Before the Deal: Who Gets Acquired, and Why?
This chapter examines the pre-deal selection of innovative start-ups and established acquirors. The question is not yet what an acquisition causes, but who enters the transaction and how the relationship between their technologies and markets shapes that choice. Understanding this selection pattern lets us interpret the later post-deal decline without mistaking a firm’s pre-existing attractiveness for an effect of ownership.
What the OECD sample means by an innovative target
The OECD study is deliberately not a census of every young company. It defines the relevant start-up as a young firm oriented toward innovation and growth: its first venture-capital financing round or first patent must occur during the study window, and that first relevant event must happen before the firm is ten years old. An acquisition is a transaction in which an established company obtains more than a 50.01% stake. That threshold matters because the analysis is about a transfer of control, not merely a financing relationship.
The resulting population is broad but selected. It contains 93,855 start-ups across 60 countries and 204 three-digit industries from 2001 through 2021. Of these, 17,732 are acquired start-ups. The associated sample of companies that either acquire a start-up or invest in one through a minority stake contains 156,681 firms across 74 countries and 217 industries after the study’s restrictions. These figures define the world to which the conclusions apply: innovation-visible young firms that can be linked across start-up, acquisition, and patent data. They do not describe all small businesses, all acquisitions, or all forms of innovation.
That selection is useful for two reasons. First, the research question concerns start-ups whose possible future competition and technologies matter, rather than firms that are simply young because they have not yet grown. Second, the study needs to know whether a target was acquired or not. It therefore keeps firms that can be linked to the acquisition database and compares acquired start-ups with firms in the same broad population that were not acquired. The acquisition share is about one-fifth of the final start-up sample, but the regression’s average annual acquisition probability is only about 2.4%. Those are different denominators: one describes firms that eventually appear as acquired, while the other comes from a repeated cross-section in which non-acquired firms can appear in several years.
The study measures innovation in several ways. Patent families provide a count of inventions while reducing the problem of counting the same invention separately in different patent offices. Citation-weighted patent stocks add a rough measure of subsequent influence. Radicalness asks whether a patent draws on technologies and applies them in a substantially different field. Venture capital is used as a proxy for an input into innovation, since early-stage firms often have little revenue and rely on external financing to fund research. None of these measures is innovation itself. Together, however, they make different parts of a young firm’s technical trajectory visible.
Those measurement choices also shape who can appear in the comparison. A start-up that has raised venture capital but has never filed a patent can still enter the broader target analysis, while a technological-proximity comparison requires a patent portfolio on both sides. A firm absent from one database is not necessarily non-innovative; it may be young, private, operating in a field where patenting is uncommon, or simply difficult to match across company records. The OECD therefore restricts the main population to firms that can be linked to acquisition information and uses country, industry, age, and financing information to make comparisons more meaningful. The resulting estimates are about observable, linkable start-ups, not about the unobserved tail of informal or very early experimentation.
The use of standardised patent flows is another important choice. Patent offices, technologies, countries, and years differ in how much they generate filings, so the OECD adjusts invention counts before aggregating them at the firm-year level. This makes a filing in a high-patenting technology less mechanically dominant than it would be in a raw count. It does not solve every comparability problem. It cannot reveal an invention that was never patented, and it cannot tell us whether an unfiled idea was abandoned, kept secret, or incorporated into a product. The measurement is disciplined enough to compare trajectories, but not rich enough to settle the meaning of every trajectory.
The pre-deal comparison is clear. Over the sample period, acquired start-ups average 3.5 patents, compared with 2.1 for start-ups that are never acquired. Their average citation-weighted patent stock is 101.9, compared with 58.4. But the gap shrinks sharply when the comparison is limited to firms that patent at least once. Among patent-holding start-ups, acquired firms average 7.7 patents against 7.0 for never-acquired firms, and 224.6 citation-weighted patents against 197.5. The main difference is therefore often whether a firm has become visibly innovative at all, not whether it has accumulated a dramatically larger portfolio after crossing that threshold.
The regression evidence gives that extensive-margin difference a more controlled interpretation. Holding age and industry–country–year conditions constant, having at least one patent at the previous point in time raises the probability of acquisition by 0.84 percentage points in the extensive-margin specification. The mean probability in that sample is 2.4%, so the estimated premium is about 34.9% of the baseline probability. Venture-capital stock is positively associated with acquisition as well. That result could mean that financing signals quality to a buyer, or that firms needing more money are doing more research and are therefore more attractive for technological reasons.
Among firms that patent at some point, larger patent stocks and higher citation-weighted stocks are also associated with acquisition. The first patent, however, appears more consequential than every additional patent. This is a useful clue about what the buyer may learn from the target. A patent can signal that a technically uncertain firm has crossed an important development threshold. It can reduce information asymmetry by making the direction of the technology more legible. It can also reveal specialized knowledge or provide legal protection around an idea. Those functions make the first visible patent valuable even when the next few patents add less information about the target’s existence or capability.
