[Patents] PATNEWS: A critique of Bessen/Maskin's innovation and patents paper

[Gregory Aharonian srctran@world.std.com]

!20010131  A critique of Bessen/Maskin's innovation and patents paper

    There is a debate going on over in Europe on the patenting of software.
Some of those fighting software patents in Europe (and wanting to keep the
current minimalist prohibition, 52.2.c in the law) are trying to raise the
issue of the inherent evilness and utter reprehensibility of software
patents. One weapon they use is to cite a paper published a year ago by
two economists in the Boston area, Eric Maskin of MIT's Economics Dept.,
and James Bessen, a consultant in the Boston area.  I feel that it is wrong
to use this paper in a policy argument because a) the paper's models are
unrealistic, and b) their use of correlating data is questionable. What
follows are comments on these two concerns.


    They propose a simple economic model to argue that in sequential
innovation industries like software, semiconductors and computers, that
patent protection may reduce overall innovation and social welfare.  The
paper is titled "Sequential Innovation, Patents and Imitation", January
2000 (it's on the Web).  Here is the abstract:

        How could such industries as software, semiconductors, and
        computers have been so innovative despite historically weak
        patent protection?  We argue that if innovation is both
        sequential and complementary - as it certainly has been in
        those industries - competition can increase firms' future
        profits thus offsetting short-term dissipation of rents.
        A simple model also shows that in such a dynamic industry,
        patent protection may reduce overall innovation and social
        welfare.  The natural experiment that occurred when patent
        protection was extended to software in the 1980's provides
        a test of this model.  Standard arguments would predict
        that R&D intensity and productivity should have increased
        among patenting firms.  Consistent with our model, however,
        these increases did not occur.  Other evidence supporting
        our model includes a distinctive pattern of cross-licensing
        in these industries and a positive relationship between
        rates of innovation and firm entry.


The problem is that these models are too simplistic to represent the
complicated dynamics of the software world, and the economic data they
use represents only part of the software industry.  To then use this
paper to argue against software patents is intellectually dishonest.

For example, consider the introduction to the static version of their
model:

        We consider an industry consisting of two (ex ante symmetric)
        firms.  Each firm can undertake R&D to discover and develop
        an innovation with expected (social) value V.  The cost of
        R&D is C, and if a single firm incurs that cost, the
        possibility of successful innovation is P.  We assume that
        if a firm is not copied, it can capture the social value of
        its innovation.  However, the other firm, unless prevented by
        patent protection, can costlessly develop an imitation.  And 
        if it does so, the innovating firm obtains only a fraction S
        (S < 1/2) of the value of V (to simplify matters, we assume
        that the imitating firm captures this same faction s).

and a dynamic version

        Let us now enrich the model to accommodate a sequence of
        potential innovations, each of which builds on its immediate
        predecessor.  More specifically, we will suppose that, for a
        firm to have a realistic chance of developing the innovation
        of the next generation, it must have market experience with
        that of the current generation.

Anyone recognize the US software R&D community (companies, universities,
government labs, individuals communicating over networks) in these models?
Additionally, marketing plays a big role in driving R&D (not only for
software, but pharmaceuticals as well).  Finally, while patent protection
is a useful tool, I don't think it is a strategic tool that shapes R&D
along the lines of these two models.   In short, they propose two models
fun to explore mathematically, but that's it.  Thus, as a message to the
Europeans, one has to be really careful in applying conclusions from the
analysis of such models to the real world for policy purposes.

What is of additional concern to me is one such application of their
models to understand the US software patenting scene.  They create an
either-or explanation for a situation which has many potential
explanations. Arguing about software, they write:

        A series of court decisions in the early 1980's had the effect
        of extending patent protection to many software ideas.
        Consequently the number of patents issued annually covering
        software grew exponentially from the mid-80's to about 7,000
        in 1995.  Within the software industry, this has sometimes
        been described as a case of "fixing what ain't broke".
        Advocates counter, arguing along the lines of the static
        model, that increased patent protection should increase 
        software innovativeness.

