The cumulative abnormal return, or CAR, is the stock market’s snap reaction to an acquisition announcement. Over the last five decades, CAR has dominated academic finance: more than 92% of M&A studies in the top journals use it to measure deal quality. Its influence extends well beyond academic research: CAR is the standard framework taught in business schools, and event studies built on announcement returns are routinely used by expert witnesses in deal litigation and by regulators evaluating the competitive effects of mergers. If CAR were a reliable measure, the implications for corporate governance would be substantial—boards could use it to evaluate management’s dealmaking, compensation committees could tie incentive pay to deal-level value creation, and antitrust investigators could benchmark their enforcement decisions against the market’s verdict.

But what if CAR does not actually measure value creation?

In our article recently published in the Journal of Finance (published version; SSRN), we present comprehensive evidence that it does not. Using more than 47,000 acquisition announcements over nearly four decades (1980–2018), we find that the market’s initial reaction to a deal bears essentially no relation to how that deal actually turns out.

The Measure Everyone Uses—But No One Has Validated

How dominant is CAR? We reviewed every M&A article in the Journal of Finance, Journal of Financial Economics, and Review of Financial Studies from 1972 to 2021. Of the studies that measure acquisition value creation, 92% use CAR. That is 202 articles, with no declining trend. Yet apart from two small-sample studies based on data from the 1970s and early 1980s, no one has tested whether CAR actually measures what everyone assumes it measures: value creation.

Announcement Returns Do Not Predict Deal Outcomes

We assess CAR’s reliability against multiple measures of how acquisitions actually perform, drawing on different data sources to capture different facets of deal success:

• Goodwill impairment: we manually track whether the goodwill associated with a specific transaction was materially written down within five years, a direct signal that the target turned out to be worth less than what was paid;

• Abnormal operating performance: changes in the acquirer’s return on assets relative to its pre-deal performance and industry benchmarks, measured over both short-term (three-year) and long-term (six-year) horizons; and

• Deal completion: whether the announced transaction was ultimately completed or withdrawn.

Despite capturing different aspects of deal success and coming from entirely different data sources, these outcome measures are significantly correlated with each other, a reassuring sign that they pick up the same underlying signal of deal quality.

The problem is that CAR is correlated with none of them. Three-day or eleven-day announcement windows, with or without controls, in-sample or out-of-sample, full sample or dozens of subsamples split by time period, deal type, acquirer characteristics, or target characteristics: the result is always the same. Announcement returns are essentially uninformative about how the deal will perform.

We even conduct a brute-force data-mining exercise across complex subsample formations, deliberately searching for a “golden subset” of deals in which CAR reliably predicts outcomes. We cannot find one.

Publicly Available Information Predicts Outcomes, but CAR Does Not Capture It

If deal outcomes were inherently unpredictable, CAR’s failure might be unsurprising. But they are not. We build a simple benchmark model using standard deal and acquirer characteristics known at the time of the announcement: relative deal size, method of payment, target’s public status, acquirer size, leverage, and past returns. This straightforward model predicts acquisition outcomes reasonably well, both in-sample and out-of-sample.

The implication is striking. Publicly available information has meaningful predictive power for deal success, yet the stock market’s reaction fails to reflect it.

To illustrate the gap, we sort acquirers into deciles based on predicted outcomes from either CAR or the characteristics model and track their long-term stock returns. The characteristics model produces a five-year return spread of 8% to 11% between the best- and worst-predicted deals. CAR produces a spread of less than 3%. The market’s initial reaction simply does not sort good deals from bad ones the way that readily available deal characteristics do.

“Listening” to CAR Destroys Value

Should managers use the stock market’s reaction to guide their decision to complete or withdraw a deal? Prior literature has suggested that managers “listen to the market,” completing deals the market applauds and withdrawing those it punishes. If CAR were a reliable signal, this would be good governance.

We test this directly. Following CAR’s advice (completing positive-CAR deals and withdrawing negative-CAR ones) generates a five-year return loss of approximately −5% relative to doing the opposite. Following the advice of our simple characteristics model, by contrast, generates a long-term return spread exceeding 20%. A board that relies on announcement returns for its “go or no-go” decision would, on average, make worse choices than one using a basic checklist of observable deal features.

CAR Gives Misleading Answers About Which Deals Create Value

Much of what practitioners and governance professionals believe about which types of deals create value comes from CAR-based research. The conventional wisdom holds that cash deals outperform stock deals, that private-target acquisitions outperform public-target ones, and that smaller acquirers do better than larger ones. These beliefs are built largely on patterns in announcement returns.

We compare how CAR ranks different deal types against how they actually perform. Using the four characteristics most commonly studied (form of payment, target’s public status, acquirer size, and relative deal size), we create 16 clusters of transactions. The results are striking: the cluster ranked as the best value creator by CAR has the worst ex-post outcomes among all 16 clusters. Conversely, the cluster ranked as the biggest value destroyer by CAR is associated with above-median ex-post outcomes.

