Compulsory LicensingEvidence from the Trading with the

Enemy Act

Petra Moser and Alessandra Voena, American Economic Review, Vol. 102,

No. 1, February 2012, pp. 396-427

What is compulsory licensing?• License domestic patents by foreign firms to domestic

firms without the consent of foreign patent owners• Permissible under TRIPS and WTO Doha 2001• Delivers life-saving drugs to millions of patients

– Kremer 2002, Galvão 2002, Gostin 2006– HIV, Malaria, tuberculosis, influenza, cancer– Thailand 2007, Brazil 2008 issue CL for Merck’s HIV drug

efavirenz, Taiwan 2005 issue of Tamiflu for swine flu• But may reduce access to foreign inventions

– Weakens property rights of foreign inventors – Discourages technology transfer and foreign invention

• What are the effects on domestic invention?

What are the effects on domestic invention in the licensing country?

• Access to foreign inventions at below market rates may weaken incentives to invest in domestic invention

• Ability to produce foreign inventions may encourage domestic invention– Opportunities for learning by doing (Arrow 1962, Irwin

and Klenow 1994)– Incentives to invest in skills and education (Landau

and Rosenberg 1992)

Compulsory licensing after WWI

• Trading with the Enemy Act (TWEA)– Passed by Congress on October 6, 1917

• Confiscate all enemy-owned property– Destroy “Germany’s great industrial army on

American soil,” “nests of sedition”– “dislodging the hostile Hun within our gates”

• By February 1919, TWEA administers all German property in the United States– Enemy-owned patents licensed to U.S. firms 1919-

1926– Most patents licensed 1919-1921

Compulsory licensing lead to a substantial increase in domestic invention

• Compare U.S. patents by U.S. inventors before and after the TWEA in chemical technologies where U.S. firms received a license compared with other technologies

• Compulsory licensing increases domestic invention by a minimum of 20 percent– First increases after 3 to 5 years– Strongest effects after 7 to 10 years

• Robust to controls for intensity of treatment, triple differences, placebo tests for other non-German inventors, intent to treat, and instrumental variables

Robustness checks• Subclass-specific time trends, treatment-specific time

trends, controlling for broad technology areas, block bootstraps, placebo

• Restricting the sample to primary subclasses, subclasses that existed prior to the TWEA, subclass-specific time trends, block-bootstrap

• Control for tariffs, demand shocks• Learning effects are strongest for firm’s own patents (4x)• All results robust to dropping largest firms• Significant overall effects on production and scientific

citations

Compulsory licenses of enemy-owned patents

• By February 1919, TWEA administers all German property in the United States

• Patents licensed to U.S. firms 1919-1926

$4.7 billion in 2008 using a GDP deflator$88 billion in 2008 using share of GDP

Royalties per U.S. firm

• Records of Chemical Foundation• 973 licenses• License number, firm, product name, royalty

$-

$500,000

$1,000,000

$1,500,000

$2,000,000

$2,500,000

$3,000,000

$3,500,000

$4,000,000

$4,500,000

0 100 200 300 400 500 600 700 800 900 1000License number

Roy

altie

s (in

rea

l 200

8 U

S $)

Ammonin, Atmospheric Nitrogen Corp.

Dye - Indanthrene Blue and Yellow, E. I. duPont de Nemours & Co.

Dye - Olive Green Vat, The Newport Chemical Corp.

Dyes - Hydron, E. I. duPont de Nemours & Co.

Naphthol A. S. Dyes, Consolidated Color & Chem. Co.

