Technological Disruptive Potential and the Evolution of IPOs and Sell-Outs

Since the late 1990s, the number of private firms exiting via initial public offerings (IPOs) in U.S. markets has sharply declined. At the same time, the number of exits via acquisitions (i.e., sell-outs) has soared. Successful firms are nowadays more likely to sell-out to other (public or private) companies than seek independent public listings, a phenomenon that has recently garnered considerable attention in the media and policy circles.

In this paper, we show that U.S. startups developing technologies with disruptive potential are more likely to exit through IPOs than selling out. The role of technology in explaining exit choices is especially prevalent for startups with the potential to disrupt established technological areas as opposed to those introducing new technologies. We further document that the average disruptive potential of startups has markedly decreased in recent years and estimate that changes in startups' technological traits can explain 20% of the recent decline in IPOs and 50% of the surge in sell-outs.

Our analysis of startups' exits builds on the idea that rational entrepreneurs (and their backers) choose the exit option that maximizes the value of their equity stake. Using the theory of the firm (Grossman and Hart (1986)) and Hart and Moore (1990)) and the exit theory of Bayar and Chemmanur (2011) as a foundation, we propose that these exit-specific valuations depend on technological characteristics such as the potential to disrupt established technologies, or to complement existing inventions through synergies. The relative attractiveness of IPOs compared to sell-outs thus hinges on the interactions between potential buyers' technologies and those of the startup, as well as the allocation of payoffs between parties. We posit that technologies with disruptive versus synergistic potential differ notably along both dimensions, and therefore trigger different exits. By design, startups developing disruptive technologies offer limited synergistic value to other parties because disruptive inventions tend to be substitutes that can replace existing technologies (Acemoglu, Akcigit and Alp (2014)). In addition, the economic success of startups with disruptive potential can be achieved while remaining independent, as it is not necessary to integrate or seek strategic assistance from potential acquirers (Bayar and Chemmanur (2011)). This independence avoids the need to share future payoffs with another party. Hence startups with technologies with high disruptive potential should favor exiting via a public listing.

Testing this hypothesis requires the ability to measure technological characteristics at the patent and startup level. We exploit the voluminous text of all patents filed with the U.S. Patent and Trademark Office (USPTO) between 1930 and 2010 (6.6 million patents). We define a patent's disruptive potential as its potential to change the path of technological evolution and eventually disrupt established technologies or create new ones (Dahlin and Behrens (2005)). We measure a patent's “technological disruptive potential” (henceforth “disruptive potential”) based on the extent to which its vocabulary is new or growing fast across all contemporaneous patent applications. For example, the use of genetics words such as “peptide”, “clone”, or “recombinant” soared in 1995, reflecting concurrent breakthroughs in genome sequencing. Our measure would classify patents using such words in 1995 specifically as having high disruptive potential.

A key advantage of this new measure, which is based solely on the text in a patent and contemporaneous patents, is that it is measurable immediately when a patent is filed. In contrast, citation based measures typically are only available 6-8 years after a patent application and require fixes for truncation bias (Lerner and Seru (2017)). This advantage is crucial for our tests, which estimate predictive models of startup exit. That text-based patent measures offer near immediate measurability also increases their utility to practitioners wishing to predict outcomes or returns and regulators interested in real time assessments of the state of innovation or the impact of new policies.

We validate our measure of patent disruptive potential in four ways. First, we examine whether patents with higher ex ante disruptive potential change the path of technological evolution.  We find that these patents receive significantly more citations and they simultaneously trigger reductions in the citation trajectory of preexisting and related patents. Second, disruptive patents have higher economic value (estimated using stock returns as in Kogan, Papanikolaou, Seru, and Stoffman (2016)), suggesting that the market recognizes their potential. Third, we focus on historical breakthrough patents recognized by USPTO between 1930 and 2010 (including the television, computer, helicopter, and advances in modern genetics) and find that these patents have significantly higher ex ante disruptive potential.  Fourth, we use textual similarity analysis to identify the publicly traded industry rivals for each startup. As direct validation, these public firms discuss market disruption more in their 10-Ks when they operate alongside startups with disruptive potential.

Our primary analysis focuses on the exit decisions of 9,167 VC-backed U.S. startups (94,703 patents) over the 1980-2010 period. Our main result is that startups with one standard deviation more disruptive potential are 25.2% more likely to go public and 18.8% less likely to exit by selling out. This result is robust to controlling for startups' age, size, financing rounds, market conditions, and other patent traits such as technological “breadth” (patents using vocabulary from diverse bodies of knowledge), similarity to other firms, patent citations, average word age, originality and fixed effects for startup cohorts, geographic locations, and technological categories. The results are also robust to several alternate estimation techniques, present in both the early and late part of our sample period, and present in all patent technological areas including biotechnologies and health science. In an important follow up test, we decompose disruptive potential into two parts which focus respectively on the potential to change the nature of established technology markets and the potential to lead to entirely new technologies. We then reexamine exits and find that disrupting established markets is far more important than new market creation in predicting IPOs and sell outs. These results strongly separate our study from the existing literature given that most existing studies focus on new market innovation.

Our second major finding is that aggregate economy-wide disruptive potential has declined substantially since the 1950s, and this trend accelerated in the 1990s. Although the overall trend is interrupted by occasional spikes during the 1970s (i.e., computers), the 1980s (i.e., genetics), and the 1990s (i.e., the internet), following each spike, the trend quickly reverts fully back to the long-term decline. As was the case for our main cross-sectional results for exits, this trend is also stronger for established technology spaces as we observe little change in the potential to disrupt by creating new technologies. These results are consistent with new ideas becoming harder to discover and develop (Jones (2009) and Bloom, Jones, van Reenan, and Webb (2017). Our findings indicate that this conclusion is particularly true in established markets.

Combined, these two conclusions suggest a new explanation for the aggregate trends away from IPOs and toward sell-outs noted in recent studies, and our estimates suggest that shifting technology traits of U.S. startups within explain 20% of the recent drop in IPOs and 50% of the rise in sell-outs. This conclusion is based on fitting a cross-sectional exit model (with and without our explanatory variables) over an initial period (1980-1995), and applying this model out-of-sample (1996-2010) to compute expected exit rates. A model that excludes our technological characteristics predicts an out-of-sample quarterly IPO rate of 0.84 percentage points. The actual rate was just 0.33 percentage points, confirming that IPOs were abnormally rare in recent years. Adding our new technology variables to the fitted model reduces this gap by roughly 20% overall. Similar analysis shows that changing technology traits are especially salient for explaining the increase in sell-outs and the decline in small firm IPOs, which is the market segment displaying the sharpest decline, see Gao, Ritter, and Zhu (2013).