained below, capital structure changes affecting existing share holdings require two adjustments to restore consistency between annual reports. Fully paid share counts are adjusted for two factors. The first factor concerns the consequences of new issue/consolidation related changes that dilute/inflate cash flow entitlements. This impact is identified and added back as increased/decreased ownership for a representative shareholder. (59) The second adjustment is similar to the first except it applies to dividend payments. Its purpose is to reduce differences in long term cash flows arising from variations in dividend payout ratios between firms. The first or Bonus Element adjustment is the more influential. A simple example illustrates its essential features. Suppose a one for one bonus issue occurs. The number of issued shares doubles and the reported cash flow per share halves. The cash flow entitlements of each pre-bonus-issue-shareholder are, however, unchanged. What changes is that two post-bonus issue shares must be combined to accurately compare cash flows with pre-bonus-issue periods. The Bonus Element adjustment makes this correction. It identifies the bonus element in all capitalisation changes and inflates/deflates share ownership of a representative investor to preserve cash flow entitlements through time. In this example, post-bonus-issue share ownership doubles for the representative investor. Consequently, adjusted cash flows are twice the reported cash flows per issued share. The second adjustment to share numbers takes account of differing dividend payment policies. Different payment policies are relevant because firms with low payout ratios retain greater funds for reinvestment. If retained funds are reinvested in positive NPV projects, low payout firms will, over time, produce higher residual cash flows. To reduce the effect of differing payout ratios we assume a representative investor immediately reinvests all dividends by purchasing new shares with the dividends at the share's market value in the ex-dividend date month. (60) The Dividend adjustment provides a consistent basis for comparison by converting dividends into share equivalents and increasing the proportional ownership of a representative investor. More formally, the Bonus Element Adjustment in month t for firm i is 1 + [CapitalisationAdjustmet.sub.1]/[Price.sub.t] = 1 + [a.sub.i,t]. It corresponds to the total number of shares a representative investor owns alter including the bonus element from capitalisation changes offered to existing shareholders and causing [CapitalisationAdjustment.sub.1]. In the one for one bonus issue example, the number of issued shares increases by 100% and cash flows per share are reduced by 50%. Under these circumstances, [CapitalisationAdjustment.sub.1] = [Price.sub.t] and 1 + [a.sub.i,t] = 2. This is the number of ex-bonus shares that are combined to equalise reported cum- and reported ex- cash flows per share. (61) Equalisation is necessary because capitalisation changes offered to cure-change shareholders redistribute cash flow entitlements across a different number of shares. Cum-change shareholders capture the benefits of unfunded equity either as holders of the newly issued shares or by selling their entitlements to it. Once changes take effect, the fact that cash flows are distributed over twice the number of shares, say, or that old shareholders sold a portion of their ownership through rights trading, is not recognised in reported per share cash flows. Cash flow changes from altered operating conditions can therefore be hidden when capital structure changes occur. The Bonus Element Adjustment restores comparability to reported cash flows by adding back the entitlements of unfunded issues. The Dividend Adjustment in month t for firm i is 1 + [Dividend.sub.t]/[Price.sub.t] = 1 + [d.sub.i,t]. It corresponds to the number of fully paid shares a representative investor could own in the firm after reinvesting all dividends. (62) Both the Bonus Element and Dividend adjustment are compounded over months between adjacent annual reports for firm i as [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.] Similarly, the adjustment over multiple years is [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.] Letting T={0} represent the offer year balance sheet date, Bonus Element and Dividend adjustments are generated for all available years between T-{-5} and T={+5} for every company examined. Each company's adjustment series is then standardised by its own offer year value. This means all companies share the same offer year adjustment of one and surrounding year adjustments describe changes relative to offer year values. Let (1 + [[??].sub.i,{T}])(1 + [[??].sub.i,{T}]) represent the combined Bonus Element and Dividend adjustment in year {T} relative to the offer, standardised against the offer year value. Adjusted residual cash flows per share, [[??].sub.i,{T}], are then measured as [[??].sub.i,{T}] = RC[F.sub.i,{T}]/[n.sub.i,{T}] (1 + [[??].sub.i,{T}])(1 + [[??].sub.i,{T}]) where [n.sub.i,{T}] is the SPPR number of issued fully paid shares at tile end of the balance sheet date month for the annual report in year {T}. [[??].sub.i,{T}] corresponds to the cash flow entitlements for the number of year {T} shares that are equivalent to one offer year share. In a sense [[??].sub.i{T}] is the real cash flow per share after removing the inflationary/deflationary effects of capitalisation changes and dividends. Finally, both residual and operating cash flows are also compared with offer year values. For example, in the case of residual cash flows, [[??].sub.i,{T}] = [[??].sub.i,{T}/[[??].sub.i,{0}]. Standardising by offer year values generates growth rate estimates for cash flows. It is particularly useful to highlight sudden changes in cash flow growth rates for individual companies. Times of sudden growth rate change indicate when investors are most likely to be surprised by unexpected news and when previous cash flow forecasts are updated. Just before such shifts is when share prices can be expected to adjust. Appendix 2 Bootstrap Procedures for the Market Model when Applied to Bidder Samples 1. Biased Procedure-Size, Survival and Non-Takeover Controlled (a) Identify the relative size of each firm by the market capitalisation of its fully paid shares two months before the current month. Classify all firms into size deciles and sub-deciles within the top decile. (b) At each month between January 1976 and June 1995 find all firms with the same survival qualities required of bidder sample members. That is, find all firms with at least ten valid monthly returns during [-34,-11] relative to the current month, plus at least five valid returns in [+8,+31], plus at least twenty returns during [-34,-11] and [+8,+31] combined, plus a valid return in the current month, [0]. (c) Exclude any firm who received or announced a takeover offer near the current month. Firms whose pre- or post-offer estimation periods ([-34,-11] and [+8,+31]) overlap with those for the current period are excluded. Such firms are not excluded from selection at other times. This filter ensures any differences found between sampled bidders and look-alikes are not due to comparisons with other bidders or targets at different stages in their own takeovers. (d) For each sampled bidder randomly select with replacement one survivor/non-target/non-bidder look-alike firm in the same size category as a bidder at the time an offer is announced. Repeat look-alike selection until 100 samples of look-alikes are obtained. 2. Size, Survival, Non-Takeover, Pre-Offer RAR and Pre-Offer Beta Jointly Controlled (a) Modifies the size procedures above by combining size categories into quintiles, that is, random selections are made from the same size quintile, not deciles/sub-decile. Size deciles are combined to accommodate the additional restriction that look-alike firms share similar pre-offer RAR and systematic risk characteristics with bidders. (b) Pre-offer RARs and systematic risk are estimated for each firm by the simple market model using a single equal-weighted market index. Decile rankings across the market are determined for each estimate in models where market related risk exceeds 1%. These are used to assign the following classification, which are similarly assigned to bidder firms.
Joint Classification of Pre-Offer RAR and Beta ([double dagger])
Pre-offer Beta Decile
4 [less than
or equal to]
Beta Decile RAR decile Beta Decile
[less than [less than or [greater than
Pre-offer RAR Decile equal to] 3 equal to] 7 equal to] 8
RAR decile [less than or 1 2 3
equal to] 3
4 [less than or equal 4 5 6
to] RAR decile [less
than or equal to] 7
RAR decile [greater than 7 8 9
or equal to] 8
Note: ([double dagger]) Beta refers here to the slope estimate in a
market model based on a single market index of equal-weighted returns.
(c) For each sampled bidder randomly select with replacement one survivor/non-target/non-bidder/equal market capitalised firm with the same pre-offer RAR/Beta coding as the bidder at the offer announcement date. If at least five alternative look-alike are not available in the same joint RAR/Beta segment combine across beta categories until five or more alternatives exist. (Beta categories were collapsed for one hundred and sixty one bidders.) Repeat look-alike selection until 100 samples of look-alikes are obtained. 3. Size, Survival. Non-Takeover, Pre-Offer RAR Controlled (a) Apply market capitalization size quintiles as in the previous procedure. (b) Identify the pre-offer RAR decile ranking of all firms. (c) Match sampled bidders with equivalent survivor/non-target/non-bidder/equal size quintile and preoffer RAR decile. If no more than four potential matches exist for any bidder combine pre-offer RAR deciles into quintiles. If no more than four potential matches are still found, combine pre-offer RAR deciles into three groups: the smallest three deciles, the largest three deciles and the remainder. (One hundred and sixty seven bidder entries required coarser RAR grouping than deciles.) Randomly select with replacement one look-alike for each bidder. Repeat look-alike selection until 30 samples are obtained. 4. Size, Survival, Non-Takeover, Pre-Offer Beta Controlled (a) Apply market capitalisation quintiles. (b) Identify the pre-offer beta decile ranking of all firms. (c) Match sampled bidders with equivalent survivor/non-target/non-bidder/equal size quintile and equal pre-offer beta decile. Collapse beta deciles first to quintiles and then into three groups (bottom 30%, top 30% and the rest) until five or more look-alike alternatives are available. Using random selection with replacement choose one look-alike firm for each bidder. Repeat look-alike selection until 30 samples are obtained.