Royalties (in real $)

License number License Modified spelling Product Royalty paid1 E. R. Squibb & Sons Neo-Arsphenamine 41,477.40$ 2 E. R. Squibb & Sons Arsphenamine 7,652.80$ 3 E. R. Squibb & Sons Oxyarylarsino Acids4 E. R. Squibb & Sons Medical Preparations5 E. R. Squibb & Sons Medical Preparations6 Dye Products & Chemical Co. Salicine Black 1,492.07$ 7 Calco Chemical Co. Cinchophen 27,806.14$ 8 John Campbell Co. John Campbell & Co. Hydron Blue Dye9 John Campbell Co. John Campbell & Co. Hydron Blue Dye

10 Radium Luminous Material Co. Radium Luminous Material Corp. Anticathode11 Mansfield Co. Reinforced Paper 6,731.85$

Was the TWEA exogenous?• Timing is exogenous• Confiscated technologies are exogenous

– U.S. confiscates all enemy-owned property• Licensing may not be exogenous

– More likely to license in areas where demand for domestic inventions is high

– More likely to license where U.S. invention is weak• U.S. can only license patents of enemies

– Germany and Austria– Not France, Britain, or Switzerland– French firms cannot license enemy-owned patents

Empirical strategy• Difference in differences

– Compare annual U.S. patents by U.S. inventors before and after the TWEA

– Compare technological subfields subject to the same shocks except for compulsory licensing

– Unit of observation: annual patents by U.S. inventors per USPTO subclass, 1875 – 1939

• Control for number and age of patents• Year-specific treatment effects• Triple differences, placebo treatment on French

inventors, intent to treat, instrumental variables• Control for demand effects• Control for time trends, block bootstrap, etc.

Main equation

• Outcome – U.S. patents by U.S. inventors per subclass and year– Include both primary and secondary subclasses

• Measures of treatment (TREATc ∙ postTWEAt )– No treatment until 1918– For treated subclasses, treatment at 1919 level from 1919

onwards• Patents by foreign inventors (Z)• Year fixed-effects (dt)• Subclass fixed-effects (fc)

Measure compulsory licensing bylicensed enemy patents

• Subclass is treated if it received at least one license under the TWEA

• 699 licensed German (and Austrian) patents – 1910, 956348, R. Herz, Cassella, Blue vat dye

• USPTO subclass – 548/442 “Organic compounds”– Treated if at least one license– 336 of 8,422 subclasses are treated

• Licensed to U.S. firms– 956348, E. I. Du Pont de Nemours & Co.

Subclass 548-422• Organic compounds wherein one of the benzene

rings and an additional carbocyclic ring bonded directly to the same acyclic nitrogen

Outcome variable: U.S. patents

• 165,400 chemical patents in 21 classes of organic chemicals, 1875-1939– Year granted, subclass– From www.uspto.gov

• 8,422 subclasses – 336 treated subclasses – Primary and secondary subclasses

• Match with specific dyes and firms– E.g., Indigo, Du Pont, National Aniline

Potential sources of bias• Systematic differences in patenting across

subclasses – Broad versus narrow classes– Variation in propensity to patent – Include subclass fixed effects

• Untreated classes are treated– Knowledge spillovers across subclasses– Underestimate the effects of compulsory

licensing

Nationalities of inventors• Lexis Nexis Chronological Patent Files, 1790 - 1970

– Keyword search in full text and inventor field• Search for inventors from enemy and other countries

– Germany, Austria– Argentina, Australia, Belgium, Brazil, China, England, France, Italy, the

Netherlands, Russia, Scotland, Switzerland, Spain

Hand-check• OCR may misread words• Small countries are omitted• Hand-collect 625 dye patents, 1900-1943

FIGURE 4 – HAND-CHECKED AND ALGORITM-ASSIGNED NATIONALITES BY USPTO CLASS (1900-1943)

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

8 101 106 204 430 528 534 536 544 546 548 549 552 562 564 568 570

USPTO main class

Inve

ntor

s by

cou

ntry

Germany .U.S Other country

Hand-collected sample Machine-collected sample

7 FIGURE 6 – HAND-CHECKED VS. ALGORITHM ASSIGNED NATIONALITIES (1900-1943)

Hand-collected sample Machine-collected sample

Inve

ntor

s by

coun

try

FIGURE 1 – PRE-TWEA TIME TRENDS IN PATENTING BY DOMESTIC INVENTORS: TREATED VERSUS UNTREATED SUBCLASSES

Coef

ficie

nt fo

r yea

r dum

mie

s

Treated Subclasses 95% CI 95% CIUntreated Subclasses

Control for number of licensed patentsFIGURE 2 – LICENSED PATENTS PER SUBCLASS

0

2

4

6

8

10

12

14

16

18

Lice

nsed

pat

ents

USPTO subclass

Control for age of licensed patentsFIGURE 3 – REMAINING LIFETIME OF LICENSED PATENTS PER SUBCLASS

Block bootstrap• Potential problem

– Serial correlation in the outcome variable– Understate standard error of β

• Block bootstrap – Bertrand, Mullainathan, and Duflo 2004– Draw subclasses at random with replacement

t-statistic tr = abs ( b̂ r - b̂ )/SE(b̂ r ).