For boards, investors, and advisors who rely on these established patterns to evaluate deal proposals, the implication is significant: the received wisdom about what makes a “good” acquisition may be systematically wrong.

The Smoking Gun: Dollar CAR Tracks the Acquirer, Not the Deal

The results above show that CAR fails empirically. But why? The answer becomes vivid when we convert announcement returns into dollars and ask: does the implied value creation scale with the size of the deal, or the size of the acquirer?

Consider Microsoft. In 2022, it acquired Activision Blizzard for $68 billion, and its stock rose 2.42% around the announcement, implying value creation of roughly $55 billion. Large, but at least in the same ballpark as the purchase price. Two years earlier, Microsoft purchased CyberX for $0.16 billion. The stock rose by a similar 2.25%, this time implying value creation of $33.5 billion. If CAR measures deal NPV, then investors believed Microsoft transformed a $160 million investment into $33.6 billion, a staggering 200× multiple, and that CyberX’s shareholders sold their company for pennies on the dollar. That interpretation is not credible.

This is not a cherry-picked case. We examine all 140 acquisitions by Cisco Systems, one of the most prolific serial acquirers in U.S. history and a staple of business-school case studies. For each deal, we ask whether the dollar value implied by CAR falls within a plausible range: is the implied gain or loss smaller than the deal price itself? An astounding 91% of Cisco’s deals fall outside this range. The announcement returns look roughly the same regardless of whether the target costs $50 million or $5 billion. Whatever the market is reacting to, it is not the specific deal.

The pattern holds across the full sample of more than 47,000 deals. Take CAR at face value, and in about 27% of positive-CAR deals the implied value creation exceeds the entire purchase price, meaning target shareholders supposedly sold at a discount of more than 50%. In about 27% of negative-CAR deals, the implied value destruction exceeds the full amount invested. Over one-quarter of all observed CARs yield economically implausible valuations.

Why? Acquisitions are not random events. Companies acquire in response to specific triggers: a failed R&D project, a competitive threat, a strategic pivot, a new CEO. When the deal is announced, investors update their beliefs not only about the deal itself but also about the acquirer’s standalone prospects and the circumstances that prompted the transaction. Because the acquirer is typically much larger than the target, even a small percentage revision in investors’ view of the acquirer can dwarf any deal-specific signal.

Our formal analysis confirms this: the dollar magnitude of CAR comoves with acquirer size 6 to 13 times more than with deal size. The stock price movement at announcement tells you far more about what investors think of the acquirer than what they think of the deal. Without knowing what the market is learning about the acquirer’s standalone value, one simply cannot extract deal-related value creation from the announcement return.

Implications for Governance, Litigation, and Investor Decision-Making

If CAR were a reliable barometer of deal quality, it should be harnessed to improve economic efficiency and corporate governance. For example, executives’ incentive pay and promotion could be tied directly to the value created in specific deals; directors could use value de-struction indicated by negative CARs as cause to revisit strategy or management; and the judgment of antitrust authorities could be benchmarked against the information conveyed in announcement returns. Our findings suggest that such reliance would be misplaced:

• Litigation and regulatory proceedings. Event studies anchored on acquisition announcement returns are routinely used in deal litigation, including appraisal proceedings and fiduciary duty challenges, and have been proposed as a tool for evaluating the competitive effects of mergers. Our results suggest that courts and regulators should exercise caution in interpreting these studies as definitive evidence of value creation or destruction.

• Board oversight of M&A. If boards were to evaluate management’s acquisition track record based on announcement returns, they could draw the wrong conclusions. A CEO whose deals consistently generate negative CARs is not necessarily destroying value, and one whose deals are cheered by the market is not necessarily creating it. Operating performance and other outcome measures offer a more reliable basis for assessment.

• Executive compensation. To the extent that incentive structures reward or penalize executives based on stock performance around deal announcements, those structures risk rewarding noise rather than skill. Compensation committees should consider whether announcement-period returns actually capture deal quality.

• Academic research and conventional wisdom. Many widely accepted conclusions in the M&A literature about which deal structures, payment methods, or acquirer types create value are derived from CAR. As we document, these conclusions can be systematically wrong when measured against actual outcomes, and they may need to be reexamined.

The bottom line: when a deal is announced, the stock price moves six to thirteen times more with acquirer-related information than with deal-related information. What looks like the market’s verdict on the transaction is overwhelmingly the market updating its view of the company. For fifty years, researchers, judges, directors, and investors have been reading a signal that is dominated by noise about the acquirer rather than news about the deal. It is time to develop, and adopt, better measures of acquisition value creation.

The published paper is available here (SSRN version).