When Corporate R&D Becomes a New Company
The subject in this chapter is the link between corporate research and development and employee-founded ventures. The specific question is whether more R&D makes employees more likely to leave for young firms, and whether those departures appear to carry new intellectual opportunities rather than only general workplace skills. Answering it helps distinguish a plausible causal pathway from a simple correlation between innovative companies and ambitious people.
From an R&D budget to a founding team
A firm’s R&D spending can produce several kinds of return. It may improve an existing product, solve a problem in the parent’s main business, or create an option that is too uncertain or too distant from the firm’s current strategy to develop internally. The last kind of return is easy to miss in conventional measures of corporate performance. An employee may recognize its value before the parent has decided that it belongs in the company at all.
That possibility creates a puzzle. R&D might make a firm more attractive and give employees better reasons to stay. It might also increase the firm’s growth prospects, making departure less appealing. But R&D can have the opposite effect if it generates ideas that employees can pursue more effectively in a new company, where they receive stronger ownership incentives and do not have to fit an unfamiliar project into the parent’s existing priorities.
Babina and Howell study this question using employer-employee records linked to public-company financial data. Their parent firms are public companies that report positive R&D spending. The researchers observe workers at those firms, follow them across employers, and identify a young firm as a “spawn” when a former employee becomes one of its top five earners within three years. The measure is designed to capture a founding team and important early employees, because administrative employment data do not label legal founders directly.
In this population, the average entrepreneurial-spawning rate is 1.3% over the relevant follow-up period. That number is a share of an establishment’s employees, not the probability that a whole corporation launches a startup. A change in the rate can therefore reflect a small but economically meaningful shift in how many people move into new firms. It also means that the estimated percentage changes must be read relative to a low baseline rather than as a change in the number of all employees who leave.
The baseline relationship is positive. In the authors’ preferred controlled ordinary least squares specification, a one-standard-deviation increase in R&D is associated with an 18.9% increase in entrepreneurial spawning relative to the sample mean. A 50% increase in R&D corresponds to roughly a 3% increase in spawning. The regression includes firm or establishment fixed effects, industry-year and state-year controls in the more demanding specifications, and time-varying measures of size, age, growth, assets, investment, and other firm characteristics.
These controls address several straightforward alternatives. Firm fixed effects compare a company partly with itself over time rather than treating every persistent difference between an innovative company and an ordinary one as an R&D effect. Industry-year controls absorb changes in opportunities within narrow sectors. State-year controls absorb regional shocks that could affect both R&D and new-firm creation. The result remains when the industry definition is made more detailed, so it is not easily explained by a broad sector simply becoming fashionable.
Still, a controlled association is not automatically causal. An unobserved technological opportunity could cause a firm to raise R&D and cause its employees to found companies at the same time. A company might also increase R&D because it already expects entrepreneurial opportunities to emerge. The researchers address these concerns with an instrumental-variables strategy based on changes in federal and state R&D tax credits. Those policy changes affect the user cost of R&D and therefore shift the firm’s incentive to spend, while being less directly tied to the individual startup decision.
The first-stage relationships are strong by the study’s reported diagnostics, with F-statistics around 25. The authors also investigate the exclusion concern: the tax credits should affect spawning through R&D rather than by directly subsidizing new firms or creating a startup opportunity of their own. They argue that these credits are generally not useful to startups without taxable income, and they find no direct relationship between state tax credits and startup creation. State credits can also induce firms to relocate or reclassify some R&D, which would weaken the instrument rather than make it cleanly stronger.
The preferred instrumental-variable estimate is substantially larger than the ordinary least squares estimate. It implies that a 50% increase in R&D is associated with an 18.3% increase in spawning, or that a one-standard-deviation increase is associated with approximately a doubling of spawning. That difference is informative but not a license to select the larger number as the “real” effect. Instrumental variables identify a local average treatment effect for firms whose R&D responds to the tax-price changes. Those firms may be unusually sensitive to incentives or may be conducting the kind of marginal research most likely to generate opportunities outside the parent.
