Research method and the long-run performance of acquiring firms.

Keywords:

TAKEOVERS; ACQUIRING FIRMS; LONG-RUN PERFORMANCE.

1. Introduction

It is well established in the event-study literature that apparently innocuous variations in sample characteristics are associated with systematic differences in empirical estimates of relative performance. These findings may be explained, at least in part, by the research method employed. For instance, Roll (1983) finds that over half the small firm effect can be accounted for by the method used to calculate returns. Similarly, Conrad and Kaul (1993) show that bias in cumulated returns can explain much of the evidence that otherwise suggests long-term market overreaction to portfolios consisting of either `winner' or `loser' firms. In the same vein, Brown and Pope (1994) argue that survival issues and the determinants of the bid-ask spread may account for much of the apparent drift in earnings over the post-announcement period.(1)

As noted in many studies, long-run returns to acquiring firms are vulnerable to research design bias. The typically large size of acquiring firms, their higher than usual survival characteristics, their exceptional performance in the pre-bid period and the estimation of their long-run returns by cumulating periodic (typically, monthly) returns are all associated with demonstrated bias in the assessment of relative performance over long horizons. This suggests that these factors may well account for the impressive body of evidence which indicates that acquiring firms experience systematically poor performance over the long-run in the post-bid period.

The impact of firm size on assessments of the performance of acquiring firms has been investigated by Franks, Harris and Titman (1991) and Agrawal, Jaffe and Mandelker (1992). Read together, their findings indicate that controlling for the (typically, negative) empirical association between firm size and return only ameliorates the negative drift in post-bid returns to acquiring firms. Agrawal, Jaffe and Mandelker conclude that `the resolution of this anomaly remains a challenge to the profession' (p. 1620).

Notwithstanding Agrawal, Jaffe and Mandelker's conclusion, their results are consistent with the argument that long-run returns to acquiring firms are vulnerable to multiple sources of research design bias. The studies by Franks, Harris and Titman, and Agrawal, Jaffe and Mandelker also highlight that two issues are prominent in event studies of takeovers: measurement of shareholder wealth effects and assessment of their significance. In this paper we focus principally on assessing the significance of takeover-related share market performance by acquiring firms after controlling for firm size, survival and method of return computation.(2) The `look ahead' approach we adopt to match the experimental and control firms on survival relies on information unknown at the portfolio formation date--that the firms would survive for the next N months. Consequently the performance of the experimental firms (relative to their controls) is not an abnormal gain or loss from an investment strategy that can be implemented.

Research design manipulations that yield revised estimates of the significance of acquiring firms' performance in the post-bid period may also lead to similar revisions of their performance over the pre-bid and bid announcement periods. And so, in this paper we review relative performance over the following event periods expressed relative to the bid announcement month: [-36,-6] months, [-3, +3] months and [+6, +36] months.(3)

Our findings indicate that after controlling for survival, firm size, and method of return computation, and using non-parametric resampling tests of significance, the performance of acquiring firms relative to control firms is consistent with the outcome expected from informationally efficient capital markets. Pertinently, controlling for survival, firm size and return computation has a varying impact on the significance of the relative performance of acquiring firms over the pre-, post- and bid periods. The significantly positive exceptional performance exhibited by acquiring firms in the period [-36, -6] months does not depend on controls for survival, firm size and method of return computation. Their generally positive returns over the period [-3, +3] months are significantly exceptional only when mean portfolio return is calculated on a buy-and-hold (BH) as opposed to a periodic (monthly) rebalanced (RB) basis and after controlling for firm size and survival. Over the period [+6, +36] months, the performance of the acquiring firms is unexceptional on a BH return basis if survival and firm size are controlled. However, calculating returns on an RB basis dramatically revises this finding. On an RB basis, the performance of the acquiring firms over the post-bid period is unequivocally significantly negative. We posit an explanation that connects this finding to higher frequency trading in acquiring firms' shares relative to the control firms' shares. The explanation is elaborated in section 4.

The rest of the paper is organised as follows. The data are discussed in section 2. Research method, including construction of the experimental and control portfolios, is the focus of section 3. Section 4 includes the results and associated discussion. A summary and conclusions comprise section 5.

2. Sample

Our monthly share market data are sourced from: (a) the Share Price and Price Relative (SPPR) database compiled by the Centre for Research in Finance (CRIF) at the Australian Graduate School of Management (AGSM); and (b) the Statex Database compiled by the Australian Stock Exchange (ASX). All share returns are adjusted for changes in the basis of quotation (e.g. dividends, new issues and capitalisation changes). At the time of this study, share market data were available for the period spanning January 1974 to June 1996. We review returns on just fully paid ordinary shares, which are the most senior equity security on issue.