Table 1a
Pre-Offer Return Characteristics of Surviving Bidders Making Offers
Between January 1976 and June 1995
Bidder Sample
Market Benchmark N Median Mean t-[ratio
Robustness .sup.1]
Restriction Set
Panel A: Market Model RAR
EW Market
No Restrictions 1118 0.9% 1.2% 10.92
All Restrictions ([double dagger]) 1053 0.9% 1.1% 11.94
Size Adjusted EW
Market
No Restrictions 1118 0.7% 1.1% 10.37
All Restrictions ([double dagger]) 1049 0.7% 1.0% 11.28
Survivor Adjusted
EW Market
No Restrictions 1092 0.9% 0.7% 3.18
All Restrictions ([double dagger]) 1012 0.8% 1.0% 10.37
Survivor & Size
Adjusted EW Market
No Restrictions 1090 0.7% 0.7% 2.86
All Restrictions ([double dagger]) 1002 0.6% 0.9% 10.19
Panel B: Market Model Beta
EW Market
No Restrictions 1118 0.85 1.01 0.28
All Restrictions ([double dagger]) 1050 0.86 0.98 -0.81
Size Adjusted EW
Market
No Restrictions 1118 0.98 1.08 3.26
All Restrictions ([double dagger]) 1054 1.00 1.08 3.09
Survivor Adjusted
EW Market
No Restrictions 1092 0.87 1.07 1.57
All Restrictions ([double dagger]) 1011 0.87 1.00 0.03
Survivor & Size
Adjusted EW Market
No Restrictions 1090 0.96 1.12 2.30
All Restrictions ([double dagger]) 1009 0.97 1.05 2.09
Bidder Sample
Market Benchmark t-[ratio. % < C
Robustness sup.2] ([dagger])
Restriction Set
Panel A: Market Model RAR
EW Market
No Restrictions 15.07 35.5%
All Restrictions ([double dagger]) 14.49 36.0%
Size Adjusted EW
Market
No Restrictions 14.37 35.6%
All Restrictions ([double dagger]) 13.69 36.0%
Survivor Adjusted
EW Market
No Restrictions 13.20 36.6%
All Restrictions ([double dagger]) 12.88 36.9%
Survivor & Size
Adjusted EW Market
No Restrictions 13.25 36.7%
All Restrictions ([double dagger]) 12.58 36.9%
Panel B: Market Model Beta
EW Market
No Restrictions -8.66 57.5%
All Restrictions ([double dagger]) -7.64 57.4%
Size Adjusted EW
Market
No Restrictions -2.15 50.7%
All Restrictions ([double dagger]) -1.55 49.8%
Survivor Adjusted
EW Market
No Restrictions -7.70 57.1%
All Restrictions ([double dagger]) -6.57 57.2%
Survivor & Size
Adjusted EW Market
No Restrictions -3.55 52.0%
All Restrictions ([double dagger]) -2.27 51.4%
Medians and Means of
100 Random Samples *
Market Benchmark N Median Mean t-[ratio
Robustness .sup.1]
Restriction Set
Panel A: Market Model RAR
EW Market
No Restrictions 1118 0.5% 0.6% 6.41
All Restrictions ([double dagger]) 1014 0.5% 0.6% 7.14
Size Adjusted EW
Market
No Restrictions 1118 0.5% 0.6% 6.22
All Restrictions ([double dagger]) 1016 0.5% 0.6% 7.22
Survivor Adjusted
EW Market
No Restrictions 1092 0.5% 0.5% 3.07
All Restrictions ([double dagger]) 977 0.5% 0.5% 5.67
Survivor & Size
Adjusted EW Market
No Restrictions 1089 0.5% 0.6% 2.10
All Restrictions ([double dagger]) 973 0.5% 0.5% 6.27
Panel B: Market Model Beta
EW Market
No Restrictions 1118 0.79 0.98 -0.60
All Restrictions ([double dagger]) 1014 0.79 0.95 -1.71
Size Adjusted EW
Market
No Restrictions 1118 0.90 1.03 1.26
All Restrictions ([double dagger]) 1021 0.91 1.01 0.45
Survivor Adjusted
EW Market
No Restrictions 1092 0.79 0.99 -0.24
All Restrictions ([double dagger]) 977 0.80 0.96 -1.39
Survivor & Size
Adjusted EW Market
No Restrictions 1089 0.88 1.00 -0.04
All Restrictions ([double dagger]) 978 0.89 0.98 -0.65
Medians and Means of
100 Random Samples *
Market Benchmark t-[ratio. % < C
Robustness sup.2] ([dagger])
Restriction Set
Panel A: Market Model RAR
EW Market
No Restrictions 9.95 40.5%
All Restrictions ([double dagger]) 9.20 40.6%
Size Adjusted EW
Market
No Restrictions 9.81 39.6%
All Restrictions ([double dagger]) 9.29 39.4%
Survivor Adjusted
EW Market
No Restrictions 8.63 41.8%
All Restrictions ([double dagger]) 7.87 41.9%
Survivor & Size
Adjusted EW Market
No Restrictions 8.95 40.6%
All Restrictions ([double dagger]) 8.47 40.4%
Panel B: Market Model Beta
EW Market
No Restrictions -15.49 61.0%
All Restrictions ([double dagger]) -12.73 60.9%
Size Adjusted EW
Market
No Restrictions -8.81 55.5%
All Restrictions ([double dagger]) -6.88 55.1%
Survivor Adjusted
EW Market
No Restrictions -15.08 61.0%
All Restrictions ([double dagger]) -12.37 61.0%
Survivor & Size
Adjusted EW Market
No Restrictions -10.28 56.6%
All Restrictions ([double dagger]) -8.12 56.5%
Prop'n Random Samples
[less than or equal to]
Bidder Sample
Market Benchmark N Median Mean t-[ratio
Robustness .sup.1]
Restriction Set
Panel A: Market Model RAR
EW Market
No Restrictions 100% 100% 100% 100%
All Restrictions ([double dagger]) 100% 100% 100% 100%
Size Adjusted EW
Market
No Restrictions 100% 100% 100% 100%
All Restrictions ([double dagger]) 100% 100% 100% 100%
Survivor Adjusted
EW Market
No Restrictions 100% 100% 92% 47%
All Restrictions ([double dagger]) 100% 100% 100% 100%
Survivor & Size
Adjusted EW Market
No Restrictions 100% 100% 66% 71%
All Restrictions ([double dagger]) 100% 100% 100% 100%
Panel B: Market Model Beta
EW Market
No Restrictions 100% 99% 83% 84%
All Restrictions ([double dagger]) 100% 99% 81% 80%
Size Adjusted EW
Market
No Restrictions 100% 100% 96% 98%
All Restrictions ([double dagger]) 100% 100% 99% 99%
Survivor Adjusted
EW Market
No Restrictions 100% 99% 94% 97%
All Restrictions ([double dagger]) 100% 98% 94% 94%
Survivor & Size
Adjusted EW Market
No Restrictions 100% 99% 96% 100%
All Restrictions ([double dagger]) 100% 99% 99% 99%
Prop'n Random Samples
[less than or equal to]
Bidder Sample
Market Benchmark t-[ratio. % < C
Robustness sup.2] ([dagger])
Restriction Set
Panel A: Market Model RAR
EW Market
No Restrictions 100% 0.0%
All Restrictions ([double dagger]) 100% 0.0%
Size Adjusted EW
Market
No Restrictions 100% 0.0%
All Restrictions ([double dagger]) 100% 1.0%
Survivor Adjusted
EW Market
No Restrictions 100% 0.0%
All Restrictions ([double dagger]) 100% 0.0%
Survivor & Size
Adjusted EW Market
No Restrictions 100% 0.0%
All Restrictions ([double dagger]) 100% 1.0%
Panel B: Market Model Beta
EW Market
No Restrictions 100% 2.0%
All Restrictions ([double dagger]) 99% 2.0%
Size Adjusted EW
Market
No Restrictions 100% 0.0%
All Restrictions ([double dagger]) 100% 0.0%
Survivor Adjusted
EW Market
No Restrictions 100% 1.0%
All Restrictions ([double dagger]) 99% 2.0%
Survivor & Size
Adjusted EW Market
No Restrictions 100% 0.0%
All Restrictions ([double dagger]) 100% 1.0%
Note: ([dagger]) % < C shows the percentage of the sample with values
less than C. C = 0 for RARs and C = 1 for Betas.
* N, Median and % < C are all median values across random samples.
Mean, t-[ratio.sup.1] and t-[ratio.sup.2] are all random samples
grand means.
[t-ratio.sup.1] is formed by estimating the cross-portfolio variance
from separate company parameter estimates.
[t-ratio.sup.2] is formed from individual company t-ratios for each
parameter estimate. It assumes stationary serial return series and
independent company returns.
t-[ratio.sup.1] and t-[ratio.sup.2] test the null hypotheses:
[alpha] = 0 for RARs and [beta] = 1 for Betas
([double dagger]) All Restrictions indicates [greater than or equal to]
5 return observations in the pre- and post-offer estimation periods +
systematic risk > 1% + estimates trimmed at 3 standard deviations from
the mean.
Table 1b
Pre-Offer Return Characteristics of Surviving Bidders Making Offers
Between January 1976 and June 1995
Bidder Sample
Market Benchmark N Median Mean t-
Robustness [ratio.
Restriction Set sup.1]
Panel A: (0,1) Market Model CARS
EW Market
No Restrictions 1118 19.9% 24.3% 11.53
All Restrictions ([double dagger]) 1052 18.9% 23.5% 11.78
Size Adjusted EW Market
No Restrictions 1118 17.9% 23.2% 11.44
All Restrictions ([double dagger]) 1049 17.6% 23.2% 12.31
Survivor Adjusted EW Market
No Restrictions 1092 17.6% 21.8% 10.30
All Restrictions ([double dagger]) 1012 16.9% 21.7% 10.69
Survivor & Size Adjusted EW
Market
No Restrictions 1090 15.5% 21.4% 10.41
All Restrictions ([double dagger]) 1004 16.1% 22.5% 11.60
Medians and Means of
100 Random Samples *
Market Benchmark % < 0 N Median
Robustness
Restriction Set
Panel A: (0,1) Market Model CARS
EW Market
No Restrictions 37.7% 1118 9.9%
All Restrictions ([double dagger]) 37.9% 1019 10.0%
Size Adjusted EW Market
No Restrictions 35.3% 1118 10.5%
All Restrictions ([double dagger]) 35.4% 1020 10.6%
Survivor Adjusted EW Market
No Restrictions 38.8% 1092 7.9%
All Restrictions ([double dagger]) 39.1% 980 8.3%
Survivor & Size Adjusted EW
Market
No Restrictions 36.5% 1089 9.1%
All Restrictions ([double dagger]) 36.2% 977 9.5%
Medians and Means of
100 Random Samples *
Market Benchmark Mean t-[ratio. % < 0
Robustness sup.1]
Restriction Set
Panel A: (0,1) Market Model CARS
EW Market
No Restrictions 14.6% 7.32 42.0%
All Restrictions ([double dagger]) 14.8% 7.73 41.9%
Size Adjusted EW Market
No Restrictions 14.2% 7.56 40.2%
All Restrictions ([double dagger]) 14.3% 8.10 40.0%
Survivor Adjusted EW Market
No Restrictions 12.2% 6.09 43.6%
All Restrictions ([double dagger]) 12.6% 6.50 43.3%
Survivor & Size Adjusted EW
Market
No Restrictions 12.4% 6.44 41.3%
All Restrictions ([double dagger]) 12.8% 7.00 41.0%
Prop'n Random Samples
[less than or equal to]
Bidder Sample
Market Benchmark N Median Mean
Robustness
Restriction Set
Panel A: (0,1) Market Model CARS
EW Market
No Restrictions 100% 100% 100%
All Restrictions ([double dagger]) 100% 100% 100%
Size Adjusted EW Market
No Restrictions 100% 100% 100%
All Restrictions ([double dagger]) 100% 100% 100%
Survivor Adjusted EW Market
No Restrictions 100% 100% 100%
All Restrictions ([double dagger]) 100% 100% 100%
Survivor & Size Adjusted EW
Market
No Restrictions 100% 100% 100%
All Restrictions ([double dagger]) 100% 100% 100%
Prop'n Random
Samples
[less than or
equal to]
Bidder Sample
Market Benchmark t-[ratio. % < 0
Robustness sup.1]
Restriction Set
Panel A: (0,1) Market Model CARS
EW Market
No Restrictions 100% 0.0%
All Restrictions ([double dagger]) 100% 0.0%
Size Adjusted EW Market
No Restrictions 100% 0.0%
All Restrictions ([double dagger]) 100% 0.0%
Survivor Adjusted EW Market
No Restrictions 100% 0.0%
All Restrictions ([double dagger]) 100% 0.0%
Survivor & Size Adjusted EW
Market
No Restrictions 100% 0.0%
All Restrictions ([double dagger]) 100% 0.0%
Note: * N, Median and % < 0 are all median values across random
samples. Mean, t-[ratio.sup.1] and t-[ratio.sup.2] are all random
samples grand means.
t-[ratio.sup.1] is formed by estimating the cross-portfolio variance
from separate company parameter estimates.
t-[ratio.sup.1] tests the null hypothesis: [alpha] = 0 for CARs
([double dagger]) All Restrictions indicates [greater than or equal to]
5 return observations in the pre- and post-offer estimation periods +
systematic risk > 1% + estimates trimmed at 3 standard deviations from
the mean.