The authors offer several reasons why the two estimates might differ. The marginal R&D dollar may be more exploratory and less central to the parent’s core, so it may produce more externally deployable ideas. Firms whose R&D is easy to adjust may also be less damaged by losing an idea, making them less likely to prevent employee departures. Or the IV may simply be selecting firms with a larger causal response. The defensible conclusion is narrower: the tax-credit evidence strengthens the case that increased corporate R&D leads to more employee spawning in the studied population, while the precise magnitude for all firms remains uncertain.
The timing tests reinforce that interpretation. If new firms were causing public companies to increase R&D, past spawning should predict current R&D. It does not in the reported tests. Nor does R&D appear to cause a general increase in turnover: the spending change does not significantly alter the shares of employees who remain at the parent, move to another incumbent, leave the observed labor force, or move to firms of unknown age. The pattern is more specific than “R&D makes workers more mobile.” It is especially connected to movement into young firms.
The sequence is not meant to imply that every R&D project creates a startup or that every employee observes a transferable invention. It shows the decision point the data are trying to identify. R&D changes the set of opportunities employees encounter; employees then sort among staying, moving to another incumbent, and joining a young firm. The new company may develop an idea, skill, or network that would have remained dormant inside the parent. The evidence is strongest on the change in spawning, not on observing each invisible step between research and founding.
Why the evidence favors intellectual capital
The central mechanism question is what R&D changes in the employee. One possibility is human capital: R&D work teaches people to manage uncertainty, coordinate technical teams, or become more confident entrepreneurs. Another is intellectual capital: research produces ideas or technical insights that can be deployed in a new firm. These channels can coexist, but they make different predictions about which parents, workers, and startups should show the largest response.
The observed pattern fits the intellectual-capital channel more closely. The R&D-spawning relationship is concentrated among high-tech parents rather than spread evenly across firms with any positive R&D. It is stronger for firms whose patents are more general-purpose—ideas cited across a wider range of technology classes—and for more inventive firms. That is what we would expect if the relevant output were a novel or broadly useful opportunity, rather than only the project-management experience acquired while doing research.
The characteristics of the new firms point in the same direction. Within the population of spawns, more parent R&D is associated with high-tech ventures, higher-wage founding-team members, and venture-capital-backed startups. Only about 2% of the spawns receive venture capital, so the relationship is not merely a broad shift in all new businesses. It is a relationship with a small group of ventures commonly used as a proxy for high-risk, high-growth opportunities. Parent R&D is also associated with a greater chance that the startup exits through an IPO or acquisition within the measured horizon.
These outcomes are not direct observations of idea transfer. Venture capital, high wages, and later exits are indicators of opportunity type and growth potential, not proof that a specific research result crossed the firm boundary. They could also reflect selection: employees who are already unusually capable may choose both high-R&D parents and ambitious startups. The mechanism evidence becomes persuasive because several proxies point in the same direction, not because any one proxy settles the question.
Industry distance provides another clue. If R&D-induced ventures were mainly copies of the parent’s existing business, we would expect them to remain in the same product area. Instead, only 16.8% of spawns are in the parent’s two-digit industry, and more parent R&D makes same-industry spawning less likely. The opportunity may therefore be new to the venture and peripheral to the parent. A researcher can carry knowledge of a technical possibility into a business that is not a direct replica of the employer’s market.
The human-capital explanation is not disproved, but several of its simple predictions are weaker. If employees become entrepreneurial mainly because of broad learning-by-doing, small firms might be the strongest source: workers at small organizations often have wider responsibilities. The observed effect is instead driven more by large parents. A young firm might also offer more entrepreneurial learning than an old one, yet the effect does not vary in the expected way with parent age. Capital spending could similarly develop project-management skills, but its per-dollar association with spawning is much smaller and less robust than that of R&D.
