Marginal Patents and the Supply of Ideas
2.3 Results and Discussion
2.3.1 Patent Counts
Figure 2.2 shows the time series of inventor productivity, split by whether the original examiner of the patent participates in the opposition division. Absence of the examiner is associated with a higher likelihood of invalidation. In line with our identification assumption, in the years leading up to the outcome of opposition, we find no visible differences in the average number of patent applications. After the outcome, however, inventors with participating examiners are relatively more productive. This effect starts around three years after the opposition outcome.
To assess whether these reduced-form effects are statistically significantly different from zero and whether they are robust to the inclusion of control variables and fixed effects, we repeat this analysis in an event-study framework. Figure 2.3 shows the time-varying reduced form impact of examiner participation on subsequent patent filings. In line with the prior graph, “lucky” inventors whose original examiner takes part in the opposition division sub-sequently file more patent applications. The effect materializes after around three years, as before. In line with the identification assumption, there are no statistically significant
differ-Figure 2.3: Reduced form effect of examiner participation on the number of applications
10 8 6 4 2 0 2 4 6 8 10
Year relative to opposition outcome 0.06
0.04 0.02 0.00 0.02 0.04 0.06
Reduced form coeff: Participation x Year
Sign. levels p< 0.10 p< 0.05 p< 0.01 Parameter estimate
(95% confidence interval)
Notes: Coefficient estimates for years relative to the opposition outcome from a reduced form regression on (inventor, year relative to opposition outcome)-level of the number of applications on the examiner partic-ipation instrumental variable. The dependent variable yi t = Napp,i t counts the number of applications on DOCDB family level which inventori has filed in period t. The corresponding specification is given by yi t =
∑10
τ=−10βτ1(Exam parti)1(t =τ) +at+bt−taf+ci+εi t. i andt are the indices for the inventor and the year relative to opposition outcome, respectively; fixed effects are described in the main text.τ=−10, . . . , 10 denotes years relative to opposition outcome. Error bars indicate the respective coefficient’s 95% confidence interval. Stars at the bottom of the figure indicate the significance levels of the coefficients.
ences in patent filings before the decision of the opposition division.
Figure 2.4 shows the instrumented yearly effect of patent invalidation on the number of applications that inventors file. Again, there is no statistically significant difference between inventors in the treatment and in the control group before the opposition outcome. In response to the outcome, inventors whose patent was invalidated due to the absence of the original examiner in the opposition division file significantly fewer patent applications. The effect starts around three years after the opposition outcome and is statistically significantly different from zero in most years. In the appendix, Figure B.1 shows the instrumented impact of patent invalidation on the likelihood of filing for a patent. In line with our identification assumption, there is no differential application propensity in the years prior to the outcome of opposition.
Yet, after an invalidation, inventors are significantly less inclined to file for a patent, with the effects being most pronounced around four years after the opposition outcome.
Table 2.3 displays the regression results of our preferred specification. Without instrument-ing the invalidation decision, Column (1) shows the partial correlation of patent invalidation and the number of subsequent patent applications. The inventor fixed effects regression
re-Figure 2.4: Effect of invalidation on the number of applications
10 8 6 4 2 0 2 4 6 8 10
Year relative to opposition outcome 1.0
0.5 0.0 0.5 1.0
IV FE coeff: Invalid x Year
Sign. levels p< 0.10 p< 0.05 p< 0.01
Parameter estimate (95% confidence interval)
Notes:Coefficient estimates of invalidation for years relative to opposition outcome from an instrumental variable (2SLS) fixed effects regression on (inventor, year relative to opposition outcome)-level using Schaffer (2010). The dependent variableyi t=Napp,i tcounts the number of applications on DOCDB family level which inventorihas filed in periodt. The corresponding specification is given byyi t=∑10
τ=−10βτ1(Invalidatedi)1(t=τ)+at+bt−taf+ci+ εi t.iandtare the indices for the inventor and the year relative to opposition outcome, respectively; fixed effects are described in the main text. The interactions are instrumented withzi,tτ =1(Examiner participationi)1(t=τ), whereτ=−10, . . . , 10 denotes years relative to opposition outcome. Error bars indicate the respective coefficient’s 95% confidence interval. Stars at the bottom of the figure indicate the significance levels of the coefficients. We find similar results for a dummy dependent variableyi t=1(NApp,i t>0), indicating whether inventorihas filed a patent application in periodt(see Figure B.1 in the appendix).