Table 2
Pre-offer Return Characteristic Estimates by Offer Order
for Surviving Bidders Making Offers Between
January 1976 and June 1995
Bidder Sample
Model Parameter
Sample Partition N Median Mean
Pre-Offer Risk Adjusted Returns
First-time 377 0.5% 0.7%
Apprentice 351 0.6% 0.9%
Experienced 277 0.9% 1.1%
First-time -- Apprentice
First-time -- Experienced
Apprentice -- Experienced
Pre-Offer (0,1) Market Model ARs
First-time 377 9.6% 16.8%
Apprentice 351 12.6% 21.0%
Experienced 277 25.3% 28.6%
First-time -- Apprentice
First-time -- Experienced
Apprentice -- Experienced
Pre-Offer Systematic Risk
First-time 380 0.9 1.0
Apprentice 353 1.0 1.1
Experienced 278 1.0 1.1
First-time -- Apprentice
First-time -- Experienced
Apprentice -- Experienced
Bidder Sample
Model Parameter t-[ratio t-[ratio %<C
Sample Partition .sup.1] .sup.2] ([dagger])
Pre-Offer Risk Adjusted Returns
First-time 4.30 6.11 40.1%
Apprentice 6.42 6.95 37.9%
Experienced 8.37 9.20 31.4%
First-time -- Apprentice
First-time -- Experienced
Apprentice -- Experienced
Pre-Offer (0,1) Market Model ARs
First-time 4.70 41.4%
Apprentice 6.33 37.6%
Experienced 9.58 28.2%
First-time -- Apprentice
First-time -- Experienced
Apprentice -- Experienced
Pre-Offer Systematic Risk
First-time 0.62 -2.97 54.5%
Apprentice 1.21 -1.99 52.4%
Experienced 2.91 1.41 45.7%
First-time -- Apprentice
First-time -- Experienced
Apprentice -- Experienced
Medians and Means of
100 Random Samples *
Model Parameter N Median Mean
Sample Partition
Pre-Offer Risk Adjusted Returns
First-time 376 0.4% 0.3%
Apprentice 348 0.4% 0.5%
Experienced 279 0.6% 0.5%
First-time -- Apprentice
First-time -- Experienced
Apprentice -- Experienced
Pre-Offer (0,1) Market Model ARs
First-time 378 6.2% 7.7%
Apprentice 349 7.9% 12.2%
Experienced 280 7.8% 12.0%
First-time -- Apprentice
First-time -- Experienced
Apprentice -- Experienced
Pre-Offer Systematic Risk
First-time 378 0.8 0.9
Apprentice 349 0.8 1.0
Experienced 281 0.8 0.9
First-time -- Apprentice
First-time -- Experienced
Apprentice -- Experienced
Medians and Means of 100
Random Samples *
Model Parameter t-[ratio t-[ratio %<C
Sample Partition .sup.1] .sup.2] ([dagger])
Pre-Offer Risk Adjusted Returns
First-time 1.86 3.87 43.3%
Apprentice 3.36 5.00 40.9%
Experienced 4.41 5.72 36.9%
First-time -- Apprentice
First-time -- Experienced
Apprentice -- Experienced
Pre-Offer (0,1) Market Model ARs
First-time 2.38 45.1%
Apprentice 4.04 42.1%
Experienced 4.28 41.3%
First-time -- Apprentice
First-time -- Experienced
Apprentice -- Experienced
Pre-Offer Systematic Risk
First-time -1.13 -8.33 59.4%
Apprentice -0.86 -7.11 59.3%
Experienced -2.11 -9.56 62.4%
First-time -- Apprentice
First-time -- Experienced
Apprentice -- Experienced
Prop'n Random Samples
[less than or equal to]
Bidder Sample
Model Parameter N Median Mean
Sample Partition
Pre-Offer Risk Adjusted Returns
First-time 61% 95% 100%
Apprentice 74% 94% 100%
Experienced 38% 100% 100%
First-time -- Apprentice 49% 44%
First-time -- Experienced 28% 27%
Apprentice -- Experienced 23% 36%
Pre-Offer (0,1) Market Model ARs
First-time 42% 90% 100%
Apprentice 73% 92% 100%
Experienced 31% 100% 100%
First-time -- Apprentice 37% 54%
First-time -- Experienced 0% 5%
Apprentice -- Experienced 0% 3%
Pre-Offer Systematic Risk
First-time 73% 97% 91%
Apprentice 81% 100% 99%
Experienced 32% 100% 100%
First-time -- Apprentice 40% 43%
First-time -- Experienced 5% 5%
Apprentice -- Experienced 8% 6%
Prop'n Random Samples
[less than or equal to]
Bidder Sample
Model Parameter t-[ratio t-[ratio %<C
Sample Partition .sup.1] .sup.2] ([dagger])
Pre-Offer Risk Adjusted Returns
First-time 100% 99% 9%
Apprentice 100% 98% 15%
Experienced 100% 100% 0%
First-time -- Apprentice 43%
First-time -- Experienced 71%
Apprentice -- Experienced 79%
Pre-Offer (0,1) Market Model ARs
First-time 99% 8%
Apprentice 98% 5%
Experienced 100% 0%
First-time -- Apprentice 56%
First-time -- Experienced 100%
Apprentice -- Experienced 99%
Pre-Offer Systematic Risk
First-time 91% 100% 3%
Apprentice 99% 100% 0%
Experienced 100% 100% 0%
First-time -- Apprentice 75%
First-time -- Experienced 99%
Apprentice -- Experienced 100%
Note: ([dagger]) % < C shows the percentage of the sample with values
less than C. C = 0 for RARs and C = 1 for Betas.
* N, Median and % < C are all median values across random samples.
Mean, t-[ratio.sup.1] and t-[ratio.sup.2] are all random samples
grand means.
t-[ratio.sup.1] is formed by estimating the cross-portfolio variance
from separate company parameter estimates.
t-[ratio.sup.2] is formed from individual company t-ratios for each
parameter estimate. It assumes stationary serial return series and
independent company returns.
t-[ratio.sup.1] and t-[ratio.sup.2] test the null hypotheses:
[alpha] = 0 for RARs and [beta] = 1 for systematic risk (betas)
([double dagger]) All Restrictions indicates [greater than or equal to]
5 return observations in the pre- & post-offer estimation periods +
systematic risk > 1% + estimates trimmed at 3 standard deviations from
the mean.