The most balanced interpretation is therefore that research creates intellectual opportunities, while human capital and networks may help employees recognize and exploit them. The evidence does not require a choice between “idea” and “person.” It says that a purely generic training story is insufficient to explain why high-tech parents, general-purpose inventions, and high-growth startups are disproportionately involved.
When a spillover is not a replication
Knowledge spillovers are benefits from one firm’s innovation that reach other firms without being fully embodied in a product sold by the original innovator. Employee spawning offers a concrete transmission channel. The parent conducts research, an employee gains access to an opportunity, and a separate firm develops it. The value may then appear as a new product, jobs, competition, or further innovation rather than as revenue booked by the parent.
This interpretation depends on two questions. First, does the new firm impose a large cost on the parent? Second, does the parent normally internalize the venture’s private benefits through investment, acquisition, or another contractual arrangement? Babina and Howell find no large observable cost in their tests. The R&D-spawning effect persists in states that enforce non-competes, does not vary with an industry measure of patentability, and is not concentrated in ventures that directly copy the parent’s industry. They also find no reported relationship between R&D and establishment employment growth or labor productivity.
Those findings should be read as absence of evidence for large measured costs, not evidence that no cost exists. The tests may miss lost project continuity, legal disputes, licensing, strategic hiring, or a temporary reduction in innovation that appears only in a different dataset. A parent may also anticipate departures and adjust compensation or project selection before an observable loss occurs. The study’s conclusion is that it does not detect an offsetting cost large enough to erase the spawning relationship in the outcomes it can measure.
The internalization test is similarly informative but limited. In an out-of-sample dataset of venture-backed startup executives linked to former employers, only 2.3% of 9,152 parents invested in or acquired one of their spawns. Even when the denominator is restricted to parents that could be matched to an investor or acquirer, the authors’ conservative upper bound is 7.9%. This suggests that parents do not usually capture a spawn’s private value by buying or financing it. It does not rule out licensing, informal relationships, or unobserved investments, and the exercise concerns a different, VC-backed population.
The geography of the movement complicates the picture in a useful way. About 88% of spawns are located in the same state as the parent, while many operate in a different industry. Knowledge can therefore remain locally embedded even when the product-market boundary changes. The employee may retain access to regional networks, specialized labor, or tacit information, while the new venture uses the inherited opportunity in a direction the parent would not have pursued.
Calling this a spillover does not mean that the employee has stolen a finished product or that property rights are irrelevant. It means that the original R&D creates value that may be developed outside the firm’s ownership boundary. The parent may have chosen not to pursue the idea, may have been unable to write a complete contract around it, or may have judged the cost of preventing departure higher than the benefit of keeping the opportunity in-house. The social value of the new venture and the parent’s private return can consequently diverge.
What this mechanism can—and cannot—explain
The evidence supports a coherent sequence. More corporate R&D, especially at high-tech and inventive parents, is followed by more employee spawning. The resulting ventures are more likely to display characteristics associated with new-to-the-world and high-growth opportunities. Their distance from the parent’s industry is consistent with ideas that are valuable but not central to the parent’s existing business. This makes intellectual capital the leading explanation for the observed pattern.
But the sequence should not be inflated into a complete biography of every startup. The worker-level measure does not observe the idea in transit, formal founder status, or every employee’s role. The tax-credit design identifies R&D changes induced by particular incentives and may emphasize marginal projects or firms with unusually responsive spending. The mechanism tests are cross-sectional and indirect, so selection, human capital, networks, financing, and opportunity recognition may all remain active.
The most important implication is about firm boundaries. Corporate R&D is not only an input into the parent’s own products. It can also enlarge the set of opportunities available to employees, some of whom will find that a separate firm offers the right incentives or organizational space. That outward flow can create a high-quality entrant without requiring the parent’s research to have been a failure. It may be a sign that the research generated more possibilities than one organization could profitably pursue.
The next task is to place this upstream mechanism beside the performance evidence and ask how much of the causal chain is directly observed. A complete account must also follow what happens to the parent when the employee leaves. Only then can the knowledge-spillover story be weighed against the innovation capacity that the parent may temporarily lose.