The association is not uniform across the life of a start-up. The link between holding a patent and acquisition is stronger for firms roughly three to ten years old than for firms younger than three. It becomes slightly negative for firms older than ten in the age interaction reported by the study. The pattern is an inverted U: patents matter most when a young company has developed a technology but has not yet become a mature incumbent. The patent premium is also larger for acquisitions across industries or countries than for deals within them. A patent may be especially useful when a buyer must evaluate an unfamiliar technology or a firm operating in a different market context.
These results establish selection, not causation. A patent may cause no acquisition at all; both may be consequences of an unmeasured capability, a strong research team, or a market with unusual growth potential. The study’s controls reduce some obvious differences, but the pre-deal pattern must remain in view when later comparisons are made. A target that patents more before acquisition was not randomly assigned to the buyer.
The acquiror and the pair are part of selection
Selection does not operate only on the target side. The OECD study compares acquirors with companies that also interact with start-ups but do so through minority investments rather than majority acquisitions. This is a useful comparison because it asks why two kinds of firms that are already interested in young companies make different choices. Acquirors are more likely to hold patents, have larger patent stocks, and have more highly cited patents. In the descriptive table, acquiring incumbents average 18.3 patents and 392.7 citation-weighted patents over the sample, compared with 4.2 and 38.1 for firms that use minority stakes without acquiring. Among firms that patent, the gap is smaller but remains: acquiring firms average 59.7 patents against 44.0, and 1,250 citation-weighted patents against 394.9. These are aggregated descriptive measures rather than a causal effect of acquiring, but they show that the buyer population is not a neutral slice of established firms. The relationship holds both for the decision to acquire at all and, among acquirors, for the cumulative number of start-ups acquired. Radicalness does not differ significantly once the size of the patent portfolio is taken into account.
One interpretation is absorptive capacity. An incumbent with a substantial technical base may be better able to recognise the value of a start-up’s invention, integrate it with existing knowledge, and pay for implementation. Acquisition is then an external extension of an active research system. Another interpretation is strategic. Innovation leaders may be more likely to buy the start-ups most capable of changing their markets. Their technical strength would make them good buyers, but it would not tell us whether they intend to develop or neutralise what they acquire.
The composition of the acquiror sample also matters. Most acquiring firms make only one acquisition in the period, while a smaller group of serial acquirors accounts for repeated deals. This is not just a descriptive curiosity. A serial buyer may have a different acquisition strategy, integration capacity, or patenting trajectory from a company making a single purchase. Later analysis must therefore avoid treating every acquiror as if one transaction were their only exposure to M&A.
The relationship between a specific target and a specific acquiror adds another layer. To study it, the OECD compares actual deals with fictitious alternatives. Each acquired target is matched with potential acquirors from the control group, and each actual acquiror is matched with non-acquired potential targets. Some comparisons are random; others match targets on age and venture-capital level before examining the relationship of interest. The procedure is repeated across 100 random draws. The goal is not to imagine a perfect alternative history, but to ask which deal characteristics make the observed pair more likely than other plausible pairs in the same menu.
Technological proximity is measured by comparing the firms’ patent stocks across 35 broad technological fields. A score near zero means that their patent portfolios occupy largely different fields; a score near one means that they are much more similar. Actual acquisition pairs tend to have greater technological proximity than the fictitious alternatives. Pairs are also less likely when they cross industries or countries. In ordinary language, buyers tend to purchase firms they can understand, reach, and connect to their existing businesses.
The pattern changes when the dimensions are combined. Technological proximity matters more for cross-industry deals. A buyer entering a new industry may seek a start-up whose technology resembles what it already knows, because that similarity creates a bridge into an unfamiliar product market. By contrast, within an industry, the buyer may not need a technically similar target to see a strategic reason to purchase it. A start-up with a different way to make the same broad product may be a differentiated competitor or a more efficient future producer.
This is where the same observation supports competing stories. Shared technology can mean complementarity: the incumbent has the absorptive capacity and assets needed to scale the invention. Shared industry or country can make integration easier and generate operational synergies. But proximity to the incumbent’s market can also reveal a threat. If the target’s technology could displace the buyer’s product, the buyer may value control even when it has little interest in developing the target’s independent path.
The pair evidence is informative but bounded. Technological proximity can be calculated only for firms with patents, so the pair sample excludes many young companies whose innovation is not captured there. Industry and country are proxies for product-market overlap, not complete descriptions of substitutability. A firm can compete across industry boundaries, and two firms in the same industry can serve different customers. The fictitious-pair design also depends partly on choices such as the number of counterparts and the rule preventing reuse of a potential partner, although the reported results remain qualitatively stable under alternative thresholds.
The pre-deal picture can now be stated precisely. Acquisitions are disproportionately connected to visible start-up innovation, innovation-oriented incumbents, and relationships that cross a technological bridge. That is evidence that technology access and the ability to absorb innovation are important. It is also compatible with defensive acquisition, especially where the target is close enough to threaten an existing business. The data do not identify the buyer’s intention simply by showing that the buyer and target look capable of working together.
This distinction is essential for the central question. Selection tells us why acquired start-ups may have looked unusually innovative before the transaction and why the parent may have been unusually capable. It does not tell us whether the deal caused the target to patent less afterward. To answer that, the comparison must construct a credible version of what would have happened had the deal not gone through.