Table 3
Bidder Operating Margin per Sale
Years from {-5} {-4} {-3} {-2}
Offer
RAR Range
Panel A: Complete Data Firms
High Average 11.5% 13.7% 13.6% 14.2%
0.747 0.796 0.857 0.928
Median 7.0% 8.2% 8.7% 9.1%
0.660 0.717 0.704 0.779
Obs 89 109 108 108
Mid Average 8.2% 8.5% 8.1% 9.2%
0.823 0.813 0.928 0.833
Median 6.3% 6.4% 6.3% 6.4%
0.869 0.896 0.870 0.888
Obs 134 155 153 154
Low Average 13.1% 10.0% 9.4% 10.2%
1.154 1.063 0.884 0.814
Median 8.1% 8.1% 7.7% 8.0%
0.945 0.932 0.942 0.921
Obs 62 70 70 69
Panel B. All Available Data
High Average 10.7% 9.8% 11.5% 14.0%
0.715 0.686 0.773 0.884
Median 7.3% 7.3% 8.7% 9.9%
0.677 0.685 0.733 0.853
Obs 152 184 213 245
Mid Average 8.9% 9.2% 7.7% 10.4%
0.819 0.820 0.911 0.847
Median 6.3% 6.4% 6.2% 6.9%
0.866 0.873 0.887 0.911
Obs 215 240 265 291
Low Average -2.6% 10.3% 12.9% 10.3%
1.099 0.990 0.792 0.705
Median 7.4% 7.5% 7.3% 7.6%
0.946 0.914 0.927 0.889
Obs 91 101 115 122
Years from {-1} {0} {1} {2}
Offer
RAR Range
Panel A: Complete Data Firms
High Average 15.2% 17.1% 15.5% 15.0%
0.930 1 0.945 1.000
Median 10.7% 13.2% 12.2% 12.4%
0.898 1 0.913 0.855
Obs 108 108 107 108
Mid Average 9.8% 10.3% 11.1% 9.7%
0.868 1 0.928 0.807
Median 6.5% 7.3% 7.5% 6.8%
0.944 1 0.992 0.972
Obs 152 152 156 153
Low Average 10.4% 10.0% 12.6% 8.1%
0.924 1 0.986 0.855
Median 8.6% 8.2% 7.6% 7.2%
0.946 1 0.972 0.895
Obs 68 68 70 69
Panel B. All Available Data
High Average 11.1% 21.4% 6.5% 1.0%
1.071 1 0.782 0.612
Median 12.2% 10.8% 10.8% 9.5%
0.952 0.881 0.881 0.814
Obs 273 289 282 272
Mid Average 9.4% 8.5% 4.6% -5.4%
0.913 1 0.862 0.787
Median 7.4% 7.8% 7.6% 6.9%
0.950 1 0.933 0.938
Obs 309 311 316 293
Low Average 12.5% 10.7% 6.7% 0.0%
0.800 1 0.761 0.742
Median 7.3% 8.3% 7.3% 6.2%
0.943 1 0.910 0.846
Obs 123 120 115 113
Years from {3} {4} {5}
Offer
RAR Range
Panel A: Complete Data Firms
High Average 14.2% 11.2% 11.2%
1.030 0.847 1.081
Median 11.0% 8.7% 9.8%
0.839 0.849 0.953
Obs 108 108 93
Mid Average 8.3% 6.2% 7.1%
0.836 0.737 0.780
Median 6.8% 6.4% 6.1%
0.965 0.909 0.846
Obs 151 153 134
Low Average 8.2% 9.7% 5.5%
0.913 0.876 0.416
Median 6.4% 7.1% 6.1%
0.908 0.943 0.710
Obs 68 69 65
Panel B. All Available Data
High Average 1.2% 12.7% 4.0%
0.540 0.917 0.768
Median 8.5% 9.5% 9.3%
0.694 0.849 0.926
Obs 246 214 191
Mid Average 3.2% 2.0% 6.9%
0.833 0.710 0.767
Median 6.6% 6.6% 6.4%
0.884 0.869 0.844
Obs 267 249 214
Low Average 8.9% 7.4% 6.7%
0.663 0.832 0.618
Median 6.4% 6.8% 6.1%
0.787 0.862 0.710
Obs 106 94 87
Table 4
Bidder Residual Cash Flow per Adjusted Share
Years from {-5} {-4} {-3} {-2}
Offer
RAR Range
Panel A. Complete Data Firms
High Average $0.082 $0.118 $0.164 $0.230
0.153 0.234 0.308 0.461
Median $0.078 $0.122 $0.153 $0.195
0.154 0.241 0.316 0.395
Obs 128 139 141 139
Mid Average $0.127 $0.153 $0.200 $0.220
0.263 0.373 0.562 0.573
Median $0.105 $0.128 $0.161 $0.196
0.313 0.438 0.541 0.641
Obs 141 160 160 160
Low Average $0.133 $0.143 $0.166 $0.185
0.554 0.330 0.684 0.459
Median $0.110 $0.116 $0.166 $0.177
0.403 0.397 0.506 0.613
Obs 55 68 69 68
Panel B. All Available Data
High Average $0.083 $0.093 $0.135 $0.173
0.182 0.227 0.256 0.378
Median $0.077 $0.086 $0.125 $0.149
0.173 0.238 0.316 0.400
Obs 184 216 250 281
Mid Average $0.126 $0.136 $0.166 $0.176
0.431 0.429 0.666 0.683
Median $0.099 $0.118 $0.134 $0.154
0.359 0.468 0.534 0.647
Obs 213 247 279 302
Low Average $0.109 $0.110 $0.137 $0.124
0.520 0.433 0.376 0.552
Median $0.096 $0.104 $0.118 $0.120
0.389 0.386 0.498 0.607
Obs 77 91 111 128
Years from {-1} {0} {1} {2}
Offer
RAR Range
Panel A. Complete Data Firms
High Average $0.356 $0.482 $0.519 $0.624
0.721 1 1.158 0.921
Median $0.302 $0.391 $0.485 $0.561
0.714 1 1.138 1.075
Obs 141 137 139 139
Mid Average $0.241 $0.322 $0.375 $0.399
0.700 1 1.103 1.025
Median $0.223 $0.289 $0.322 $0.354
0.806 1 1.156 1.242
Obs 158 159 158 159
Low Average $0.205 $0.263 $0.298 $0.273
1.122 1 1.023 0.872
Median $0.198 $0.289 $0.271 $0.224
0.749 1 1.178 1.115
Obs 68 69 69 69
Panel B. All Available Data
High Average $0.275 $0.318 $0.229 $0.215
0.706 1 0.803 0.555
Median $0.242 $0.271 $0.240 $0.190
0.709 1 1.024 0.988
Obs 314 324 319 294
Mid Average $0.189 $0.242 $0.272 $0.305
0.732 1 0.863 0.972
Median $0.158 $0.205 $0.221 $0.236
0.782 1 1.121 1.230
Obs 327 340 343 317
Low Average $0.130 $0.159 $0.195 $0.183
0.950 1 1.028 0.303
Median $0.114 $0.147 $0.149 $0.148
0.723 1 1.139 0.982
Obs 133 137 132 126
Years from {3} {4} {5}
Offer
RAR Range
Panel A. Complete Data Firms
High Average $0.778 $0.886 $1.177
1.584 1.867 1.926
Median $0.515 $0.447 $0.624
1.324 1.335 1.513
Obs 140 139 122
Mid Average $0.484 $0.409 $0.465
1.072 1.187 1.044
Median $0.353 $0.324 $0.387
1.429 1.504 1.695
Obs 158 159 140
Low Average $0.278 $0.334 $0.288
0.957 1.118 1.564
Median $0.288 $0.328 $0.265
0.958 1.041 1.307
Obs 68 68 63
Panel B. All Available Data
High Average $0.427 $0.606 $0.775
0.877 1.690 2.057
Median $0.229 $0.285 $0.475
0.891 1.254 1.513
Obs 264 227 191
Mid Average $0.342 $0.382 $0.439
1.433 1.741 1.427
Median $0.271 $0.259 $0.357
1.332 1.454 1.634
Obs 287 260 225
Low Average $0.247 $0.326 $0.291
1.420 1.150 1.201
Median $0.207 $0.221 $0.168
0.958 1.080 1.233
Obs 114 101 94
Table 5
Annual Growth Rates for Relative Residual Cash Flows of
Complete-Data Firms by RAR Type
RAR Type {-4} to {-2) {-2} to {-1} {-1} to {+1}
Panel A. Non-Takeover Control Firms
High-range 22.7% 45.8% 14.5%
Mid-range 20.2% 21.3% 19.1%
Low-range 20.3% 0.4% 15.2%
Kruskal-Wallis p-value 0.19 0.00 0.04
Panel B. Bidder Firms
High-range 28.0% 80.8% 26.3%
P(Controls [less than or 0.8 1.0 1.0
equal to] Bidders)
Mid-range 20.9% 25.7% 19.8%
P(Controls [less than or 0.6 0.9 0.7
equal to] Bidders)
Low-range 24.3% 22.3% 25.4%
P(Controls [less than or 0.7 0.9 0.9
equal to] Bidders)
RAR Type {+1} to {+4} {-1} to {+4}
Panel A. Non-Takeover
Control Firms
High-range 10.5% 11.3%
Mid-range 10.0% 13.4%
Low-range 8.4% 11.8%
Kruskal-Wallis p-value 0.74 0.03
Panel B. Bidder Firms
High-range 5.5% 13.3%
P(Controls [less than or 0.3 0.8
equal to] Bidders)
Mid-range 9.1% 13.3%
P(Controls [less than or 0.2 0.5
equal to] Bidders)
Low-range -4.0% 6.8%
P(Controls [less than or 0.0 0.0
equal to] Bidders)
Table 6
Proportional Distribution of Bidder Pre-Offer RAR Estimates (%)
RAR Decile Group 1 2 3 4 5
Pre-Offer Prop (a) 2.6 7.9 8.7 10.7 7.9
Deviation from -7.4 -2.1 -1.3 0.7 -2.1
Expectations
RAR Decile Group 6 7 8 9 10
Pre-Offer Prop (a) 9.8 10.4 13.4 13.7 15.0
Deviation from -0.2 0.4 3.4 3.7 5.0
Expectations
Table 7
CARS [-3, +3] for Surviving Bidders Making Offer Between January 1976
and June 1995 (for Survivor & Size Adjusted EW Market): Bootstrap
Samples Matched with Bidders on Size/Survival/Non-Takeover-Participant
Status
Bidder Sample
Plus Extra Controls for
Restriction Set N Median Mean
No Extra Controls
No Restrictions 1090 -1.7% -0.1%
At least 5 obs pre & post 1025 -1.8% -2.8%
+ Systematic Risk > 1% 1018 -1.8% -2.8%
+ Trimmed at 3 Std Devs 998 -1.7% 1.6%
Pre-offer RAR and Beta
No Restrictions 1090 -1.7% -0.1%
At least 5 obs pre & post 1025 -1.8% -2.8%
+ Systematic Risk > 1% 1018 -1.8% -2.8%
+ Trimmed at 3 Std Devs 998 -1.7% -1.6%
Pre-offer RAR
No Restrictions 1090 -1.7% -0.1%
At least 5 obs pre & post 1025 -1.8% -2.8%
+ Systematic Risk > 1% 1018 1.8% -2.8%
+ Trimmed at 3 Std Devs 998 1.7% -1.6%
Pre-offer Beta
No Restrictions 1090 -1.7% -0.1%
At least 5 obs pre & post 1025 -1.8% -2.8%
+ Systematic Risk > 1% 1018 -1.8% -2.8%
+ Trimmed at 3 Std Devs 998 -1.7% -1.6%
Bidder Sample
Plus Extra Controls for [t-ratio. [t-ratio.
Restriction Set sup.1] sup.2] % < 0%
No Extra Controls
No Restrictions -0.04 -1.25 52.0%
At least 5 obs pre & post -2.02 -1.05 52.3%
+ Systematic Risk > 1% -2.02 -1.09 52.5%
+ Trimmed at 3 Std Devs -1.32 -1.28 52.2%
Pre-offer RAR and Beta
No Restrictions -0.04 -1.25 52.0%
At least 5 obs pre & post -2.02 -1.05 52.3%
+ Systematic Risk > 1% -2.02 -1.09 52.5%
+ Trimmed at 3 Std Devs -1.32 -1.28 52.2%
Pre-offer RAR
No Restrictions -0.04 -1.25 52.0%
At least 5 obs pre & post -2.02 -1.05 52.3%
+ Systematic Risk > 1% -2.02 -1.09 52.5%
+ Trimmed at 3 Std Devs -1.32 -1.28 52.2%
Pre-offer Beta
No Restrictions -0.04 -1.25 52.0%
At least 5 obs pre & post -2.02 -1.05 52.3%
+ Systematic Risk > 1% -2.02 -1.09 52.5%
+ Trimmed at 3 Std Devs -1.32 -1.28 52.2%
Medians and Means
of Random Samples *
Plus Extra Controls for
Restriction Set N Median Mean
No Extra Controls
No Restrictions 1089 -1.6% -2.8%
At least 5 obs pre & post 1000 -2.0% -2.6%
+ Systematic Risk > 1% 991 -2.1% -2.7%
+ Trimmed at 3 Std Devs 971 -1.9% -2.3%
Pre-offer RAR and Beta
No Restrictions 1089 -4.8% -5.3%
At least 5 obs pre & post 1073 -4.7% -6.0%
+ Systematic Risk > 1% 1068 -4.8% -6.0%
+ Trimmed at 3 Std Devs 1049 -4.6% -5.2%
Pre-offer RAR
No Restrictions 1090 -6.2% -8.6%
At least 5 obs pre & post 1074 -6.3% -8.2%
+ Systematic Risk > 1% 1069 -6.3% -8.2%
+ Trimmed at 3 Std Devs 1050 -6.2% -7.7%
Pre-offer Beta
No Restrictions 1089 -2.4% -1.8%
At least 5 obs pre & post 1073 -2.4% -3.0%
+ Systematic Risk > 1% 1069 -2.5% -3.0%
+ Trimmed at 3 Std Devs 1050 -2.5% -3.0%
Medians and Means
of Random Samples *
Plus Extra Controls for [t-ratio. [t-ratio.
