Article de Emilio Raiteri.
Travail qui sera présenté au sein du groupe de travail R&D et Innovation défense, le jeudi 13 novembre 2014.
The idea that demand might be a major source of innovation dates back to the contribution of
Schmookler(1962, 1966) and Kaldor(1966). Despite the slowdown in the study of this relation due to
the critics by Mowery and Rosenberg(1979), the demand side approach has recently regained attention.
With the resurrection of the demand side also the debate on the role of public demand has been re-
vitalized. Innovative public procurement has been increasingly considered as a form of public support
to private innovation activities both from innovation scholars and policy makers (Edler and Georghiou
2007), grounding the need for demand oriented innovation policy.
Economic historians suggested an even more fundamental role for public procurement in setting the
pace of technological change, acknowledging government demand as a crucial factor in developing some
of the most in uential technologies of the 20th century. In particular Ruttan (2006) and Mowery (2008,
2011) report how U.S. military and aerospace related procurement had a major impact for the emergence
and di usion of so called general purpose technologies (GPT) such as semiconductors, computing and
the internet. GPTs are usually considered as engines of economic growth, fostering widespread produc-
tivity gains through massive relocation and reorganization of the economic activity. According to the
GPT literature this potential growth is achieved only if the virtuous cycle of innovation complementar-
ities is triggered between the sector that unveils the new technology (upstream sector) and the sectors
applying the new technology (downstream). Technological levels of the upstream and downstream sec-
tors are hence strategic complements and widespread di usion stems from the coordination of beliefs
between GPT producer and application sectors (Bresnahan and Trajtenberg, 1995). Coordination fail-
ures and larger uncertainty tied to drastic innovations may therefore provide little market incentives for
adoption in the downstream sectors, potentially leaving an economy locked-in on inferior technological
trajectories . Bresnahan and Trajtenberg (1995) already suggested that public procurement may play a
very important role to overcome this potential market failure, injecting the virtuous cycle of innovation
Despite the economic historians’ contributions and Bresnahan and Trajtenberg’s suggestion, no em-
pirical work has so far tried to nd evidence of the link between public procurement and technological
generality. This paper tries to ll this gap. Following the intuition provided by Bresnahan and Trajten-
berg (1995) and conceiving the arrival of a GPT “as a process unfolding in time rather than a single
homogeneous shock” (Cantner and Vannuccini, 2012), I surmise that procurement might represent one of
the most important element in creating the right soil to “cultivate” a technology that may (or may not)
have the potential to reach high levels of pervasiveness. I will hence hypothesize that public procurement
can raise the degree of generality of upstream technologies, triggering innovation complementarities in
the downstream sectors for which the market is not providing su cient incentives.
To empirically test this hypothesis I make use of patent data and in particular of patent citations.
Citations can be considered as ‘paper trail’ of the linkages between an innovation and its technological
‘antecedents’ and ‘descendants’ (Trajtenberg et al. 1997). This feature allows to: i) use patent citations
to identify the connection between innovations related to public procurement and their antecedents; ii)
measure generality of patents looking at the extent to which the follow-up technical advances are spread
across di erent technological elds through a Generality Index (Trajtenberg et al. 1997; Hall, 2002) 1.
On the basis of this 2 consideration the hypothesis stated above can be expressed in a more formal way,
I will in fact hypothesize that receiving a citation from a patent related to public procurement raises the
generality level of the cited patent.
To perform the analysis we exploit data from 3 di erent sources: i) NBER patent data project that
collects data for patents granted by the USPTO in the period 1976-20062, together with citations data; ii)
Federal Procurement Data System (FPDS),3 that includes several information for each Federal contract
awarded from 2000 onwards; iii) the Compustat North America Database which gathers nancial and
market information on public companies in the U.S. .
I hence design a quasi experiment in which we compare the change in the generality level (measured
through the Generality Index) at two di erent points in time, 1999 and 2006, between treated and
a control patents, whose application date falls in the period 1993-1997 and who received at least 10
forward citations in 2006. Public procurement is the treatment variable and, in particular, a patent is
put into the treatment group if it receives a citation from a patent related to public procurement in scal
In order to deem a patent as ‘related to public procurement’ I rely on 2 considerations: i) U.S.
laws: the regulation states that a patent for an invention made by a contractor in the performance of
work under a government contract should include a ‘Government interest statement’. ii) I Aggregate
procurement contracts data for scal year 2000 at the rm level and then match them through entities
names with USPTO patents whose priority date falls in year 1999-2000 (NBER- patent data project).
A patent is then considered as ‘related to public procurement’ if it belongs to a rm who won at least a
procurement contract in scal year 2000 and if it includes the ‘Government interest statement’.
Taking simple di erence in averages of the change in the Generality Index among 1999 and 2006
between the treated and control group of patents would lead to biased results due to multiple endogeneity
issues (mainly selection bias). To mitigate the bias we hence use as control patents only patents that
are similar to the ones in the treated group along several dimensions (as it was done by Czarnitki et.
al., 2011 and Fier and Pyka 2012). I therefore adopt the conditional di erence-in-di erences approach
(Heckman et al., 1998). As a rst step I use propensity score matching method to tackle the selection
on observables problem, estimating the propensity score through a probit regression on several patents’
and assignees’ characteristics. Once that the matching is implemented I use di erence-in-di erence to
eliminate potential biases due to selection on unobservables.
Preliminary results of the average treatment e ect provided by the CDiD approach suggest a positive
and signi cant impact of innovative public procurement upon the generality of a patent. In particular,
on average receiving a citation by a patent related to public procurement raises the Generality Index
of a 4%, con rming the initial hypothesis. Public demand seems hence to be of crucial importance
in increasing the pervasiveness of a technology, calling for the need of Schumpeterian demand policies