TABLE 4 – CONFIDENCE INTERVAL OF THE BLOCK BOOTSTRAP COEFFICIENTS Treatment coefficient 99% confidence interval BDM test Subclass includes at least one license 0.0346864 0.1942624 99% Number of licenses 0.0203172 0.1149504 99% Remaining lifetime of licensed patents 0.0018647 0.0103889 99%

Note: Data from www.uspto.gov and the Lexis Nexis Chronological Patent Files (1790-1970). Our data consist of all 165,400 patents between 1875 and 1939 in 21 USPTO main classes that contained at least one licensed enemy patent. These 21 main classes are subdivided into 8,422 subclasses. Data on inventor nationality are based on a key word search of the Lexis Nexis. Confidence interval obtained by running OLS regressions on 79 block bootstrap samples of the original dataset as in Bertrand et al. (2004), that draw entire subclasses to maintain the structure of correlation constant.

APPENDIX TABLE D1 – CONFIDENCE INTERVAL OF THE BLOCK BOOTSTRAP COEFFICIENTS

Treatment specific time trends• Control for pre-existing differences in time trends across

treated and untreated subclasses• Treatment-specific linear and quadratic time trends

• Controlling for trends, compare changes in domestic patents across treated and untreated subclasses– Untreated coefficients of year dummies d– Treated coefficient of year dummies d

+ year-specific treatment effect b

YEAR SPECIFIC TREATMENT EFFECTS (1919-1939) DIFFERENCE-IN-DIFFERENCES, CONTROLLING FOR LINEAR TIME TRENDS FOR THE TREATED

SUBCLASSES

0

0.5

1

1.5

2

2.5

1915 1920 1925 1930 1935

Treated subclasses

Untreated subclasses δ

δ + β

Why the long lag?• Stock market crash and Great Depression

– New technology areas with unsteady income streams more vulnerable to budget cuts and restructuring

• Incomplete descriptions in German patents– BASF patents withheld essential information for Haber-Bosch

• “The Badische Company had effectively bulwarked this discovery with strong, broad patents…but cleverly avoided particulars as to the catalysts employed or their preparation. This last information was the core of the process so far as its practical operation was concerned.” (Haynes 1945, pp. 86-87.)

• U.S. firms took > 10 years (and $500 million in public funds) to replicate Haber Bosch– Du Pont wrestled with obscure descriptions

• “Acting under a license issued under the Trading-with-the-Enemy Act, Du Pont wrestled with the obscure descriptions in the German patents to work out a practical process only after long experimentation” (Haynes, 1945)

• Low initial skill levels of U.S. firms– More likely to license in subclasses where domestic invention is weak– Longer lead times to catch up and apply foreign inventions

U.S. firms were more likely to license in subclasses with low pre-TWEA share of U.S. inventors

FIGURE 4 – NUMBER OF SUBCLASSES WITH AND WITHOUT LICENSES BY PRE-TWEA SHARE OF DOMESTIC INVENTORS

Notes: Data from www.uspto.gov and the Lexis Nexis Chronological Patent Files (1790-1970). Our data consist of all 165,400 patents between 1875 and 1939 in 21 USPTO main classes that contained at least one licensed enemy patent. These 21 main classes are subdivided into 8,422 subclasses. Data on inventor nationality are based on a key word search for country names in Lexis Nexis.