Restriction Set sup.1] sup.2] % < 0%
No Extra Controls
No Restrictions -0.95 -1.17 52.2%
At least 5 obs pre & post -1.85 -1.77 52.9%
+ Systematic Risk > 1% -1.87 -1.79 53.1%
+ Trimmed at 3 Std Devs -1.90 -1.97 53.0%
Pre-offer RAR and Beta
No Restrictions -2.89 -5.26 56.0%
At least 5 obs pre & post -4.37 -5.27 56.2%
+ Systematic Risk > 1% -4.35 -5.26 56.1%
+ Trimmed at 3 Std Devs -4.45 -5.4 56.0%
Pre-offer RAR
No Restrictions -3.93 -6.71 57.9%
At least 5 obs pre & post -5.82 -6.81 58.1%
+ Systematic Risk > 1% -5.84 -6.83 58.2%
+ Trimmed at 3 Std Devs -6.50 -7.01 58.2%
Pre-offer Beta
No Restrictions -0.90 -2.28 53.2%
At least 5 obs pre & post -2.09 -2.40 53.2%
+ Systematic Risk > 1% -2.10 -2.42 53.2%
+ Trimmed at 3 Std Devs -2.48 -2.62 53.4%
Prop'n Random Samples
[less than or equal
to] Bidder Sample
Plus Extra Controls for
Restriction Set N Median Mean
No Extra Controls
No Restrictions 100% 50% 82%
At least 5 obs pre & post 100% 60% 47%
+ Systematic Risk > 1% 100% 59% 48%
+ Trimmed at 3 Std Devs 100% 63% 71%
Pre-offer RAR and Beta
No Restrictions 100% 100% 97%
At least 5 obs pre & post 0% 100% 100%
+ Systematic Risk > 1% 0% 100% 100%
+ Trimmed at 3 Std Devs 0% 100% 100%
Pre-offer RAR
No Restrictions 100% 100% 100%
At least 5 obs pre & post 0% 100% 100%
+ Systematic Risk > 1% 0% 100% 100%
+ Trimmed at 3 Std Devs 0% 100% 100%
Pre-offer Beta
No Restrictions 100% 77% 80%
At least 5 obs pre & post 0% 73% 50%
+ Systematic Risk > 1% 0% 73% 57%
+ Trimmed at 3 Std Devs 0% 73% 90%
Prop'n Random Samples
[less than or equal
to] Bidder Sample
Plus Extra Controls for [t-ratio. [t-ratio.
Restriction Set sup.1] sup.2] % < 0%
No Extra Controls
No Restrictions 82% 49% 46%
At least 5 obs pre & post 42% 75% 34%
+ Systematic Risk > 1% 42% 75% 36%
+ Trimmed at 3 Std Devs 69% 77% 31%
Pre-offer RAR and Beta
No Restrictions 97% 100% 0%
At least 5 obs pre & post 100% 100% 0%
+ Systematic Risk > 1% 100% 100% 0%
+ Trimmed at 3 Std Devs 100% 100% 0%
Pre-offer RAR
No Restrictions 100% 100% 0%
At least 5 obs pre & post 100% 100% 0%
+ Systematic Risk > 1% 100% 100% 0%
+ Trimmed at 3 Std Devs 100% 100% 0%
Pre-offer Beta
No Restrictions 80% 87% 23%
At least 5 obs pre & post 50% 87% 23%
+ Systematic Risk > 1% 50% 87% 27%
+ Trimmed at 3 Std Devs 83% 87% 23%
Note: * 100 bootstrap samples were selected for No Extra Controls and
Pre-offer RAR and Beta. 30 samples were chosen for remaining controls.
N, Median and % < 0 are all median values across random samples. Mean,
[t-ratio.sup.1] and [t-ratio.sup.2] are all random samples grand
means. [t-ratio.sup.1] is formed by estimating the cross-portfolio
variance from separate company abnormal return estimates.
[t-ratio.sup.2] is formed from individual company t-ratios for each
abnormal return estimate. It assumes stationary serial return series
and independent company returns.
Table 8
(0,1) Market Model CARs [-3,+3] for Surviving Bidders Making Offers
between January 1976 and June 1995 (for Survivor & Size Adjusted EW
Market): Bootstrap Samples Matched with Bidders on Size/Survival/
Non-Takeover-Participant Status
Bidder Sample
Plus Extra Controls for
Restriction Set N Median Mean
No Extra Controls
No Restrictions 1090 3.7% 3.1%
At least 5 obs pre & post 1025 3.6% 3.3%
+ Systematic Risk > 1% 1018 3.4% 3.2%
+ Trimmed at 3 Std Devs 1003 3.4% 3.5%
Pre-offer RAR and Beta
No Restrictions 1090 3.7% 3.1%
At least 5 obs pre & post 1025 3.6% 3.3%
+ Systematic Risk > 1% 1018 3.4% 3.2%
+ Trimmed at 3 Std Devs 1003 3.4% 3.5%
Pre-offer RAR
No Restrictions 1090 3.7% 3.1%
At least 5 obs pre & post 1025 3.6% 3.3%
+ Systematic Risk > 1% 1018 3.4% 3.2%
+ Trimmed at 3 Std Devs 1003 3.4% 3.5%
Pre-offer Beta
No Restrictions 1090 3.7% 3.1%
At least 5 obs pre & post 1025 3.6% 3.3%
+ Systematic Risk > 1% 1018 3.4% 3.2%
+ Trimmed at 3 Std Devs 1003 3.4% 3.5%
Bidder Sample
Plus Extra Controls for [t-ratio. [t-ratio.
Restriction Set sup.1] sup.2] % < 0%
No Extra Controls
No Restrictions 2.94 6.10 44.2%
At least 5 obs pre & post 3.08 6.17 44.7%
+ Systematic Risk > 1% 3.02 6.07 44.8%
+ Trimmed at 3 Std Devs 3.67 5.87 44.7%
Pre-offer RAR and Beta
No Restrictions 2.94 6.10 44.2%
At least 5 obs pre & post 3.08 6.17 44.7%
+ Systematic Risk > 1% 3.02 6.07 44.8%
+ Trimmed at 3 Std Devs 3.67 5.87 44.7%
Pre-offer RAR
No Restrictions 2.94 6.10 44.2%
At least 5 obs pre & post 3.08 6.17 44.7%
+ Systematic Risk > 1% 3.02 6.07 44.8%
+ Trimmed at 3 Std Devs 3.67 5.87 44.7%
Pre-offer Beta
No Restrictions 2.94 6.10 44.2%
At least 5 obs pre & post 3.08 6.17 44.7%
+ Systematic Risk > 1% 3.02 6.07 44.8%
+ Trimmed at 3 Std Devs 3.67 5.87 44.7%
Medians and Means
of Random Samples *
Plus Extra Controls for
Restriction Set N Median Mean
No Extra Controls
No Restrictions 1089 2.4% 1.5%
At least 5 obs pre & post 1000 2.0% 1.2%
+ Systematic Risk > 1% 991 1.9% 1.2%
+ Trimmed at 3 Std Devs 974 2.0% 1.7%
Pre-offer RAR and Beta
No Restrictions 1089 1.6% 0.2%
At least 5 obs pre & post 1073 1.6% 0.2%
+ Systematic Risk > 1% 1068 1.6% 0.2%
+ Trimmed at 3 Std Devs 1050 1.8% 1.4%
Pre-offer RAR
No Restrictions 1090 1.1% 0.0%
At least 5 obs pre & post 1074 1.1% 0.1%
+ Systematic Risk > 1% 1069 1.1% 0.0%
+ Trimmed at 3 Std Devs 1051 1.2% 0.8%
Pre-offer Beta
No Restrictions 1089 1.6% 0.2%
At least 5 obs pre & post 1073 1.6% 0.2%
+ Systematic Risk > 1% 1069 1.6% 0.2%
+ Trimmed at 3 Std Devs 1052 1.8% 1.1%
Medians and Means of
Random Samples *
Plus Extra Controls for [t-ratio. [t-ratio.
Restriction Set sup.1] sup.2] % < 0%
No Extra Controls
No Restrictions 1.39 4.93 46.0%
At least 5 obs pre & post 1.11 3.97 46.5%
+ Systematic Risk > 1% 1.04 3.91 46.6%
+ Trimmed at 3 Std Devs 1.75 3.48 46.4%
Pre-offer RAR and Beta
No Restrictions 0.15 3.59 47.0%
At least 5 obs pre & post 0.19 3.57 47.1%
+ Systematic Risk > 1% 0.18 3.57 47.0%
+ Trimmed at 3 Std Devs 1.46 3.10 46.7%
Pre-offer RAR
No Restrictions 0.03 3.17 48.1%
At least 5 obs pre & post 0.06 3.26 48.2%
+ Systematic Risk > 1% 0.01 3.21 48.3%
+ Trimmed at 3 Std Devs 0.80 2.70 48.0%
Pre-offer Beta
No Restrictions 0.17 3.46 47.3%
At least 5 obs pre & post 0.23 3.45 47.3%
+ Systematic Risk > 1% 0.18 3.40 47.3%
+ Trimmed at 3 Std Devs 1.19 2.99 47.1%
Prop'n Random Samples
[less than or equal
to] Bidder Sample
Plus Extra Controls for
Restriction Set N Median Mean
No Extra Controls
No Restrictions 100% 89% 90%
At least 5 obs pre & post 100% 95% 97%
+ Systematic Risk > 1% 100% 94% 97%
+ Trimmed at 3 Std Devs 100% 93% 97%
Pre-offer RAR and Beta
No Restrictions 100% 99% 100%
At least 5 obs pre & post 0% 98% 100%
+ Systematic Risk > 1% 0% 97% 100%
+ Trimmed at 3 Std Devs 0% 97% 98%
Pre-offer RAR
No Restrictions 100% 100% 100%
At least 5 obs pre & post 0% 100% 100%
+ Systematic Risk > 1% 0% 100% 100%
+ Trimmed at 3 Std Devs 0% 100% 100%
Pre-offer Beta
No Restrictions 100% 97% 100%
At least 5 obs pre & post 0% 97% 100%
+ Systematic Risk > 1% 0% 97% 100%
+ Trimmed at 3 Std Devs 0% 97% 100%
Prop'n Random Samples
[less than or equal to]
Bidder Sample
Plus Extra Controls for [t-ratio. [t-ratio.