Our experimental sample firms are drawn from the population of ASX listed firms that made successful takeover bids for other ASX listed firms over the period January 1973 to June 1995. Successful takeover bids are defined as all bids in which the offerer acquired more than 50% of its target's shares. Although effective control may be achieved through a holding of less than 50% of a firm's issued shares, our operational definition of a successful takeover bid ensures that bids reviewed in our study are only those where it is unambiguously clear that control of the target has passed to the bidder firm. Over 3,000 bids for ASX listed firms were made over the period of our sample. However, just 731 of these bids satisfy all the selection criteria.(4) The sample size is reduced further if an experimental sample's event period [-36, +36] months falls outside the period, January 1974 to June 1996, for which monthly share return data are available. For instance, the event-window [-36, +36] months fell within the period January 1974 to June 1996 for just 415 of the firms involved in the 731 bids.

3. Assessing Long Horizon Returns: Issues and Research Method

Our research method is based on the premise that controlling for qualitative factors missing from theoretical models of expected return such as the Sharpe-Lintner CAPM better places us to assess the proposition that the post-merger share returns to acquiring firms reflect an informationally efficient market. Following Beedles, Dodd and Officer (1988), we say qualitative because although the identified factors are empirically associated with systematic regularities in returns calculated from historical price series they do not have equilibrium bases and hence cannot be used to reject rigorously specified null hypotheses.

Given the qualitative factors we control for do not have equilibrium bases, in this paper we focus on determining if, over as given event period, the performance of acquiring firms is statistically exceptional.(5) We do this by comparing the mean return to a portfolio of such firms against an empirically generated distribution of the mean returns to 1,001 portfolios, each comprising firms matched on size-decile and survival with the experimental sample firms but otherwise selected randomly. In other words, our control is the empirical distribution under the null hypothesis.

The specific resampling technique we employ to construct the control portfolios is described later in section 3. In the interim we discuss our reasons for matching on firm size and survival and review the measurement issues connected with computing long horizon returns.

3.1 Firm Size

The rationale for controlling for firm size when assessing the long Horizon performance of acquiring firms is well established by Franks, Harris and Titman (1991), and Agrawal, Jaffe and Mandelker (1992). The principal reasons are the widespread and robust finding that there is a (typically, negative) empirical association between firm size and return and that acquiring firms are mostly large firms. This implies that, other things being equal, if firm size is not controlled we may expect to observe a negative mean abnormal return to a portfolio of acquiring firms.

Given that Agrawal, Jaffe and Mandelker (1992) find that controlling for firm size does not eliminate the negative drift in post-merger return,(6) a pertinent concern is whether our matching criteria capture all relevant cross-sectional variation in expected return. For instance, although Fama and French (1992) show that both size and book-to-market equity (BE/ME) are empirically associated with average returns to otherwise randomly selected portfolios, we do not match on the basis of BE/ME. We do not match on the basis of BE/ME principally because of lack of practical access to the requisite data. However, research by Anderson, Lynch and Mathiou (1990) indicates that, in Australia, the size-effect drives the BE/ME effect. Further, Rau and Vermaelen (1996), in reviewing the postacquisition performance of firms classified as either `glamour' or `value' acquirers, match on the basis of both size and BE/ME and find that their results are similar to those obtained on the basis of matching on a size-based benchmark alone (see their fn. 3, p. 8). Given that the empirical association of BE/ME with average return may be expected to be particularly strong in sample portfolios comprising either `glamour' firms or `value' acquirers, Rau and Vermaelen's results suggest that in a study where no such partitioning of acquirers is undertaken there is even less reason to expect different results by matching on the basis of BE/ME.

Notwithstanding the above, in principle there is an indefinite number of potential variables on which sample firms may be matched with control firms. The ideal set of matching criteria consists of all variables that are empirically associated with cross-sectional variation in share returns; and we can be sure that not all these variables have been discovered.(7) If two or more portfolios display no significant difference in return despite failure to match on all the ideal criteria then we may presume that the portfolios are cross-sectionally diversified on the excluded criteria. That is, failure to match on all potentially relevant criteria increases our confidence--although we cannot be certain-that there is no significant difference in performance across the groups, if we fail to find such a difference.

