American Economic Review 2012, 102(1): mpulsory Licensing: Evidence from theTrading with the Enemy Act†By Petra Moser and Alessandra Voena*Compulsory licensing allows firms in developing countries to produce foreign-owned inventions without the consent of foreign patent owners. This paper uses an exogenous event of compulsorylicensing after World War I under the Trading with the Enemy Act toexamine the effects of compulsory licensing on domestic invention.Difference-in-differences analyses of nearly 130,000 chemical inventions suggest that compulsory licensing increased domestic inventionby 20 percent. (JEL D45, L24, N42, O31, O34)Compulsory licensing allows firms in developing countries to produce foreigninventions without the consent of foreign patent owners.1 Countries such as Brazil,Thailand, and India have used the policy to procure life-saving drugs for millionsof patients and are proposing it as a means to access foreign technologies to combatclimate change.2 Opponents of compulsory licensing, however, fear that the policymay reduce access to critical innovations that are invented abroad, as it weakensincentives for foreign firms to transfer new technologies into developing countries.For example, the US pharmaceutical company Merck criticized Brazil’s licensingof its HIV drug efavirenz as an “expropriation of intellectual property” which will“hurt patients who require new life-saving therapies” (,May 7, 2007).* Moser: NBER and Department of Economics, Stanford University, 579 Serra Mall, Stanford, CA 94705(e-mail: [email protected]); Voena: Harvard Kennedy School, 79 JFK Avenue, Cambridge, MA 02138 (e-mail:alessandra [email protected]). We thank David Autor, Tim Bresnahan, Iain Cockburn, Giacomo De Giorgi,Christina Gathmann, Eric Hilt, Todd Newman, Tom Nicholas, Joerg Ohmsted, Andrea Pozzi, Mike Scherer, GavinWright, and seminar participants at Arizona, Berkeley, Chicago Booth, ISNIE, the NBER, and Stanford, for helpful comments. We owe special thanks to Ryan Lampe for helping us collect the US patent data, and KathrynSteen for sharing copies of archival records. Shaun Hayes at the American Heritage Center of the University ofWyoming helped us to access the personal papers of the Alien Property Custodian. Adam Tarhouni, Emily Rains,Leah Karlins, Marina Kutyavina, and Stephanie Lee provided excellent research assistance. Petra Moser thanksthe National Science Foundation for its financial support through grant SES0921859, and the NBER’s Group onInnovation Policy and the Economy and the Kauffman Foundation for funding the early stages of this project.Alessandra Voena acknowledges financial support through the Ely Fellowship at SIEPR and the Clayman Institutefor Gender Studies at Stanford.†To view additional materials, visit the article page at general, TRIPS Art.31 allows compulsory licenses after negotiations for voluntary licenses have failed. Incases of emergency, TRIPS allows governments to grant compulsory licenses without first trying to negotiate. TheWorld Trade Organization (WTO) Doha Declaration of 2001 emphasized developing countries’ rights to issuecompulsory licenses: “Each member has the right to grant compulsory licenses and the freedom to determine thegrounds upon which such licenses are granted” (WT/MIN(01)/DEC/1, Art. 5.b).2Thailand and Brazil have used compulsory licenses to produce antiretrovirals for AIDS patients and Indiahas indicated plans to use compulsory licensing to combat swine flu (Kremer 2002; Galvão 2002; Gostin 2006;Steinbrook 2007).396

VOL. 102 NO. 1Moser and Voena: Compulsory Licensing397Policy debates have, however, neglected an important aspect of compulsorylicensing: does compulsory licensing increase or discourage domestic invention incountries that license foreign technologies? Compulsory licensing may discouragedomestic invention if access to foreign inventions at below-market rates weakensincentives to develop alternative technologies domestically. The ability to produceforeign inventions could, however, also enable domestic firms to establish their ownindependent production, which strengthens incentives to invest in complementaryresearch and skills and creates opportunities for learning by doing (e.g., Arrow1962; Stokey 1988; Irwin and Klenow 1994).To test whether compulsory licensing increases or decreases domestic invention,we take advantage of an exogenous episode of compulsory licensing as a result ofWorld War I. On October 6, 1917, Congress passed the Trading with the EnemyAct (TWEA). Section 10 of the Act permitted US firms to violate enemy-ownedpatents if they contributed to the war