Changes

====Guzman and Stern====

We use data on venture capital-backed startups as their venture capital investment provides a direct performance measure to identify their clusters. However, other entrepreneurial firms may have clusters that influence clusters of venture-backed firms. In particular, pre-venture capital startups, firms that have experienced an acquisition or an initial public offering, and high-growth startups that do not raise venture capital (see Guzman and Stern, 2015, and related papers) may all affect the agglomeration of venture-backed startups. We leave it to future research to examine the relationships between clusters of different types of startups.

We use data on venture capital-backed startups as their venture capital investment provides a direct performance measure to identify their clusters. However, we expect that other entrepreneurial firms, such as those identified from business registration data in Guzman and Stern (2015), may have a second-order effect on the clustering of venture-backed firms.

Guzman and Stern (2015) use business registration to find non-venture-backed startups. We do not use this data as it does not have a performance measure to demonstrate the representative layer's selection. However, understanding the relationships between clusters of different types of startups is an exciting topic for future research.

Bibtex:

=====Previous attempts to include Guzman and Stern=====

Finally, Guzman and Stern (2015a,b) and related papers use business registration data to suggest that non-venture high-growth, high-technology startups outnumber their venturebacked counterparts by around three or four to one. These papers map and analyze nonventure startup activity within states, cities, zip codes, and for an individual street. A natural extension to this paper would consider agglomeration economies of non-venture startups, as well as their effects on the agglomeration of venture-backed firms.

Catalini et al. (2019) suggest that there are high-growth, high-technology startups that are not supported by venture capital. These firms may have been turned down by venture capitalists, may never approach venture capitalists, or may not have had their venture investment recorded at the time the authors conducted their study. Regardless, according to Catalini et al. (2019), non-venture HGHT startups outnumber venture-backed firms by around three or four to one. One obvious question is then whether they enhance or detract from agglomeration economies among venture-backed startups.

=====Comments from the Associate Editor and two reviewers at Management Science=====

Multiple referees recommend supplementing your analyses with additional datasets. An expensive but informative path would be the National Establishment Time Series, which has street addresses for every business that reports data to Dun & Bradstreet; this would solve your longitudinal address problem. Alternatively, at no cost you can take advantage of Guzman & Stern’s Startup Cartography project data, which provides counts of the high-potential startups you are probably trying to capture with VX (and also see high-potential startups that don’t raise VC).

First, the analysis only considers startup ventures which have already received venture funding—an important component of the ecosystem, for sure, but not necessarily the most important for agglomeration purposes. This results in an incomplete and possibly distorted picture of what agglomeration actually means. if a district has 50 startups that have not yet received VC funding and one that has, would that qualify as agglomeration? And what about companies that have been acquired or have gone public? At a minimum, a more accurate count of active ventures would be required to draw any legitimate conclusions about the effects of agglomeration, along the lines of what Andrews, Fazio, Guzman, Liu, & Stern (2019) have done with the Startup Cartography Project, especially if it is true that “non-venture high-growth, high-technology startups outnumber their venture-backed counterparts by around three or four to one” (p. 5).

===Back and Forth===