3.2 Survival

Brown, Goetzmann and Ross (1995) point out that selecting a sample on criteria related to survival can give rise to apparent abnormal performance. Kothari and Warner (1997) show that detectable differences in mean abnormal return and the standard deviation of return arise from conditioning on survival. The differences are a consequence of the difference in average return to firms that survive over a given period and the average return recorded by firms that delist or are newly listed. For instance, newly listed firms recorded systematically negative CARs in the US over the 1970s and 1980s (Loughran & Ritter 1995). As Kothari and Warner (1997) note, any selection criteria that have the consequence of systematically excluding newly listed firms will impart a positive bias to the average CAR to the selected firms.

The direction of the bias imparted to the selected firms by omitting delisted firms is difficult to specify ex ante. In Australia, as in the US (Shumway 1997), around half the delisted firms become so as a result of a takeover or merger (Singh 1997). Acquisition-related delistings typically spell `good news' for the shareholders of the delisted firms in that they earn positive abnormal returns subsequent to the announcement of a bid. Pertinently, we are aware of this because the share trades that make manifest these positive abnormal returns are recorded in places such as the CRSP and the AGSM's SPPR databases. However, firm delistings that occur due to bankruptcy or insufficient capital--what Shumway (1997) terms `performance-related reasons'--have less well documented prices leading down to termination day. Shumway (1997) finds that only 11.7% of the 1,029 performance-related delistings by NYSE/AMEX firms that he investigates have delisting returns available on CRSP; a similar pattern of lacunae in price data exists in Australia (Singh 1997). This matters because, as Shumway documents for US firms, performance-related delistings have adverse effects on the value of firms' shares.

One consequence of the availability of terminal returns being systematically associated with reason for delisting is that the net impact on portfolio performance of firms that subsequently delist is unclear. Attempts to estimate the unavailable returns may impart their own biases. For instance, Barber, Lyon and Tsai (1997) adopt an estimating procedure that assumes the proceeds of delisted firms from a given portfolio are re-invested in the remaining firms within the respective portfolio (on an equally weighted basis), that is, missing monthly returns are filled with the mean monthly return of the remaining firms in the portfolio. The prevalence of positive skewness in the distribution of firms' returns suggests that the Barber, Lyon and Tsai method yields an upward biased estimate of the missing return, given that positive skewness results in the mean return being higher than the median return. If, as they contend, survival and skewness have a significant impact on measures of relative performance then the upward bias in the return to the reference portfolio induced by their procedure is problematic, particularly when the experimental sample portfolio does not possess the same incidence of survival as the control portfolio(s).(8)

One procedure to ameliorate the impact of survival bias without encountering the above problem is to match the sample firms with control firms on the basis of survival. Given that survival is priced ex ante, matching on observed survival rates over a given period does not ensure that differences in ex post returns between the sample and the control firms are wholly unconnected to survival-related pricing.(9) However, if investors expectations of firms survival are unbiased, matching on actual survival over a period will ameliorate the impact of survival-related differences in performance. We investigate this proposition by reviewing the relative performance of the bidding firms before and after explicitly conditioning on survival.

Under the survival condition, for each experimental sample member (a bidding firm) we select the control firms only from the population of firms that were listed over the full period spanning [-36, +36] months relative to the respective sample firm's bid announcement date. This means that if we review relative performance over, say, the period [-3, +3] months, the firms included in the control portfolios (and in the experimental sample portfolio) still have to meet the criterion of survival over at least the period [-36, +36] months. We settle on the period [-36, +36] as the minimum survival period because it encompasses all the event periods over which we assess performance. This implies that when we review the relative performance of the acquiring firms in the pre-, post- and bid periods we can discount the possibility that survival-related differences between the sample and control firms account for any significant differences in performance.(10)

Where we do not condition on survival, we measure returns on a month-by-month basis over the event-window and do not impose any survival requirement on either the sample firms or the control firms beyond the minimal requirement that prices exist for two successive months. The mean monthly returns are cumulated to obtain estimates of portfolio returns over the event period. Notwithstanding that our sample of 731 successful bidders mostly survived, the number of acquiring firms varies significantly across event-months because inclusion in the experimental portfolio depends on the event-month falling within the period January 1974 to June 1996. In general, the further away the event-month is from the bid announcement month, the fewer the acquiring firms with available share price data. For instance, in month [-36] 559 acquiring firms have share price data available but in month [-6] there are 665.

Given that the variation in the number of firms in the experimental portfolio over each event-month is mostly unrelated to acquirers being delisted or newly listed, the associated biases are not expected to impact on estimates of the experimental portfolio's performance. However, although our research method ensures that the number of firms in each control portfolio equals the number in the experimental portfolio, the control portfolios include, in each event-month, a proportionate number of newly listed firms and firms that delist in subsequent months. In so far that these survival-related biases significantly affect the mean monthly returns to the control portfolios, they impact on measures of the relative performance of the acquiring firms.