effort.3 As the war dragged on, the TWEAbecame more and more punitive (Steen 2001). One week before the Armistice atCompiègne on November 11, 1918, Congress amended the TWEA to confiscate allenemy-owned patents. By February 1919, German-owned patents were systematically licensed to US firms.To measure the effects of compulsory licensing, we compare changes in the number of patents by domestic inventors across technologies that were differentiallyaffected by the TWEA. This allows us to control for alternative factors that mayhave encouraged domestic invention across chemical technologies, such as improvements in education and scientific training (e.g., Landau and Rosenberg 1992) or tariff barriers intended to protect the US chemical industry (Eichengreen 1989; Irwin1998). Technologies are measured at the level of subclasses of United States Patentand Trademark Office (USPTO) patents in organic chemistry. Chemical inventionsin all of these subclasses were affected by tariff barriers and improvements in education, but only some subclasses were affected by compulsory licensing.Three complementary variables measure compulsory licensing. A binary variable identifies subclasses that received at least one license under the TWEA. Twoadditional variables control for differences in the number and in the novelty oflicensed patents.Changes in domestic invention are measured by the number of US patents grantedto US inventors per subclass and year. To construct the data, we collected information on all 19 USPTO classes of organic chemicals that received at least one of 727compulsory licenses of enemy-owned patents under the TWEA. These 19 classesproduced a total of 128,953 patents between 1875 and 1939 and covered 7,248 subclasses; 336 of these subclasses were treated.These data reveal a substantial increase in domestic invention in subclasses thatwere affected by compulsory licensing. In subclasses that received at least onelicense, domestic inventors produced an average of 0.151 additional patents per yearafter the TWEA compared with other subclasses. This implies an increase in domestic patents of nearly 25 percent relative to an average of 0.619 patents per subclassbetween 1919 and 1939. Tests that control for the number of compulsory licenses312 USC. § 95a. Today, Cuba is the only country still affected by the TWEA.

398THE AMERICAN ECONOMIC REVIEWfebruary 2012indicate that each additional license generated 0.072 additional patents per subclassand year. In subclasses where US firms licensed patents that were 10 years younger,domestic inventors produced 0.060 additional patents per year.We also examine the timing of effects, which may help shed some light on themechanisms by which licensing encourages domestic invention. If licensing increasesdomestic invention through learning by doing, effects may take several years to materialize, as domestic firms learn to produce foreign inventions and build their own production capacities. This process might be especially slow if domestic inventors need“time to learn,” as Arora and Rosenberg (1998, p. 79) suggest to have been the casefor organic chemicals in the United States.4 In fact, our data on US patents suggestthat pre-TWEA levels of domestic invention were especially low in treated subclasses.Estimates of annual treatment effects confirm that the full impact of compulsorylicensing occurred with a lag of eight to nine years. Enemy-owned patents werelicensed from 1919 to 1926, with most licenses being granted from 1919 to 1922(Steen 2001). Although annual treatment effects become significant as early as1927, the strongest effects occur for patents that were granted after 1931. Given thatpatent grants occur two to three years after applications in our data, this implies thatthe largest effects on applications began in 1928—six to nine years after most patents had been licensed. Effects remained large and significant at nearly 60 percentadditional patents per subclass and year throughout the 1930s.One caveat with these results is that the licensing decisions of US firms may nothave been exogenous, even though the timing of the TWEA and the types of technologies that could be licensed were exogenous. Most importantly, US inventorsmay have been especially eager to license foreign inventions in subclasses where thedemand for domestically produced goods was high, so that the observed effect maybe the result of an interaction between the demand for domestic production and theability to license foreign inventions. On the other hand, the demand for licenses mayhave been highest in subclasses where levels of