turns a negative, significant coefficient of patent invalidation. Because of the potential endo-geneity of the invalidation decision, we use the examiner participation instrumental variable in all subsequent columns. In a first step, Column (2) shows the reduced form coefficient. The presence the original examiner (lower likelihood of invalidation) increases subsequent patent applications. Column (3) presents the instrumented coefficient of invalidation. It shows that the magnitude of the effect is substantial: On average, the local average treatment effect im-plies that inventors file half a patent less per year. The sizeable difference to the coefficient in Column (1) indicates that an important source of endogeneity is time-varying and cannot be controlled for by individual fixed effects. This is in line with the findings of Galasso and Schankerman (2015, 2018). Column (4) shows that our results are robust to using the log number of applications to account for the skewness of the dependent variable. Following in-validation due to examiner (non-)participation, inventors file around 20% fewer patents. To get a sense of whether the productivity effects are driven by the extensive or the intensive margin, Column (5) shows the effect on the probability of filing a patent application at all in
Table2.3:Effectofinvalidation:Numberofapplications (1)(2)(3)(4)(5)(6)(7) EstimationmethodFEFEIVFEIVFEIVFEIVFEIVFE DependentvariableNappNappNapplog(1+Napp)1appNnat appNWO app ApplicationauthorityEPEPEPEPEPNationalWIPO 1(Invalidated)×1(Post)−0.042∗∗∗ −0.515∗∗∗ −0.204∗∗∗ −0.151∗∗∗ −0.0010.082 (0.008)(0.150)(0.049)(0.038)(0.103)(0.104) 1(Exampart)×1(Post)0.031∗∗∗ (0.008) Yeareffects(reltooppo)Yes∗∗∗Yes∗∗∗Yes∗∗∗Yes∗∗∗Yes∗∗∗Yes∗∗∗Yes∗∗∗ Yeareffects(reltoappl)Yes∗∗∗ Yes∗∗∗ Yes∗∗∗ Yes∗∗∗ Yes∗∗∗ Yes∗∗∗ Yes∗∗∗ Underidentificationtest76.976.976.976.976.9 Weakidentificationtest77.577.577.577.577.5 Numberofoppositions29,00929,00929,00929,00929,00929,00929,009 Numberofinventors65,41565,41565,41565,41565,41565,41565,415 Observations1,276,7291,276,7291,276,7291,276,7291,276,7291,276,7291,276,729 Standarderrorsclusteredattheoppositionlevelinparentheses ∗p<0.1,∗∗p<0.05,∗∗∗p<0.01 FixedEffects(Column1),reducedformfixedeffects(Column2)andinstrumentalvariable(2SLS)fixedeffects(Columns3–7)regressionson(inventor,yearrelative oppositionoutcome)-level.Columns(1)–(3)usedifferentspecificationsforthesamedependentvariable,thenumberofapplications.Columns(3)–(5)usethesameIVFE fordifferentfunctionalformsofthedependentvariable:alinear-,alog-andanindicatorvariablespecification.Totestwhetherthereductionisdrivenbyashift nationalortransnationalpatenting,Columns(6)and(7)displaytheeffectonthenumberofpatentfamilies,whichdonotcontainanEPapplication.First,inColumn onlypatentfamiliesarecounted,whichcontainanationalapplicationinaEuropeancountry,butdonotcontainEPorWIPOapplications.Second,inColumn(7),only familiesarecounted,whichcontainatleastoneWIPOapplication,butnoEPapplication.ForColumns(6)and(7),wehaveusedthesamesetofinventorsasinthe columns.Ifwerestrictthesampletoinventorswhohaveatleastone“national”oratleastone“WIPO”patentfamilyinoursamplingperiod,wefindqualitatively results.AllvariablesarecountedontheDOCDBfamilylevel.Thepostperiodisdefinedasthetimewindowfrom0to10yearsafteropposition.Standarderrors clusteredattheoppositionlevel.Theunderidentificationandweakidentificationtestsaretheheteroskedasticity-robustKleibergenandPaap(2006)rkLMandWaldF respectively,asreportedbythextivreg2Statacommand(Schaffer,2010).ForananalogoustableusingtheMorrisonetal.(2017)inventordisambiguation,see ableB.6intheappendix.ForanalogoustablesonthesubsamplesofEuropeanandforeigninventors,seeTablesB.7andB.8.
a given year. Having a patent invalidated in opposition reduces the likelihood of subsequently filing a patent by 15 percentage points.
In principle, the decrease in patent applications after invalidation could be independent of innovative activities and merely reflect a change in filing strategies. First, inventors, their firms, or their patent attorneys could steer patenting away from the EPO and instead patent directly at the desired national patent offices, avoiding a potential centralized opposition pro-cedure. To investigate this channel, Column (6) uses the number of patent families with (Eu-ropean) national patent applications as the dependent variable (not counting patent families that contain EP or WIPO applications). We find no change in national patenting. Alternatively, innovators could substitute EP patenting with WIPO’s centralized application procedure. To investigate this possibility, the dependent variable in Column (7) counts patent families con-taining a WIPO, but no EP application. While the point estimate is positive, it is not significantly different from zero. Besides, its magnitude is substantially smaller than our main effects. Our results therefore reflect an actual decrease in patent filings rather than a shift to substitute patent authorities. Second, inventors or their firms could be more selective in choosing inven-tions for patenting and could be more reluctant to split connected inveninven-tions into several ap-plications. In either case, one would expect an immediate change in filing behavior. However, Figure 2.4 shows that the invalidation effect is most pronounced 3-5 years after the opposition outcome. This could reflect the delayed effect of a real change in innovative activities.
In the appendix, we show that our main productivity results are robust to excluding outliers (such as the top 5% of inventors with respect to prior filings and technology areas) and to restricting the sample to inventors who patent both before and after the opposition outcome (Table B.5). In Table B.6, we also show that our results are unaffected by using the alternative inventor disambiguation by Morrison et al. (2017). Finally, in Tables B.7 and B.8 we show that findings are very similar for European and non-European inventors.
Given that we instrument invalidation by the presence of the original examiner in the op-position division, our estimates reflect local average treatment effects. To explore whether compliers differ from the overall population of patents in opposition, in Appendix B.2 we fol-low Angrist and Pischke (2009) and document the relative incidence of certain applicant and inventor characteristics among compliers. Note that in our context, compliers are inventors whose application was invalidated because the original patent examiner did not participate in the opposition division. Table B.2 shows the complier share, which lies at around 7% on average. Table B.3 examines the characteristics of complier applications relative to the gen-eral population of patents in opposition. On average, applications whose opposition outcome changes with the examiner’s presence in the opposition division are less likely to have more than two inventors and to receive above median citations. Their family size is larger, but this is at the margin of statistical significance. Table B.4 conducts an analogous comparison for inventor characteristics. Inventors of complier patents are more likely to have below median tenure and to have filed patents in a lower number of technology areas before the invalidation.
They have also filed fewer patents before the opposition, but the difference is insignificant. In summary, however, complier patents do not differ substantially from the average patent in opposition.