3.3 Issues in Calculation and Interpretation of Long-Run Returns

It is now commonplace that long-run returns incorporate systematic biases (e.g. Brown & Warner 1980; Blume & Stambaugh 1983; Roll 1983; Conrad & Kaul 1993; Barber & Lyon 1997; Kothari & Warner 1997). For instance, Barber and Lyon (1997) demonstrate that buy-and-hold abnormal returns are subject to a skewness bias, a new listing bias, and a rebalancing bias. Cumulated abnormal returns are subject to a measurement bias, a new listing bias, and a skewness bias, although the skewness bias is less severe than for buy-and-hold abnormal returns.

The new listing bias was addressed in the preceding section. What Barber and Lyon term the rebalancing bias manifests when `buy-and-hold' abnormal returns are defined as the difference between the buy-and-hold return to the sample firms and the return on an index that assumes monthly rebalancing of all securities in it. This bias does not affect our study because we review raw returns, not abnormal returns.

Skewness is a relevant issue. Brown and Warner (1980) show that the use of monthly data significantly increases the severity of skewness. However, as Kothari and Warner (1997) observe, the use of a resampling procedure of the type we adopt to assess the degree of skewness under the null hypothesis can satisfactorily address this problem, at least where the aim is to determine if observed performance is significantly exceptional. Further, when reviewing relative portfolio performance using buy-and-hold returns, we compare the median buy-and-hold return as well as the mean to guard against any difference between the mean returns of the experimental and control portfolios being driven by outliers.

Measurement bias is a problem particularly when discrete returns are averaged across firms and then cumulated over time. Blume and Stambaugh (1983), Roll (1983), and Conrad and Kaul (1993) review the reasons that induce the bias and note that buy-and-hold returns yield estimates that more closely approximate actual investor experience.

We disentangle the impacts of methods of computing long horizon returns by reviewing results to the same portfolios using, where possible, both average buy-and-hold (BH) returns and cumulated monthly rebalanced (RB) average returns. When the survival condition is applied, both the experimental samples and their matched control firms have buy-and-hold returns available over the complete period for which performance is estimated. These buy-and-hold returns are compared with the cumulated monthly returns to the same firms. We report just the cumulated monthly average return to the portfolios compiled without conditioning on survival because not all their constituent firms have a complete set of monthly price relatives over the whole event period to enable calculation of their buy-and-hold return. Comparison of the cumulated returns before and after conditioning on survival indicates the impact of survival on reported performance.

3.4 Size-Matching and Construction of Control Portfolios

Our procedure for matching on size is revealed in the steps we follow to construct 1,001 control portfolios.(11) The control portfolios are formed under the null hypothesis of no association between the fact that a firm is involved in a takeover bid and its return. The composition of the control portfolios differs depending on whether we condition on survival. To facilitate the exposition, we describe the procedure assuming the event-window of interest is the period spanning [-3, +3] months relative to the takeover bid announcement month, given that the bidder was listed for the whole period [-36, +36] months. 3.4.1 Survival Conditioned Control Portfolios The first four steps below are repeated for each experimental sample acquirer that has price data available over the period [-36, +36] months relative to its respective bid announcement month.

1. Identify the set {LS} of all listed firms that survived over the same [-36, +36] months defined relative to the sample firm's bid announcement month.

2. Calculate the size (i.e. price per ordinary share as at the beginning of month [-36] multiplied by number of issued shares) of each firm in set {LS} and identify the size decile of the experimental firm.

3. Select a firm from the same size decile as the experimental firm, drawing randomly and sampling with replacement, and allocate it to the first of the 1,001 control portfolios.

4. Repeat step 3 one thousand times, each time moving on to the next control portfolio in the sequence of 1,001 control portfolios.

3.4.2 Non-Survival Conditioned Control Portfolios

Where we do not condition on survival, the procedure for selecting the control portfolios is the same as described above to step 4. The only differences are that: (a) steps 1-4 are repeated for each month over the period [-3, +3] months (for instance, in month [-3] the set {LS} consists of all firms that have a price relative available in the same month); and (b) market capitalisation is calculated anew each month using the share price and number of issued shares at the beginning of the month.

Note that buy-and-hold returns (as well as monthly RB returns) may be calculated only for the firms that have been conditioned on survival. The nonsurvival conditioned portfolios will not be composed of the identical set of firms over the event period and so their returns are calculated only on an RB basis, that is, by cumulating mean monthly returns.