domestic invention were initiallylow. In those subclasses domestic invention is likely to have increased more slowlybecause US firms had to bridge a larger gap to the technological frontier before theycould patent their own inventions.To control for the potential influence of alternative factors, we subject the data to aseries of additional tests. Triple difference regressions account for unobservable characteristics that may have encouraged patenting by all non-German inventors in treatedsubclasses. Specifically, we compare changes in patenting by domestic inventors withchanges in patenting by other non-German inventors before and after the TWEA.Triple difference estimates confirm that licensing encouraged patenting by domesticinventors, even relative to other non-German inventors. An alternative test artificiallyexposes French inventors, who could not license enemy patents under the TWEA to“treatment” by compulsory licensing. In this test, compulsory licensing has no effect.Also see Haber (1971); Aftalion (2001); Mowery and Rosenberg (1998). In 1923 chemical trials during a courtcase established that a skilled US chemist could not reproduce synthetic organic chemicals based on confiscatedGerman patents: Louis Freedman, who had earned degrees from Yale and Columbia proved unable to producecincophen, a drug to treat gout (Steen 2001). Additional delays may result from incomplete information in patentdocuments. The German firm BASF, for example, withheld critical information about the Haber-Bosch processfrom its patent application and US firms took nearly a decade to replicate its process (Haynes 1945).4

VOL. 102 NO. 1Moser and Voena: Compulsory Licensing399To assess the direction and size of selection bias, we estimate intent-to-treat (ITT)and instrumental variable (IV) regressions, where the number of enemy-owned patents that US firms could have licensed under the TWEA measures the ITT and IVvariables. ITT estimates are slightly smaller than OLS estimates, while IV estimatesare somewhat larger, which indicates that selection bias (such as the concentrationof licensing in subclasses with low initial skill levels) may indeed lead us to underestimate the true effects of compulsory licensing.Additional robustness checks control for preexisting time trends and variationabove the subclass level, regressions on a restricted sample of primary subclasses,and regressions for changes in patenting within a specific chemical (indigo dyes).In a final section of the paper, we perform a firm-level analysis that distinguishesthe effects of patents that were licensed to a specific US firm (Du Pont) from theeffects of patents that were licensed to other firms. Effects of own licenses are morelikely to result from learning that occurs when a firm produces foreign inventions,while other licenses capture factors that benefit the industry more broadly, such asimprovements in education. Our results suggest that both types of mechanisms wereimportant, but effects of own licenses were roughly four times as large as effects ofother firms’ licenses.The remainder of this paper is structured as follows. Section I summarizes basicfeatures of the TWEA. Section II presents our empirical strategy. Section III detailsthe data collection and discusses potential sources of bias and measurement error.Section IV presents estimation results, Section V presents robustness checks, andSection VI summarizes results of our firm-level analysis. Section VII concludes.I. The TWEA as a Natural Experiment of Compulsory LicensingCreated by an Act of Congress on October 6, 1917, the TWEA was intended to“dislodge the hostile Hun within our gates” (Alien Property Custodian 1919, p.17) to destroy “Germany’s great industrial army on American soil,” its “spy centers,” and “nests of sedition” (Alien Property Custodian 1919, p. 14). To this end,the TWEA placed all enemy property “beyond the control of influence of its former owners, where it cannot eventually yield aid or comfort to the enemy” (AlienProperty Custodian 1919, p. 13).5On March 28, 1918, the TWEA was amended to give the Custodian the powerto sell enemy property, including all enemy-owned patents, “as though he were theowner thereof” (Alien Property Custodian 1919, p. 22). Thus, the Alien PropertyCustodian began to appropriate any patent owned by “enemy persons” andcorporations doing business in Germany, Austria-Hungary, Bulgaria, and Turkey,as well as the occupied parts of Belgium, France, Russia, and the Balkans (AlienProperty Custodian 1919, p. 7), administering these properties as a trust.5The destruction of German property was al