Restriction Set sup.1] sup.2] % < 0%
No Extra Controls
No Restrictions 91% 85% 12%
At least 5 obs pre & post 98% 99% 11%
+ Systematic Risk > 1% 97% 99% 12%
+ Trimmed at 3 Std Devs 98% 99% 13%
Pre-offer RAR and Beta
No Restrictions 100% 99% 2%
At least 5 obs pre & post 100% 99% 6%
+ Systematic Risk > 1% 100% 99% 6%
+ Trimmed at 3 Std Devs 99% 99% 7%
Pre-offer RAR
No Restrictions 100% 100% 0%
At least 5 obs pre & post 100% 100% 0%
+ Systematic Risk > 1% 100% 100% 0%
+ Trimmed at 3 Std Devs 100% 100% 3%
Pre-offer Beta
No Restrictions 100% 97% 3%
At least 5 obs pre & post 100% 100% 3%
+ Systematic Risk > 1% 100% 100% 3%
+ Trimmed at 3 Std Devs 100% 100% 3%
Note: * N, Median and % < 0 are all median values across random
samples. Mean, [t-ration.sup.1] and [t-ratio.sup.2] are all random
samples grand means. [t-ratio.sup.1] is formed by estimating the
cross-portfolio variance from separate company abnormal return
estimates. [t-ration.sup.2] is formed from individual t-ratios for
each abnormal return estimate. It assumes stationary serial return
series and independent company returns.
The author wishes to thank Tom Smith for comments on an earlier version and special thanks to Peter Swan for encouragement and support. (1.) For example, see Dodd (1976), Asquith (1983), Walter (1984), Jensen and Ruback (1983), Bishop, Dodd and Officer (1987), and more recently, Brown and da Silva Rosa (1998). (2.) Franks and Harris (1989), p. 247. (3.) Eckbo, Maksimovic and Williams (1990), Acharya (1993) and Prabhala (1997) develop models recognising the private information revealed at takeover offer announcements when investors condition on known objective functions of bidder managers. A different aspect on which offers are conditional is the timing of the offer itself. Managers can be expected to announce offers at times most suitable for their objectives. This timing aspect has implications for the choice between the two normal-return models considered here. (4.) Kendig (1998) identifies takeovers from the reasons given for companies delisting from the Australian Stock Exchanges. In contrast, this paper measures takeover activity according to offer announcements. Kendig's analysis is restricted to acquisitions leading to sufficient ownership transfers of target firm shares that they cease to qualify for listing, whereas our measure includes all takeover activity. (5.) Nelson (1959), Melicher, Ledolter and D'Antonio (1983), Bishop, Dodd and Officer (1987). (6) Kothari and Warner (1997) also find market-model CARs excessively reject the zero null hypothesis in favour of a negative alternative when sample firms are selected with low book-to-market ratios at the start of randomly identified event years. Kothari and Warner attribute the negative-CAR outcome, in part, to unusually good performance by low book-to-market firms during the estimation period, leading to systematically positive market-model alphas (p. 329). (7) A different possibility to note is that bidders may also consistently release news on other aspects of existing businesses when announcing offers. If credible news with positive cash-flow consequences is more common at offer announcements than at other times, inferences on acquisition consequences based on abnormal returns will be positively biased. For example, on the day Wesfarmers Limited announced a hostile bid for Howard Smith Limited, 13th June 2001, it 'surprised on the upside ... by telling the market that it was about to report a June 30 profit rise of 27% and tbrecasting an increase of 31% for 2002.' (Knight 2001a). Abnormal returns for both the market and (0,1) market models are affected. Although Casey (1987) finds favourable information releases are a common takeover defence, our analysis of bidder cash flows (later in the paper) finds no evidence of consistent alterations in reported cash flows near offer announcements. If positive bidder profit forecast revisions are more common near offers, our evidence from subsequent annual reports indicates such forecasts lack credibility. (8.) Using Monte Carlo simulations Kothari and Warner (1997) and Barber and Lyon (1997) also find the (0,1) market model is often a superior benchmarking procedure to the market model. (9.) The identification of listed bidders required complete disclosure of parent company names in ASX documents containing offer details. Proprietary companies are frequently used by bidders of all kinds as vehicles for acquisition. Those not directly declared in source documents as subsidiaries of listed parent companies are misclassified here. (10.) Kendig (1998) finds waves in Australian takeovers and share prices are related to each other. Kendig measures takeovers by the date at which acquired targets are delisted (p. 101). Delisting typically follows an offer by 4 months (Kendig footnote 22, p. 101). Kendig measures share prices by quarterly values of the Australian All Ordinaries Index. Waves in share prices are found to lead takeovers by 3 months but not be significantly related at other leads or lags (table 5.1, p. 148). Kendig does not consider the possibility that index prices for the quarter are affected by investor reactions to offers, four months before the delist date. Our analysis is unaffected by the interaction of offer reactions and offer activity. (11.) Another possibility is that the appeal of domestic firms to foreign acquirers varies with exchange rates. A comparison of bidding trends for all bidders with those for just Australian listed bidders shows Australian listed firms generally account for around 60% of all offers, every month, across all three decades. Constant participation by domestic bidders weakens the foreign attractiveness proposition because domestic enthusiasm increases with surges in overall takeover activity. If surges are related to lower exchange rates, greater foreign competition should increase costs for domestic bidders and reduce their participation. In fact, domestic bidders are more active when overall activity increases. (12.) Typical pre-offer and post-close periods were found by measuring the average time between offer-related events and offer announcement dates. The pre-offer period ends at month [ 11] because, on average, bidders making pre-offer substantial shareholder declarations were found to do so seven months before a takeover offer. More than 400 offers between 1985 and 1990 were examined in archived ASX records at Sydney University to identify dates when substantial shareholder announcements were first made. Our estimate smaller than that implied by Figure 6 of Bishop (1991), which applies only to notices preceding acquired targets, and longer than that reported for a smaller sample of offers by Casey, Dodd and Dolan (1987) (p. 209). The pre-offer estimation period was pushed back a further three months to protect market model parameter estimates from takeover related price movements. This left [-10] as the earliest event period month. [FIGURE 6 OMITTED] The post-offer estimation period commences at month [+8] because, on average, the final closing date for offers proceeding to formal takeovers, comes four months after the first offer (which agrees with Kendig (1998). The post-offer period was defined to commence three months after the typical closing date, again, to avoid takeover related effects on bidder prices. This meant month [+7], relative to the offer month, [0], was the latest in the event period. The entire event period is [-10,+7]. The choice of twenty four months for both pre-offer and post-offer estimation periods was arbitrary but considered to provide a reasonable compromise between parameter estimate precision, for stationary return companies, and self-selection sample estimation biases between long- and short-lived companies. Forty eight month estimation periods are also accepted by commercial users of market model parameters such as those supplied by the Centre for Research in Finance, Australian Graduate School of Management. (13.) A separate collection of 100 bootstrap samples was also formed after first excluding from random selection all companies that were recently involved in their own takeovers, or would soon become so. Companies who had been takeover participants within the previous thirty one months or who would become participants before another thirty four months, were excluded. When takeover participants are included within bootstrap samples it is not clear what a significant outcome implies. Significant differences may emerge because controls are also involved in takeovers, as targets, say. Given the large gains noted for many targets, the selection of targets as controls may introduce a negative bias. The second series of bootstrap samples overcomes these problems. Similar issues arise in the construction of benchmark indices. However, benchmark biases can be overcome by bootstrapping. Bootstrapping exposes differences between bidder sample firms and many random control samples. All bidder sample measures may be biased in relation to a particular benchmark but that bias is eliminated when compared with similarly biased controls. (14.) See da Silva Rosa (1994) or Simmonds (2003) for further details. (15.) A further restriction is also sometimes applied to sample measures. This is that at least five returns must be observed in the second period, [+8,+31]. (16.) The data set only includes takeover or tender offers. Mergers, or Schemes of Arrangement, are not examined. (17.) The same robustness restrictions are maintained here as in Simmonds (2003). These are: 1) At least five monthly return observations must be present in both the pre- and post-offer estimation periods; 2) The squared correlation coefficient for each model or [R.sup.2], must exceed 1%; and 3) Model estimates further than three standard deviations from the cross-company mean are excluded. (18.) See Barber and Lyon (1997), Kothari and Warner (1997). (19.) For instance, thin trading can be a source of bias for market model parameter estimates. See Scholes and Williams (1977), Dimson (1979). (20.) [T-ratio.sup.1), the cross portfolio measure, implies the mean beta is not significantly different from one. However, [t-ratio.sup.2] indicates look-alike betas are typically less than one. [T-ratio.sup.2] is based on t-statistics for individual firms, rather than cross-sample parameter variability, and is more sensitive to sample skewness. As the columns labelled % < [C.sup.[dagger]] show, more than 50% of estimated betas are smaller than one, in both the bidder and look-alike samples. This skewness produces median beta values substantially less than one and also the significant outcome for [t-ratio.sup.2]. (21.) Each month the market benchmark is comprised of all firms eligible as size- and survivor-matched candidates for look-alike selection. It is an equal-weighted average of candidate returns. If bootstrap sampling was completely randomised, through time and across candidates, it would draw candidate firms from the benchmark population with equal chances. When the benchmark averaging procedure is also applied to the samples, each sample result represents a random portrayal of the benchmark. Averaged across many samples, a completely randomised bootstrap procedure mimics the benchmark. Complete randomization and consistent procedures ensure bootstrap methods are unbiased. Complete randomization is not applied to bootstrap samples selected here and this is the reason for non-zero pre-offer RAR estimates. The bootstrap samples are constrained to match the bidder sample at particular points in time. This timing constraint produces the departure from benchmark levels. (22.) Assuming q, B, F, fixed event costs and R, the discount rate, are constant. (23.) Responses to anticipated investment announcements, [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.], may also be significant but can be expected to cancel across an observation period with [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.], for individual firms because the unconditional expected change for anticipated investments is zero. That is, E(change) = E(change|no announcement) x Prob(no announcement) + E(change|announcement) x Prob(announcement) = (0-qB)(1-q) + (B-Bq)q,= 0 (24.) The two effects may not be independent. Updated anticipation of acquisition benefits may follow news of cash flow improvements for existing operations. Updated anticipation of acquisition benefits may also follow other news than that on cash