4. Results

4.1 Pre-Bid Performance

The performance of the experimental firms over the period [-36, -6] months relative to the bid announcement month indicates the robustness of a longstanding finding in event-studies of takeovers; bidders are typically firms that have performed exceptionally well. Nevertheless, there are significant differences in the relative performance of the acquiring firms under the varying conditions in which performance is measured.

The left most panel in table 1 shows that the 415 acquiring firms that had price data available over the event-window [-36, +36] months had a mean BH return of 148.11% over the period [-36, -6] months. None of the 1,001 control portfolios achieved a higher mean BH return; the median of the control portfolios' mean returns is 81.37%.

[TABULAR DATA 1 NOT REPRODUCIBLE IN ASCII]

Interestingly, the mean RB return to the same 415 acquiring firms is 122.08%. Roll (1983) points out that, by Jensen's inequality, the RB return will be lower than the BH return if returns are temporally independent. However, 731 of the 1,001 control portfolios have a mean RB return that is higher than their BH return. This manifests in 23 of the 1,001 control portfolios having a higher mean RB return than 122.08%. Further, the distribution of the mean RB return to the control portfolios has a median value of 85.74%, which is 4.37 percentage points higher than the same metric calculated for the portfolios' mean BH returns.

One possible explanation for the divergent impact of method of return calculation on the performance of the sample portfolio and the control portfolios is that shares in acquiring firms are more heavily traded and so their recorded monthly returns are more likely to exhibit temporal independence while the firms in the control portfolio are less likely to be frequently traded and therefore more prone to exhibiting serial dependence in their monthly returns. Serial dependence in returns can cause the BH return to fall below the RB return as the review period lengthens. In proposing this explanation, we are mindful that firm size is positively associated with trading frequency and that the firms in the control portfolios are matched on size-decile with the acquiring firms. However, it is likely that acquiring firms are more heavily traded than the typical firm in their size-decile (particularly in the deciles comprising smaller firms). Whatever the reason, the extent of serial dependence in the returns to firms in the control portfolios is evidently large enough to cause them to exhibit a mean RB return generally higher than their mean BH return.

Notwithstanding the above, it could be argued that the exceptional positive performance of acquiring firms in the pre-bid period (based on mean return) does not reflect the general performance of acquiring firms. Comparison of the median BH return to the acquiring firms against the median BH returns to the 1,001 control portfolios reveals that the acquirers' exceptional performance overall is not driven by a few outliers. We see from the left most panel in table 1 that the median BH return to the portfolio of acquiring firms is 61.82%, which is higher than the median BH return to any of the control portfolios; the distribution of their median BH returns itself has a median value of 44.17%.

The last column in the first panel of table 1 reports the relative performance of acquiring firms if sample selection is not conditioned on survival. The number of acquiring firms with available return data available in each month over the event period [-36, -6] months ranges from a minimum of 559 to a maximum of 655.(12) Over this period the mean RB return to the experimental sample portfolio is 129.75%, a performance equalled or surpassed by just two of the 1,001 control portfolios.

Interestingly, the median value of the distribution of mean RB returns to the 1,001 control portfolios is 70.85%, as opposed to 85.74% when firms are conditioned on long-term survival. The decrease of around 15% in the median value of the distribution probably reflects the strength of the under-performance of IPOs and of the firms that were delisted due to their failure, relative to the superior performance of takeover `victims'. When we do not condition on survival the control portfolios will include a number of newly-listed firms and of firms that delisted due to their poor performance, as well as firms that were taken over.

In summary, two salient findings emerge from our review of acquiring firms' pre-bid performance. One is that the exceptional positive performance of these firms is robust enough to be unequivocally significant regardless of survival or method of return computation. The other is that return effects related to survival and method of return computation impact on the degree of exceptional performance exhibited. Assessing portfolio performance on a monthly rebalanced basis decreases the recorded relative performance of the acquiring firms, notwithstanding the new listing bias that operates in their favour in such comparisons. As the analysis that follows indicates, this overall downward bias may significantly prejudice the conclusions we draw from the performance of acquiring firms over the bid and post-merger periods.

4.2 Bid-Period Performance

Share price adjustment in response to the expected wealth effects of takeover bids may begin well in advance of the bid announcement month and will not be complete until the bid outcome is known. Nevertheless, for the majority of successful takeover bids, it is reasonable to assume that the seven month period centred on the bid announcement is when most of the takeover-related share price adjustment occurs.

