The megadeal wave meets the size effect

The market has a wave of large deals again. PwC’s H1 2025 outlook reports deal values up 15 percent and deal volumes down 9 percent between the compared first-five-month periods, with 36 megadeals against 31. The report defines a megadeal as more than US$5 billion and says its H1 figures are extrapolated from the first five months of global announced transactions.

That is an activity description, dated to H1 2025. It is not a return forecast.

That boundary is the same discipline used when evidence must survive a hostile reading: the question and the evidence object stay separate.

The temptation is to reach for the older size effect. Small acquirers once appeared to do better than large acquirers. A later study reports a more positive post-2009 mega-deal result. A current wave then arrives and makes both numbers feel like competing explanations. They are not one series. They use different periods, thresholds, units, outcomes, and source versions.

The useful answer is a regime map: keep each observation intact, then ask which question it can answer.

Why is the historical M&A size effect merely a three-day announcement artifact?

Moeller, Schlingemann, and Stulz’s 2004 Journal of Financial Economics article studies 12,023 US acquisitions announced from 1980 to 2001. The sample includes purchases over US$1 million and classifies public acquirers by market capitalisation. Its headline outcome is a three-day cumulative abnormal return around the announcement.

In the equal-weighted comparison, the mean announcement CAR is 1.102 percent overall. It is 2.318 percent for small acquirers and 0.076 percent for large acquirers. The numbers are striking because they are close to one another in time and use the same event window. They are not a statement about whether the integration worked, whether the deal completed, or what happened to the business three years later.

The weighting matters too. Equal-weighted means give each deal one vote. A dollar-weighted result asks what happened to shareholder wealth across dollars invested. The 2004 paper contains both kinds of question, and they should not be collapsed into one sentence called the size effect.

For a second example of keeping a large claim attached to its distribution, see the European failure number nobody quotes.

How do long-term mega-deal return studies alter the scale comparison?

Alexandridis, Antypas, and Travlos revisit US M&A through 2015. The full-text record held for this note is their accepted manuscript, not the version of record. The publisher page confirms the Journal of Corporate Finance publication, volume 45, pages 632 to 650, but the published PDF was not held in the captured session. Every number from this study is therefore labelled as an accepted-manuscript result.

The accepted manuscript reports 26,078 US deals in its sample, including 3,811 deals from 2010 to 2015. It defines mega-deals at US$500 million and reports a 2.54 percent abnormal return in its post-2009 mega-deal comparison, together with a reported US$62.3 million shareholder gain. This is a later-period announcement result with a different threshold and a different sample construction from the 2004 size comparison.

The sign looks like a reversal. The safe statement is narrower: the accepted manuscript reports a more positive post-2009 mega-deal comparison than the older large-acquirer result. That observation could reflect a changed market regime, governance, deal selection, sample filters, or some combination of them. It is not evidence that a permanent law replaced another permanent law.

The source-version label is part of the finding. An accepted manuscript is a complete research record for the edition held here, but it is not a typeset version-of-record page. The number may be used in this note with that boundary visible. It may not be presented as a page-pinned finding from the published edition.

How does programmatic deal experience separate serial acquirers from one-off buyers?

Hu, Li, Li, and Wang provide a second reason not to speak of a single mega-deal result. The full-text record held here is also an accepted manuscript, while the publisher record confirms Journal of Empirical Finance 55, pages 119 to 142. The accepted manuscript studies 3,544 US deals announced from 1980 to 2016, with deal value above US$500 million in 2016 dollars.

The study keeps several outcomes separate: the probability of completion, the three-day announcement CAR, the 36-month stock return, and return on assets. It also separates deals by prior acquisition experience. High experience means at least 12 completed prior transactions in the reported classification.

That design makes the board question more precise. Is the concern that the market will punish the announcement, that the transaction will fail to complete, that the acquirer will underperform over three years, or that the operating business will not improve? Experience may sort those outcomes differently. The accepted manuscript reports a moderator in an observational design. It does not make experience a treatment that causes every experienced acquirer to create value.

