In prior posts about my “Upside-down VCs” research, I described my own experiences in venture capital that led me to question VCs’ choices about their internal org structures, VCs’ disincentives regarding hiring junior people, and the comments posted regarding the structural and hiring differences between early-stage and later-stage VC firms.
In this post, I’ll outline two remaining findings from the paper: (1.) other factors that affect internal org structures and (2.) tests of performance implications.
Other Factors Affecting Internal Structure
As described before, in tests of my longitudinal data on 317 American VC firms over 4 years (total number of observations: 1,033 firm-years), the stage of company in which the VC firm tended to invest — i.e., whether the VC firm was an “early-stage investor” versus a “later-stage investor” — was the strongest factor affecting the VC firm’s organizational structure, regarding both statistical significance and practical significance.
However, there were three other main effects that were significant (at the p<.05 level or better):
- Multiple US offices — VC firms that had more than one domestic office had hired more junior people than VC firms that still had a single domestic office. (In contrast, a significant but weaker result showed that the more foreign offices a VC firm had opened, the more top-heavy the firm’s internal structure.)
- Multiple industry sectors — VC firms that invested in more than one industry sector had hired fewer junior people than VC firms that focused on a single sector.
- “Fit around a table” — There was a big difference between VC firms that had fewer than 8-10 people and those that had more than 8-10 people. My field research suggested that in firms where everyone could “fit around the table,” the approaches for discussing potential and existing investments were based much more on face-to-face communication than in firms that were too big for everyone to discuss issues face to face, and that this had implications for how they structured themselves internally. (One VC vividly compared “around the table” firms to a team of Navy SEALS, which has to stay small — 7 people — in order to remain tight-knit and “lethal.”) Quantitative tests of my dataset confirmed that firms where everyone could fit around the table were more “upside down” (had fewer junior people) than firms that had grown “beyond the table,” and that most of this change occurred at the 8-10 person breakpoint.
These results hold even after controlling for differences in the firms’ number of investments per year, firm ages, total active capital, and the year in which the data were collected.
The most important implication of the prior posts is that a VC firm’s internal organizational structure should affect the firm’s investment performance. However, the impact on performance should differ between early-stage investors and later-stage investors. In the paper, the diagram summarizing this is as follows:
In short, H2 (“Hypothesis 2″) is that for VCs who invest in early-stage companies, firms that are more “upside-down” (i.e., have fewer junior people) should have better performance than firms that are more “pyramidal” (i.e., have more junior people).
The opposite should be true for the later-stage investors of H3 (“Hypothesis 3″): for VCs who invest in later-stage companies, firms that are more “upside-down” will have worse performance than firms that are more “pyramidal.”
Data note: Testing these two hypotheses was not possible when I first did this research in 2001, because there was no systematic and reliable source of data on VC-fund performance. However, after 2001, as a result of legal actions, some of the biggest limited partners (e.g., CalPERS, CalSTRS, UTIMCO, University of Michigan) began having to publish the returns of the VC firms in which they had invested. A couple of years ago, I collected the existing 9 LP releases and matched them up with my VC internal-organization data and found that I had performance data for 121 funds from 97 of the firms in my dataset. It is important to note that the sample size in the performance models (n=121) is much smaller than in the strategy-and-structure models (n=1,033).
Tests of these hypotheses strongly supported the two hypotheses, with the early-stage hypothesis (H2) supported at the p<.01 level of significance, and the later-stage hypothesis (H3) supported at the p<.005 level of significance.