The middle panel of table 1 shows that the 415 acquiring firms that had price data available over the six years centred on their bid announcement month had a mean BH return of 19.77% over [-3, +3]. Twelve of the 1,001 control portfolios achieved a higher mean BH return; a measure of the strength of the acquiring firms' performance is the median value in the distribution of the control portfolios' mean returns, which is 12.22%. Pertinently, the exceptional performance of the acquiring firms is not driven by a few firms. The median BH return to the acquiring firms is 11.52%, a performance metric surpassed by none of the 1,001 control portfolios; and the median of the 1,001 control portfolios' median BH return is 7.11%.

The unequivocally significant, positive exceptional performance recorded by the acquiring firms is an unusual finding. Jensen and Ruback (1983) review the then extant studies and observe that the shareholders of acquiring firms do not lose from takeover activity. It is a cautious conclusion endorsed by the majority of subsequent studies. In this context, it is noteworthy that the same 415 acquiring firms do not display significantly positive relative performance on an RB return basis. The mean RB return to the 415 acquiring firms is 15.38%, a performance surpassed by 113 control portfolios (i.e. more than 10% of the control portfolios). The median value of the control portfolios' mean RB returns is 12.40%. Again we observe the BH return to the acquiring firms being greater (albeit slightly) than the mean RB return to the same firms, while the reverse is true for the control portfolios.

The last column in the middle panel of table 1 reports the performance of the acquiring firms over the bid period on an RB return basis without imposing any survival constraint. Over this period, the number of acquiring firms with available price data in each month ranges from 650 to 660. We see that over the period [-3, +3] months the mean RB return to the experimental portfolio is 14.66%, a performance equalled or surpassed by just 32 of the 1,001 control portfolios. The median value of the control portfolios' mean RB returns is 9.68%. Note that the mean RB return to the acquiring firms is not greatly affected by disregarding survival; the 415 firms that met the long-term survival criteria had a mean RB return of 15.38%, that is, just 0.72% higher than mean RB return disregarding survival. However, the 1,001 control portfolios' mean RB returns are more affected by survival. The median value of those mean returns decreases from 12.40% to 9.68%, if survival is ignored. Again, it is likely that the new listing bias and the performance delist bias (both are survival related) account for this fall.

The last point draws attention to an important issue: the relative impact of the survival-related biases and the method of return calculation bias on assessments of relative performance in event studies. Our results suggest that if the survival-related biases were absent then the acquiring firms display insignificant exceptional performance over the period [-3, +3] months based on their mean RB returns. In the Australian share market and probably elsewhere too, the survival-related biases are strong enough to more than off-set the negative bias against the acquiring firms attributable to calculating returns on a mean RB basis, at least over event-periods as long as seven months.

4.3 Post-Bid Performance

Examination of the post-bid returns to bidding firms is important because it is not an uncommon belief that the share market is prone to `speculative bubbles' that result in bidders' prospects (and consequently, their shares) being systematically overrated for considerable periods. This belief is often associated with the view that a correction to the share market's unjustified over-valuation will be observed only in the long-run; that is, the share market does not act quickly to eliminate systematic bias in estimates of firms' prospects.

Efficient market sceptics who hold to some version of the above sketch will not be persuaded by the evidence here, which supports the hypothesis that takeover activity, in general, adds to the long run wealth of the bidder's shareholders. Exceptional positive share market performance by bidding firms around the pre-bid and bid periods is, in their view, consistent with takeovers being, on average, negative net present value investments. The sceptics would argue that the most reliable share market evidence on takeovers is based on the evaluation of returns to successful bidders in the post-bid period, since this is when the consequences of takeover are manifest and when the share market may make an informed assessment of them.

While some might consider that there are ample grounds to discount or even dismiss the views of the efficient market sceptics, considerable force is lent to their argument by the widespread finding that bidding firms experience significantly negative abnormal returns over one to three years after acquisition of the target (e.g. Dodd 1976; Malatesta 1983; Asquith 1983; Limmack 1991; Agrawal, Jaffe & Mandelker 1992). As Jensen and Ruback (1983) comment:

These post-outcome negative abnormal returns are unsettling

because they are inconsistent with market efficiency and suggest

that changes in stock prices during takeovers overestimate the

future efficiency gains from mergers (p. 20).