Why does dollar-weighted tail risk distort perceptions of typical M&A transactions?

The 2005 Journal of Finance paper by Moeller, Schlingemann, and Stulz supplies the warning that an equal-weighted average can hide. In its lower-tail analysis, 87 large-loss deals account for about US$397 billion of loss. The result is about dollar wealth and concentration in the lower tail, not the announcement return of a typical acquisition.

The distinction is operational. An equal-weighted announcement CAR tells you how the market reacts to the average deal in a specified sample and event window. A dollar-weighted lower-tail result tells you how much wealth can be concentrated in a small number of very large losses. Neither tells you what a current deal will return.

Which five transaction dimensions must deal committees audit before signing?

The comparison is easiest to keep honest when the evidence sits in rows rather than on one axis.

The table is intentionally awkward. A common y-axis would imply that 2.54 percent, 0.076 percent, US$397 billion, and 36 deals are observations of the same thing. They are not. The value of the map is the refusal to smooth them into a single trend.

Why is historical performance reversal an unreliable forecast for specific transactions?

Four boundaries stop the older number from becoming a current recommendation.

The periods differ. The 2004 result covers announcements from 1980 to 2001. The accepted 2017 manuscript extends through 2015. The accepted 2020 manuscript extends through 2016. PwC describes H1 2025 activity. A sequence of dates is not a controlled time series.

The thresholds differ. The older study uses purchases over US$1 million and classifies acquirer size. The accepted 2017 and 2020 studies use mega-deal thresholds around US$500 million, with the 2020 threshold stated in 2016 dollars. PwC uses more than US$5 billion. Calling every row a large deal does not make the populations equal.

The outcomes differ. Announcement CAR, completion, 36-month stock return, ROA, and aggregate dollar loss are separate dependent variables. An announcement reaction can be positive while a later operating outcome is not. A dollar-weighted tail can be severe while the equal-weighted mean remains a different object.

The source versions differ. The 2004 and 2005 published PDFs are held and page-checkable. The 2017 and 2020 full-text records are held accepted manuscripts. The publisher pages confirm the latter articles’ bibliographic records and expose previews, but no version-of-record PDF was held in the captured session. The distinction is not a footnote. It determines how confidently a number can be repeated and what kind of locator it can carry.

The prudent conclusion is not that megadeals work or fail. It is that the current wave must first be classified by the question being asked. A board discussing announcement risk should not quote a 36-month ROA result. A team discussing operating integration should not use a three-day CAR as proof of execution. A market report describing 36 megadeals should not be turned into an expected-return line.

Which four diagnostic questions should board members ask before trusting deal stats?

1. What is the unit? Deal, acquirer, shareholder wealth, operating business, or market activity?
2. What is the window? Announcement day, completion, 36 months, or a current reporting period?
3. What is the threshold and currency year? US$1 million, US$500 million, US$5 billion, or a different definition entirely?
4. What does the source version permit? Published article, accepted manuscript, practitioner report, or a filing with its own reporting boundary?

Then ask the question the statistic cannot answer: what would change the decision? If the answer is integration sequencing, customer retention, operating capacity, or financing, the event-study number is context rather than evidence for the decision.

Where are the empirical limits of transaction scale and size studies?

This note compares distinct historical and current descriptive records. It does not pool their statistics, estimate a current return, or claim that the post-2009 result is a permanent reversal. The 2017 and 2020 findings are used only as explicitly labelled accepted-manuscript evidence. PwC’s H1 2025 report is a global activity description based on its stated method and threshold, not a forecast and not a substitute for deal-level return evidence. A later version-of-record acquisition for either accepted manuscript would require a fresh source comparison.

Evidence base. The analytical frame also draws on these additional sources: Alexandridis et al. 2017; Hu et al. 2020; Moeller et al. 2005. The links identify the exact works; they support the mechanisms and boundary conditions discussed here, not every claim in isolation.