flows. On the other hand, news on cash flows may have negligible impact on acquisition intentions. Firms with improved cash flows may invest in-house rather than buy from outside. Considering the possibilities, evidence for each explanation is examined separately. (25.) Also assuming that acquisition benefits do not fall as expectations of an offer increase. For instance, bidders making many offers are not expected to obtain smaller acquisition benefits from later targets. Smaller acquisition benefits would arise if investment opportunities are limited and bidders target the most valuable opportunities first. Declining acquisition benefits are less plausible if investment opportunities are not limited or successful bidding requires skill development. Bidders making offers for a succession of companies might be expected to improve their abilities to identify target opportunities and also their acquisition skills. (26.) Pre-offer RARs should be lower for the least surprising bidders when acquisition losses are expected. (27.) The identification of previous offers spans changes in both target and bidder names. All known offers since 1972 were examined to determine whether previous bids had been made. It should be noted, however, that the identification of previous offers does not follow changes in management or corporate group structures. Such changes are relevant because these may alter a company's perceived acquisition preferences without that company ever making an offer under its own name. A predilection for acquisitions may be imported with new managers or owners. Even so, the measure we adopt does identify the level of acquisition activity undertaken by indvidual companies. (28.) It should be noted that an individual bidder company can appear in any or all of the three categories as it moves through its acquisition agenda. (29.) A different possibility is that bidders who survive sufficiently long to achieve 'Experienced' bidder status enjoy significantly higher pre-offer RARs because they are survivors, typically beyond the period over which survivor controls are applied. For instance, 'Experienced' bidders must have survived long enough to make offers for at least two targets before the present bid. It may be their perceived survival qualities and not their expected acquisition benefits that yields higher pre-offer returns. (30). Jennifer Cross at the Department of Accounting and Finance of the University of Western Australia succeeded in converting the binary files to text. Her kind and generous assistance is gratefully acknowledged. (31.) Availability of data after 1985 is the reason for using this data set rather than the Australian Graduate School of Management's Annual Report Record. (32.) The net income measure used here is after tax and interest and any adjustments for extraordinary or abnormal effects but before adjustments for revaluations to reserves (for changes brought forward such as those due to tax effect accounting, currency fluctuations, minority interests), and before deductions for minority interests, any dividend payments and any write-offs of capitalised or mining/exploration expenses, when these are recorded. (33.) EBIT is defined as net income plus income tax expense and interest paid. (34.) Operating cash flow is defined as EBIT plus depreciation expenses for intangibles, fixed assets, written off development costs and other amortisations, amortisations/write-offs of deferred costs, and share/debt issue expenses written off. (35.) Residual cash flow is defined as EBIT plus depreciation plus amortisations minus taxes minus interest minus preference dividends. (36.) Share numbers are adjusted for the effects of bonus/consolidation elements in new share issues that dilute/inflate cash flow entitlements between one annual report and the next. Appendix 1 contains further explanation and a detailed description of the adjustment method. (37.) Adding back depreciation and amortisation expenses also reduces the impact on acquirer earnings of the choice between recognising target assets under the purchase and pooling of interests methods. Under the purchase method, any premium over the target's net tangible asset backing is written up as 'goodwill' in the bidder's asset account and depreciated within twenty years. This depreciation expense is not available under the pooling of interests method. (38.) Between 1971 and 1995 annual inflation rates ranged from 1.5% to more than 16% (Annual rates at December for the Reserve Bank's Consumer Price Index: Total.) (39.) RAR rankings are applied across all available firms and also through time. This is achieved by applying the same market model specifications described above in (1) to all firms. Treating the current month as a pseudo-offer month produces RAR estimates at that time for all firms. These are ranked to obtain boundaries for the top 30%, bottom 30% and the rest. Moving forward one month the whole process is repeated to generate boundaries at the next month. The same procedure is applied for all available months and produces boundary time series. It is these boundary time series that define the population segment membership of each sampled firm. (40.) 39.2% is more than 6 standard deviations above the expected population proportion of 30% and 18.1% is more than 8 standard deviations below its mean. (41.) Averages and medians of relative values are found after forming relative values. (42.) Trimming was also applied at two standard deviations from the mean (not shown) and produced no material changes, implying the following results are not due to further extreme outcomes. (43.) A positive bias can exist even if non-contributing firms do not expire. Financially distressed firms are likely to involve greater compliance and collection costs for Statex. Annual reports of distressed companies are therefore more likely to be missing, late or incomplete in the Statex system. In contrast, companies with complete information across the data set will tend to exhibit strong performance. Hence, a survivorship bias can be expected, which favours superior performance for contributing firms. (44.) In a sense, contributing firms are winners in a lottery of economic activity. Losers cease to participate. Since only winners are observed, their economic performance is likely to be relatively good. (A similar argument justifies the ex-post creation of survivor-adjusted benchmarks to control possible survivor biases in ex ante equilibrium values when investors price survival prospects.) (45.) Reliable overall industry measures were not available. (46.) Upper tail values remain influential when All Available data are used. High-range bidders then report an average of $0.775 in {+5} and a median of $0.45. (47.) Implicit restrictions diminish because there is no requirement for sampled bidders to exist in {+5}. If performance is poor, firms can exit and still be included in the bidder sample. The same condition is not true at {-5} because sampled firms are known to be bidders in {0}. If sampled firms exist at {-5}, they must survive to {0}, when the sample is identified. They are certain to perform well enough to avoid perishing between {-5} and {0}. Sampled bidder firms may perish after {0}. (48.) It is important to note that the bidder operating margins in figure 3 do not recognise asset mix changes before and alter a takeover offer, following the possible combination of bidder and target assets. Pre-offer reports correspond to just tile bidder's original assets, whereas post-offer reports can consolidate accounts of targets as subsidiaries. Combining reported measures for each firm and comparing operating margins before and alter an offer, reveals an explanation for investor confidence; operating margins for the combined High-range firm are greater after an offer than beforehand. This is despite a drop from levels achieved with the original assets of High-range bidders. High-range bidders are able to sufficiently increase the relatively low margins of their high volume targets that margins across the combined firm grow, as do total residual cash flows. An ability to expand margins for the combined firm is not displayed by any other bidder RAR-types. Over the five years preceding an offer, {-5} to {-1}, the median operating margin is just 5.5% for trimmed targets of High-range bidders with Complete-Target-Data between {-4} and {-1}. These are the lowest and most stable target operating margins across the three bidder types. Targets of Mid-range bidders generate margins of 6.5%, while those of Low-range bidders are closer to 7%. Over this same interval, median margins for bidders are 8.6%, 6.4% and 8.1% and median revenues for bidders/targets are $130m/$69m, $156m/$66m and $187m/$50m for High-, Mid- and Low-range bidders. Assuming bidder and target firms do not trade with each other before the offer, and no efficiency improvements arise from the combination of bidder and target firms, High-range bidder firms can be expected to generate margins near 7.5% when target and bidder operations are combined (=(130*8.6%+69*5.5%)/(130+69)). Similarly, post-offer margins of 6.4% and 7.9% can be expected for Mid- and Low-range bidders. In the four years following the offer, {+1} to {+4}, median bidder margins were actually 11.6%, 6.8% and 7.1% for High-, Mid- and Low-range bidders, representing a change of +4.1%, +0.4% and -0.8% from the no synergy case. Only High-range bidders achieved an improvement in combined firm operating margins. This same outcome obtains when results from the latest year before the offer are used, {-1}, rather than pre-offer medians. Expected margins are then 9.3%, 6.8% and 8.5%, when no synergies exist and realised post-offer margins deviate from expectations by +2.4%, 0.0% and--1.4%. (49.) The balance sheet date for the average bidder sampled is actually 0.6 months after its offer date. (50.) One difficulty with arguing news for {-2} and {-1} most likely influences pre-offer RARs is that official confirmation of balance sheet outcomes publicly reaches the market later, on the earnings announcement date, not the balance sheet date. Earnings announcement dates are not widely available for companies in this research. They are available at various times in the Statex data set for 1256 companies. A review of the available earnings announcement dates found the typical delay from the end of a reporting period until a preliminary earnings announcement is approximately 74 days, or 2 1/2 months. A similar delay was found for available bidder data. When a delay of 2V2 months is included, year {-1} information typically reaches the market 9V2 months before the offer announcement month, year {-2} results arrive at [-21 1/2] and year {-3} data emerge at [-33 1/2]. Allowing for a delay of 2 1/2 months, earnings announcements for years {-3} and {-2} typically fall within [-34,-11], as do the entire reporting periods for {-2} and {-1}. The announcement of {-1} cash flows is not within [-34, -11] but is imminent and typically arrives just 1 1/2 months later. (51.) For instance, selective briefings on price sensitive information, before market-wide announcements, or 'the tweaking of profit expectations has been a part of the fabric of the corporate community for a long time.' (Knight, 2001b). Commenting on a request by the Australian Securities and Investments Commission (ASIC) for legislative authority to fine companies providing selective briefings, rather than market-wide announcements, Knight (2001b) writes, 'For those unfamiliar with why AMP is being singled out on corporate disclosure, a raft of profit downgrades simultaneously hit the market a few weeks ago after a series of one-on-one broker chats with AMP's corporate affairs manager.... [AMP] said that there was no talk of prospective profits. [However] it seems that, coincidentally, there was then a much heater market consensus on downgrades. Having got wind of this, ASIC asked AMP to make a statement to the market about its profit forecasts.' (52.) With hindsight these firms are known not to be takeover participants, however, their realized takeover activity was unknown to investors in the past. Share prices are influenced by the prospect of unknown future outcomes, hence the possibility that these firms might become involved in takeovers would have been capitalised. Nevertheless, when the market is efficient and makes unbiased predictions, firms not involved in takeovers are less likely