The results reported in the right-most panel of table 1 indicate quite clearly that the long-term performance of acquiring firms in the post-merger period is sensitive to research design induced bias, which gives rise to the appearance of exceptionally poor performance. The mean RB return over the period [+6, +36] months to the 415 acquiring firms is 53.53%, a performance surpassed by all 1,001 control portfolios. The median value of the control portfolios' mean RB returns is 79.12%. However, the mean BH return to the survival conditioned 415 acquiring firms over the same period is 74.14%, a performance that is surpassed by 618 control portfolios. The median value in the distribution of the 1,001 control portfolios' mean BH returns is 77.10%. Again we observe that the mean RB return to the acquiring firms is lower than their mean BH return. For the control portfolios, the tendency is in the opposite direction; 632 of the 1,001 control portfolios display a mean RB return higher than their mean BH return. The last column in table 1 reports RB returns over the period [+6, +36] months when portfolios are compiled without conditioning on survival. Recall that the results for the control portfolios may be expected to reveal the net effect of the new listing and the delisting biases. The new listing bias will not be present for the acquiring firms, given that they will have been listed for at least seven months (mostly, much longer) prior to the start of this period. The results confirm our expectation, at least for the control portfolios. The median value of the control portfolios' mean returns is 58.33% which is much lower than the same metric when survival is controlled. When selection is conditioned on survival the median value of the mean RB returns is 79.12%. Notwithstanding the large fall in the control portfolios' mean RB return if survival is ignored, 944 of the control portfolios achieve a higher return than the mean RB return of 48.56% to the portfolio comprising the experimental firms. This result is consistent with the findings from other studies and suggests that although the new listing bias and the performance-related delist bias decrease the performance of the control portfolios, they do not offset the negative bias against the acquiring firms due to calculating RB as opposed to BH returns.

5. Summary and Conclusions

In sum, takeover activity contributes significantly in the allocation of corporate resources to higher valued uses. Acquiring firms are typically firms that have performed exceptionally well in the share market's estimation. Their relative performance around the seven-month period centred on the bid announcement month indicates that, on average, the inter-corporate investments they make to build on that performance increase the market value of their shareholders' equity. The long-term performance of the acquiring firms in the post-merger period is consistent with the proposition that the market for corporate control is informationally efficient.

The broad tenor of these findings will not surprise a significant proportion of financial markets researchers. The more remarkable aspects of our results are their methodological implications. Earlier studies have drawn attention to the significant difference in estimates of performance when calculating portfolio returns on a buy-and-hold basis as opposed to a periodically rebalanced basis. Here we find that the direction of the divergence is systematically related to sample characteristics. Mean portfolio returns calculated on an RB basis are not invariably higher (lower) than returns calculated on a BH basis. Importantly, matching control firms on firm size and survival does not ensure that the direction of the divergence in their portfolios' mean RB and BH returns will be the same as for the experimental firms.

Our results also confirm that survival-related biases are an important issue in event studies. Experimental firms typically display survival characteristics that are significantly different from the survival characteristics of other firms. For instance, it is reasonable to suppose that our sample of acquiring firms includes a lower than usual proportion of newly-listed firms. Matching on survival potentially alleviates the problem but the question of how long a period should be used to match on survival remains moot.

(Date of receipt of final typescript: April 1998 Accepted by Garry Twite, Deputy General Editor.)

(1.) They could not account, however, for the fact that much of the drift is concentrated around subsequent earnings announcement dates.

(2.) In an earlier paper (Brown & Da Silva Rosa 1997) we review the returns to shareholders of both acquiring and target firms involved in takeover bids made between January 1974 and December 1995.

(3.) Performance over the period [-6, +6] months was also reviewed. The results are similar to those obtained for the period [-3, +3] months. The results for the shorter period, however, arguably yield more accurate estimates of bid announcement related performance.

(4.) The success rate of takeover bids in Australia is probably higher than is implied by this figure when compared to the total number of bids made. In some cases, although the bidding firm would have achieved a share holding in the target company of over 50%, the outcome was unclear from available data. These bids are excluded from our sample.

(5.) Support for our approach is found in Barber and Lyon (1997), and Kothari and Warner (1997), who show that abnormal return estimates based on a wide variety of matching procedures can be systematically non-zero. This implies it is difficult to justify a rigorous non-statistical interpretation of any single or point estimate of abnormal return.

(6.) They find that Franks, Harris and Titman's (1991) divergent results are specific to their sample period (1975-1984).

(7.) Barber and Lyon (1997) note the obvious: `as future research in financial economics discovers additional variables that explain the cross-sectional variation in common stock returns, it will also be important to consider these additional variables when matching sample firms to control firms' (pp. 370-71).

(8.) Barber, Lyon and Tsai's (1997) procedure is more or less forced on them by their preference for using buy-and-hold abnormal returns as opposed to cumulative abnormal returns (summed monthly returns) when reviewing long-horizon returns.

(9.) Brown, Goetzmann and Ross (1995) point out that given securities are priced on the basis of expected survival, survival-related effects in returns may be present even when no firm actually fails.

(10.) Our procedure for conditioning on survival includes an implicit requirement. Given that our share price database spans just the period January 1974 to June 1996, we are able to identify the firms that survived over the period [-36, +36] months only for those acquiring firms whose event-window did not extend beyond either January 1974 or June 1996.

