Anne ter Wal – Imperial College
Oliver Alexy – Technical University of Munich
Jörn Block – Universität Trier & Erasmus University Rotterdam
Philipp Sandner – Frankfurt School of Finance & Management
Haochi Zhang – Northwestern University Kellogg School of Management
Tanya Yuan Tian – Northwestern University Kellogg School of Management
Question 1. Your paper fruitfully integrates considerations of network structures and knowledge base of actors, and shows that closed-diverse and open-specialized networks are positively associated with venture performance. We are interested in learning about the motivation behind this paper more generally. As both the theory propositions and empirical context are intriguing and important, have you considered alternative framing as a more context and phenomenon-driven one than its current form?
Great question. The paper is actually the second in a line of research that we have based on CrunchBase data, on which two of us (Jörn, Philipp) have been experts for a while. From the very beginning, we had made a list of research questions that we were interested in, and the interplay of network structure and knowledge diversity was a line of thought running through all the questions.
In the first paper, we wanted to understand what predicts venture valuation (i.e., how does venture capitalists’ social capital influence the amount of money they pay for a start-up). However, it was nearly impossible to disentangle whether investors with better networks fund better ventures or whether they were willing to pay premiums for the same start-up, given they could leverage them better. We decided to ground this first paper much more strongly in the phenomenon and chose a rather explorative approach. At the same time, in writing that paper, we came to understand that the first-round valuation of venture capitalists would be a great control for venture quality, and armed with that information, we returned to an idea we had pondered for a while – can we help solve the perennial brokerage-closure debate by conceptualizing a network structure that combines the diversity advantages of a network with high brokerage and the trust or interpretation advantages of a closed network?
The question came to us quite naturally. Beyond Jörn and Philipp who are experts on the data, large-scale data analysis, and real-time data processing, Anne has a degree in economic geography (the network perspective), and Oliver in innovation studies and management information systems (the information processing view). We had hoped to write this network paper ever since we had set out working with the CrunchBase data, because it allowed us to measure network structure and knowledge diversity separately. Still, in the first version we submitted to ASQ, we downplayed context – we wanted the social network story to shine through. The reviewers then rightfully suggested that we bring some context back.
Question 2. You used a combination of fascinating data sources including CrunchBase, USPTO, and PATSTAT. We are interested in your experience working with and merging these datasets. Can you share some challenges and your solutions in this process?
Given our set of firms is somewhat limited, the two key problems we think that exist in these types of situations – name disambiguation and nested firm structures – were not as bad and unsolvable as we anticipated. In general, our problem was that we could only merge across the databases on firm name and location, and in particular patent databases are well known for their typos (even if these may later be corrected in so-called reassignments). Here, we were fortunate that Philipp’s Ph.D. was on patent and trademarks, and Jörn and he had extensively worked on the value of patents and trademarks in start-ups before. Hence, they not only had previously invested into name disambiguation, they also had collected firm histories so that for each firm in our sample, we could try and determine whether they were part of a larger firm or conglomerate. Still, in particular for the name disambiguation, that was a team effort. For example, we all read through all the investments made (both the investors and the targets) and evaluated whether two similar sounding firms were indeed the same or whether a name was just popular (indeed, we have a few firms with exactly the same name in the sample, but which are different firms!).
Question 3. The two network types found to be beneficial are closed-diverse and open-specialized networks. Do you think it might be harder to form and to sustain these two network types, considering insights such as homophily and the tendency of triadic closure?
The first criticism we got when presenting this paper was even harsher – people did not even believe that our phenomenon would exist (i.e., that the network structure would not be in the dataset)! After all, many network scholars firmly believe that open networks are diverse, and closed networks are homogenous. Only when we started showing the descriptive statistics to the audience (a simple cross-table) did presentations start to go smoothly.
In hindsight, though, we think this network structure may be more common than even we may have thought. For example, the Danish are said to keep their childhood friends for life. That means you should have a very strongly connected network of people who over time may become quite diverse, for example in terms of their professional background. Having said that, it is true that closed-diverse and open-specialized networks are quite hard to create and maintain, as tendencies toward homophily and triadic closure promote networks that are closed-specialized. In an MBA elective on strategic networking, one of the insights Anne is trying to bring across is that it is worthwhile for individuals to reach out to “pockets of specialists” in different locations that are out of their local tightly-knit networks (i.e. to create open-specialized networks), or to introduce diverse people in your network to each other to collectively make sense of diverse, potentially disjointed perspectives (i.e. to create closed-diverse networks). This would, however, require a very active approach to networking. If we leave our networks evolve without much deliberate effort, we end up with the closed-specialized that may lock us into established ways of thinking.
Question 4. You currently focused on venture performance as the dependent variable and tested the second-order effect of investor networks on this DV. But it also seems interesting to us to test the first-order effect on the investors’ performance themselves. Can you please share some thoughts on your choice of interest here?
Great question – this is a future paper we have on our radar! Seriously, we think that venture and investor performance are inherently interconnected. In our current paper, we are looking at a deal moving from the first to the second round. Here, (first-round) investors make money by capitalizing on their shares – in the next round of investment, they may actively be selling their shares to the newly incoming investor(s). The issue here is that CrunchBase, on which we heavily relied, does not identify precisely what share of ownership an investor receives, so we cannot calculate the profit rate. The closest we can do though is show that our effects remain qualitatively unchanged when we use different dependent variables, such as whether the venture was acquired (but we cannot see either whether this is a bankrupt firm being bought for a few pennies, or a highly successful exit) or achieved an IPO (this one, at last, is clear!) – separately as well as in competing risks models.
What we are keen on identifying in future work is how the past network will influence who the next investors are: that is, which network structure in the first round of investments is more likely to bring in a “cool” new investor, and, hence, a new tie that the investor may leverage in the future?
Question 5. It has been a couple of years since the publication of this article. Would you like to share some new directions related to this paper that you may’ve been developing? Thank you!
We originally started working on this paper in 2009 – so it’s even older than the 2016 publication (and older than the 2013 Academy of Management Conference Best Paper Proceedings version). We are seeing us extend this line of work in three ways – first, we are still looking at CrunchBase as a source of data, and we are hopeful that our paper has helped in ongoing efforts to establish this dataset as “legitimate” to see it used in research more broadly (Ben Hallen and Emily Cox Pahnke at UW draw on it heavily as well, for example). In one of our current projects, we combine CrunchBase data with Twitter data to derive more information about the personalities behind the investors and start-ups. Second, we are still looking into further network-related topics, such as the one mentioned above: how do structural and relational attributes of networks help explain network dynamics? How does the bond between investors and start-ups come into place? As you pointed out, it is quite difficult for start-ups and investors alike to create open-specialized and closed-diverse networks and we wish to better understand the network behaviors and strategies that allow them to build those networks. Finally, we see us reapply and extend some of the core thinking of our paper to different contexts. For example, some of us are now looking at innovation ecosystems, or business angel networks. There, too, we find a combinatorial perspective on network structure and knowledge diversity to be extremely informative, and this view has inspired and assisted us in developing some of our more recent research questions.