        If the static model is correct, then the extension of patent
        protection should have produced a sharp increase in R&D
        spending among those firms and industries applying for
        patents.  This should have subsequently been followed by an
        increase in productivity growth.  The changes should be
        measurable and large after controlling for other, possibly
        offsetting charges.

        [.....]

        To summarize, if the state model holds, relative R&D spending
        should have increased sharply, followed by productivity.  If
        the dynamic model applies, relative R&D spending and productivity
        would not have increased and might exhibit a mild decrease.

First, the court decisions favoring software patenting only partially
explains the rise in software patent applications.  The authors ignore
the parallel set of court decisions that weakened copyright protection
for software, which drove many to the patenting side.  The investment
climate also became very favorable to the software industry in the 1990s,
which also partially drove the increase in software patent applications.

Next, they analyze these models in light of NSD and NBER R&D economic
data. They conclude

        In summary, the initial high level of innovation in software,
        mixed industry support for patents, and an apparent gradual
        slowdown of R&D intensity all suggest that the applicability
        of the dynamic model.  This conclusion is also supported by
        the distinctive pattern of cross-licensing in the semiconductor
        and computer industries and by the more general positive
        relationship between innovation and firm entry.  All of this
        is difficult to reconcile with the traditional static model
        of intellectual property.

So using a simplistic two firm semi-dynamic model, and a subset of all
of the economic data relating to these industries, they make some
conclusions.  Interesting, but how relevant?  For example, consider
the phrase in their abstract:

        A simple model also shows that in such a dynamic industry,
        patent protection may reduce overall innovation and social
        welfare.

They measure innovation and social welfare for the software industry
using the NSF and NBER data from the 1974 to 1994 time period, and 
conclude "... in a dynamic world, firms may have plenty of incentive
to innovate without patents and patents may constrict complementary
innovation."  Constrict?

Whoopee.  Given all of these simplifications and "may" qualifiers, it is
dubious to bad logic for anyone to then argue that the Bessen/Maskin
article is grounds to ban software patents because they are bad.

In the real world, is there any evidence that the explosion in (bad)
software patents has stifled innovation and social welfare?  I offer
some data to show that you can't answer this question YES or NO.  The
data I offer is the new number of papers in the INSPEC engineering
abstracts database that have a keyword entry "SOFTWARE".  INSPEC has
abstracts to most of the journals and conferences where new refereed
papers are published, papers that come out of the many corporate,
academic and government R&D environments where software plays a role.


                    YEAR   NEW "SOFTWARE" PAPERS IN INSPEC
                    1999                  19,480
                    1998                  20,143
                    1997                  20,187
                    1996                  19,743
                    1995                  17,181
                    1994                  17,632
                    1993                  14,450
                    1992                  13,336
                    1991                  13,088
                    1989                  14,797
                    1988                  17,159
                    1987                  15,538
                    1986                  13,295
                    1985                  12,139
                    1984                  13,615
                    1983                  10,615
                    1982                   6,799
                    1981                   6,615
                    1980                   4,943
                    1979                   4,382
                    1978                   3,960
                    1977                   2,764
                    1976                   2,370

What we see is a gradually increasing amount of new materials published
about software over a time period when the number of issued software
patents went from a few hundred a year to a few tens of thousands a year.
During the 1990s, each year saw about a 8% growth in the number of new
software publications.

Hard to argue then that innovation and the public's desire for publication
of new ideas, has suffered at all.  Heck, as someone who has to keep
track of all of these materials for prior art searching purposes, I wish
software patents were stifling things so I would have less to track :-)

Looking at all sorts of data, it is hard for me to argue anything but that
software is just another technology with patenting problems no different
than other technologies (though I do think the quality of software patents
is lower).  While the Bessen/Maskin article is suggestive, it is too
simplistic to use as a jsutification for any policy decision.

Greg Aharonian