(11.) One thousand and one now seems a somewhat arbitrary number. It was based on an early judgement (now some three years ago) that we made about the number of trials needed to get an adequate representation of the empirical distribution under the null hypothesis given the survivorship constraint and the number of firms listed on the ASX. Our judgement was `1,000 plus one for good measure' representation of the empirical distribution under the null hypothesis given the survivorship constraint and the number of firms listed on the ASX. Our judgement was `1,000 plus one for good measure'!

(12.) For many firms in our sample, the period [-36,-6] months extends earlier than the earliest month, January 1974, for which we have price relative data.

References

Agrawal, A., Jaffe, J.F. & Mandelker, G.N. 1992, `The post-merger performance of acquiring firms: a re-examination of an anomaly', Journal of Finance, vol. 67, pp. 1605-21.

Anderson, D., Lynch, L. & Mathiou, N. 1990, `Behaviour of CAPM anomalies in smaller firms: Australian evidence', Australian Journal of Management, vol. 15, pp. 1-38.

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 test statistics', Journal of Financial Economics, vol. 43, pp. 341-72.

Barber, B.M., Lyon, J.D. & Tsai, C.L. 1997, Improved methods for tests of long-run abnormal stock returns, Working paper, Graduate School of Management, University of California, Davis.

Beedles, W.L., Dodd, P. & Officer, R.R. 1988, `Regularities in Australian share returns', Australian Journal of Management, vol. 13, pp. 1-29.

Blume, M. & Stambaugh, R.F. 1983, `Biases in computed returns: An application to the size effect', Journal of Financial Economics, vol. 12, pp. 387-404.

Brown, P. & Da Silva Rosa, R. 1997, `Takeovers: Who wins?', Journal of Australian Society of Security Analysts, no. 4 (Summer), pp. 2-5.

Brown, S.J., Goetzmann, W.N. & Ross, S.A 1995, `Survival', Journal of Finance, vol. 50, pp. 853-73.

Brown, S.J. & Pope, P.F. 1994, Post-earnings announcement drift: Market inefficiency or research design biases? Working paper series FD-94-22, Finance Department, Stern School of Business, New York University.

Brown, S.J. & Warner, J.B. 1980, `Measuring security price performance', Journal of Financial Economics, vol. 8, pp. 205-58.

Conrad, J. & Kaul, G. 1993, `Long-term market overreaction or biases in computed returns?', Journal of Finance, vol. 48, pp. 39-64.

Dodd, P. 1976, `Company takeovers and the Australian equity market', Australian Journal of Management, vol. 1, pp. 15-53.

Fama, E.F. & French, K.R. 1992, `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. & Titman, S. 1991, `The post-merger share price performance of acquiring firms', Journal of Financial Economics, vol. 29, pp. 81-96.

Jensen, M. & Ruback, R.S. 1983, `The market for corporate control: The scientific evidence', Journal of Financial Economics, vol. 11, pp. 5-50.

Kothari, S.P. & Warner, J.B. 1997, `Measuring long-horizon security price performance', Journal of Financial Economics, vol. 43, pp. 301-40.

Limmack, R. 1991, `Corporate mergers and shareholder wealth effects: 1977-1986', Accounting and Business Research, vol. 21, pp. 239-51.

Loughran, T. & Ritter, J.R. 1995, `The new issues puzzle', Journal of Finance, vol. 50, pp. 23-51.

Malatesta, P.H. 1983, `The wealth effects of merger activity and the objective function of merging firms', Journal of Financial Economics, vol. 11, pp. 155-81.

Rau, P.R. & Vermaelen, T. 1996, Glamour, value and the post-acquisition performance of acquiring firms, Working paper, INSEAD.

Roll, R. 1983, `On computing mean returns and the small firm premium', Journal of Financial Economics, vol. 12, pp. 371-86.

Shumway, T. 1997, `The delisting bias in CRSP data', Journal of Finance, vol. 52, pp. 32740.

Singh, I. 1997, Measuring the performance of delisted companies on the Australian stock exchange, Honours dissertation, Department of Accounting and Finance, The University of Western Australia, Nedlands.

Philip Brown, Raymond da Silva Rosa, Department of Accounting and Finance, University of Western Australia, Nedlands WA 6907; E-mail: [e-mail]

We acknowledge the expert computing assistance of Jennifer Cross and Paul Vowles. The paper has benefited from comments received from participants at workshops at the Melbourne Business School, the University of Sydney and at the 1997 French Finance Conference.