to experience significant acquisition related share price effects. (53.) Part of the reason for slower growth in cash flows for bidder shareholders is the change in capital structure that typically follows an offer. Before an offer, the book value of long term debt to the market value of fully paid shares (D/E) is around 40% (median) for all bidders. This jumps to 48% in {+1} and remains above 45% every year after an offer. An increase in interest payments accompanied the greater reliance on borrowed funds. Interest payments jumped from around 50% of shareholder cash flows before an offer to 57.5% in {+11, before falling back to 55% by {+3}. Interest payment obligations remained higher in every post-offer year than any pre-offer year, as a proportion of shareholder cash flows. (Preference dividends are typically negligible.) In contrast, D/E ratios for look-alike firms remain near 22% over both the pre- and post-offer periods. Look-alikes report no sudden changes in D/E in any year. Thus, the sudden jump in interest obligations for bidders at {+1} explains part of their slower post-offer growth, both relative to their own pre-offer performance and to that of look-alikes. However, increased interest payments do not explain the continued slowdown for bidders because interest obligations do not continue to rise after an offer. Notwithstanding the higher payments to lenders, the growth in bidder shareholder cash flows is slower after an offer than beforehand. (54.) This comparison of cash flow growth between bidders and non-takeover look-alikes is not a test of the impact of acquisition policies on the prospects of firms directly affected by each takeover. Look-alike firms are not matched with bidders by industry or pre-offer RAR. Comparisons occur only between RAR ranked portfolios matched on size and survival qualities. RAR rankings may be correlated between bidders and some firms. For instance, it is possible that pre-offer controls for Low-range bidders tend to be High-range firms. If so, a more appropriate test of acquisition effects would compare growth patterns between these categories. It is conceivable that growth for such High-range look-alikes may slow by more than the growth of their Low-range bidder counterparts and still be stronger than the growth of High-range bidders. Such results are consistent with acquisition related impacts on look-alike growth. Supplementary tests found RAR rankings for bidders and these look-alike firms are independent. Hence, the conclusions reached for the broader market are appropriate for the comparisons made. (55.) Earlier we found post-offer bidder operating margins exceeded weighted averages of pre-offer bidder and target margins (see footnote 47). We concluded that bidders generate wealth from target operations and this sustained post-offer growth in bidder residual cash flows (Figure 4). Despite continuing to grow, Table 5 implies bidders are not able to maintain the same rate of growth in post-offer cash flows as look-alike firms who do not acquire. (56.) Kendig (1999) argues constraints on managers relax after periods of market prosperity and that this leads to increased takeover activity. Our cash flow results support Kendig's argument. (57.) Table 6 demonstrates the bidder sample does indeed contain significantly higher proportions of positive pre-offer RAR estimates than the broader population. It presents the proportional distribution of bidder sample pre-offer RAR estimates relative to corresponding time-aligned measures for all survivor companies It is generated from decile rankings of RARs for all companies with similar survival characteristics to bidders and systematic risk exceeding 1%. The market model with a single equal-weighted benchmark index was applied to generate RAR measures. The table reports the proportion of the bidder sample falling into each population-wide pre-offer RAR decile group. Ten percent of the bidder sample would fall into each category if bidder pre-offer RAR estimates were distributed like their time-aligned survivor counterparts. (58.) This is broadly consistent with the significant book-to-market benchmark in the three factor model of Fama and French (1993) since firms with large pre-offer RARs are more likely to also possess low book-to-market values. (59.) No adjustments are applied for capital structure changes involving other parties than all existing shareholders. For example, share placements to new investors, or a selective issue to existing shareholders, do not produce adjustments even though cash flow entitlements may be affected. This is because any resulting transfers to or from existing owners is part of the firm's normal operations rather than a repackaging of entitlements for existing owners. (60.) In practice, the effect of the Dividend adjustment is not found to be material across the different bidder categories. In contrast, the Bonus Element adjustment does affect bidder groups differently. The bonus element in High-RAR Range (defined above) bidder issues exceeds that in other bidder categories. (61.) More generally, any issue offered to existing shareholders at less than the current market price involves a bonus or unfunded component, which dilutes cash flows per share and must be controlled between annual report dates. Renounceable issues, for instance, include both funded and unfunded components. Another example illustrates the point. Suppose shareholders are offered one new share at a price of 100 for every two shares already held. If the share price is 130 just before trading commences in the new entitlements, immediately afterwards it is expected to fall to 120 = 130 + 130 +100/3. Because cash flow entitlements are distributed equally across both old and new shares, the price of old shares falls, from 130 to 120, and that for newly issued shares rises, from the 100 paid to 120. The old share price falls because just 100 is raised for new investment when the net present value of existing operations is 130. [CapitalisationAdjustment.sub.1] equals the expected price change per existing share, or 10. This is the net present value of existing cash flows per share that are transferred to holders of newly issued shares. It is the amount diluted from an existing share by the new share issue. An accurate comparison of cash flows between annual reports before and after the renounceable issue requires an adjustment to maintain consistent entitlements with pre-issue conditions. The Bonus Element Adjustment makes such an adjustment. For this renounceable issue example, the Bonus Element Adjustment is 1 + 10/120 = 130/112. Multiplying post-issue cash flow entitlements, 120, by the Bonus Element Adjustment, recovers pre-issue cash flow entitlements, 130, and restores a consistent basis for comparisons. (62.) Note Price, is the first closing price ex- the event. In the case of dividends, [Price.sub.t] is the first ex-dividend closing price. In the case of capitalisation changes, [Price.sub.t] is the first ex-change closing price. It usually occurs in the same month as the event ex-date. References Acharya, S. 1993, 'Value of latent information: Alternative event study methods', Journal of Finance, vol. 43, pp. 413-29. Aitken, M. & Czemkowski, R. 1992, 'Information leakage prior to takeover announcements: The effect of media reports', Accounting and Business Research, vol. 23, pp. 3-20. Asquith, P. 1983, 'Merger bids, uncertainty and stockholder returns', Journal of Financial Economics, vol. 11, pp. 51-83. Barber, B.M. & Lyon, J.D. 1997, 'Detecting long-run abnormal stock returns; The empirical power and specification of tests statistics', Journal of Financial Economics, vol. 43, pp. 341-72. Bishop, S. 1991, 'Pre-bid acquisitions and substantial shareholder notices', Australian Journal of Management, vol. 16, no. 1, pp. 1-33. Bishop, S., Dodd, P. & Officer, R.R. 1987, Australian Takeovers: The Evidence 1972-1985, Centre for Independent Studies, Policy Monograph 12, Sydney, Australia. Brown, P. & da Silva Rosa, R. 1998, 'Research method and the long-run performance of acquiring firms', Australian Journal of Management, vol. 23, no. 1, pp. 23-38. Casey, R.S. 1987, Takeover Defence Strategies. The Law and Practice in Australia, Information Australia Group, KPMG Peat Marwick Hungerfords. Casey, R., Dodd, P. & Dolan, P. 1987, 'Takeovers and corporate raiders: Empirical evidence from extended event studies', Australian Journal of Management, vol. 12, pp. 201-20. Castagna, A.D. & Matolcsy, Z.P. 1985, 'Accounting ratios and models of takeover target screens: Some empirical evidence', Australian Journal of Management, vol. 10, no. 1, pp. 1-15. da Silva Rosa, R. 1994, The Economic Consequences of Takeovers in Australia: A Re-Examination, PhD Manuscript, University of Western Australia. Dechow, P. 1987, 'The share market's assessment of initial acquisitions by seven controversial investors', Australian Journal of Management, vol. 12, pp. 23-48. Dimson, E. 1979, 'Risk measurement when shares are subject to infrequent trading', Journal of Financial Economics, vol. 7, pp. 197-226. Dodd, P. 1976, 'Company takeovers and the Australian equity market', Australian Journal of Management, vol. 1, pp. 15-35. Eckbo, E., Maksimovic, V. & Williams, J. 1990, 'Consistent estimation of cross-sectional models in event-studies', Review of Financial Studies, vol. 3, pp. 343-65. Fama, E. & French, K. 1993, 'Common risk factors in the returns on stocks and bonds', Journal of Financial Economics, vol. 33, pp. 3-56. Franks, J.R. & Harris, R.S. 1989, 'Shareholder wealth effects of corporate takeovers: The UK experience 1955-1985', Journal of Financial Economics, vol. 23, pp. 225-49. Jensen, M.C. & Ruback, R.S. 1983, 'The market for corporate control: the scientific evidence', Journal of Financial Economics, vol. 12, pp. 13-32. Karafiath, I. & Spencer, D.E. 1991, 'Statistical inference in multiperiod event studies', Review of Quantitative Finance and Accounting, vol. 1, pp. 353-71. Kendig, C. 1998, Takeover Waves: Behaviour, Motives and Consequences, PhD Manuscript, University of Melbourne. Kothari, S.P. & Warner, J.B. 1997, 'Measuring long horizon security price performance', Journal of Financial Economics, vol. 43, pp. 301-39. Knight, E. 2001a, 'Chance to convert a weed into a rose', Sydney Morning Herald, 14th June 2001, p. 27. Knight, E. 2001b, 'Putting a culture on the stand', Sydney Morning Herald, 15th August 2001, p. 21. Malatesta, P. & Thompson, R. 1985, 'Partially anticipated events: A model of stock price reactions with an application to corporate acquisitions', Journal of Financial Economics, vol. 14, no. 2, pp. 237-50. Melicher, R.W., Ledolter, J. & D'Anotonio, L.J. 1983, 'A time series analysis of aggregate merger activity', Review of Economics and Statistics, vol. 65, pp. 423-30. Nelson, R.L. 1959, Merger Movements in American Industry: 1895-1956, Princeton University Press, New Jersey. Palepu, K. 1986, 'Predicting takeover targets: A methodological and empirical analysis', Journal of Accounting and Economics, vol. 8, pp. 3-35. Pound, J. & Zeckhauser, R. 1990, 'Clearly heard on the street: The effect of takeover rumours on stock prices', Journal of Business, vol. 63, no. 3, pp. 291-308. Prabhala, N.R. 1997, 'Conditional methods in event studies and an equilibrium justification for standard event-study procedures', Review of Financial Studies, vol. 10, no. 1, pp. 1-38. Roll, R. 1986, 'The hubris hypothesis of corporate takeovers', Journal of Business, vol. 59, no. 2, pp. 197-216. Schipper, K. & Thompson, R. 1983, 'Evidence on the capitalised vale of merger activity for acquiring firms', Journal of Financial Economics, vol. 11, pp. 85-119. Scholes, M. & Williams, J. 1977, 'Estimating betas from non-synchronous data', Journal of Financial Economics, vol. 5, pp. 309-402. Simmonds, D. 2003, 'A comparison of two widely applied models of shareholder returns in the context of takeover offer gains for Australian bidders: 1976 to 1995', Australian Journal of Management, vol. 28, no. 1, pp. 23-61. Walter, T. 1984, 'Australian takeovers: Capital market efficiency and shareholder risk and return', Australian Journal of Management, vol. 9, no. 1, pp. 63-118. David P. Simmonds, Department of Finance H03, University of Sydney, NSW 2006 and Centre for Research in Finance, Australian Graduate School of Management, UNSW, Sydney, NSW 2052. Email: davids@agsm.edu.au |
|
||||||||||||||||||
Printer friendly
Cite/link
Email
Feedback
Reader Opinion