OLS over- or underestimates true effects

• Absence of domestic competition may encourage R&D– U.S. patents increase regardless of compulsory

licensing– OLS overestimates effects of compulsory licensing

• Lower initial skill levels reduce effects– Lower domestic skill levels in subclasses with fewer

domestic inventors– Higher start-up costs for domestic inventors– OLS underestimates effects of compulsory licensing

Dearth of human capital may slow R&D

• Subclasses with fewer domestic inventors have lower initial skill levels

• Higher start-up costs for domestic inventors• OLS underestimates effects of compulsory licensing

TABLE – ADDITIONAL CONTROLS: PROFESSORS OF CHEMISTRY, CHEMISTS, AND WORKERS IN DYEING, 1880-1930 Census year Professors of chemistry % of all professors Chemists % of all workers Workers in dyeing % of all workers

1880 1 3.1% 27 0.0% 371 0.1% 1900 0 0.0% 85 0.0% 1,205 0.2% 1910 3 1.3% 158 0.0% 2,198 0.6% 1920 8 2.7% 340 0.1% 2,597 0.6% 1930 8 1.5% 488 0.1% 4,740 1.0%

Notes: Based on a 1 percent sample of from IPUMS micro data 2008. Professors of Chemistry, Chemists, and Workers in Dyeing are the only employment groups for which consistent data are available for the U.S. census during those years. Professor of Chemistry help proxy for the highest level of academic training; Chemists p roxy for high-skilled workers, and Workers in dyeing for average skill levels.

Triple Differences• Control for factors that affect patenting by all non-German inventors

– U.S. inventors, but also inventors from Britain, France, etc.– Absence of German competitors during the war– Re-entry of German competitors after the war

• Compare– changes in patents by U.S. inventors– changes in patents by non-German foreign inventors – across treated and untreated subclasses – before and after the TWEA

– where the n distinguishes U.S. and other non-German inventors

Placebo treatments• Potential problems

– Possibly random correlation between other variables independent of treatment

– Pre-TWEA German inventors patent in patent-intensive subclasses, post-TWEA other inventors catch up in these subclasses (independent of TWEA)

• Placebo test– Randomly assign 4 percent of inventors to be “treated”

• Same share of subclasses as treated U.S. inventors by TWEA

– Test for similar “effect” on patents by French inventors in U.S.• French inventors cannot license enemy-owned patents under the TWEA

FIGURE 12 - PLACEBO TREATMENT ON PATENTS BY FRENCH INVENTORS (1919-1939)

Notes: For a 95-percent confidence interval. Data from www.uspto.gov and www.lexisnexis.com. The data include all 165,400 patents between 1875 and 1939 in 21 USPTO main classes that received at least one license under the TWEA; this covers 8,422 subclasses, including 336 treated subclasses. The figure plots the coefficients of the year-specific effect of treatment on the number of patents from a French inventor.

FIGURE 10 - PLACEBO TREATMENT ON PATENTS BY FRENCH INVENTORS

Alternatively, let “treatment” begin pre-TWEA (Examples: TWEA begins in 1905)

Random treatment

• Randomly assign 4 percent to be treated • Same share as treated subclasses• Reject H0 for 45 of 50 placebos at the 5 percent level

TABLE 8 - OLS RESULTS WITH PLACEBO TREATMENT DISTRIBUTION of the t-statistic t-statistic 5th percentile -1.68 Median 0.10 95th percentile 1.68

EXAMPLE

Placebo (randomly assigning 4% of subclasses to treatment) -0.007 0.0144 Number of patents by foreign inventors 0.324*** (0.019) Constant 0.154*** (0.007) Subclass fixed effects Yes Year fixed effects Yes Observations 547,430 Number of subclasses 8,422

TABLE 7 - OLS RESULTS WITH PLACEBO TREATMENT

Intent to treat (ITT) • Selection

– U.S. firms more likely to license in subclasses where U.S. invention is weak

• Spillovers– Knowledge spillovers to untreated classes– Potentially less enforcement of enemy-owned patents

• Estimate effect of intent to treat– Define treatment as enemy patents– ITT/Prob(treat) = Treatment on treated

• Measurement error– Enemy Patents includes patents licensed voluntarily by U.S. firms before the

TWEA– Cannot be licensed under the TWEA, but allow U.S. firms to produce foreign

invention, as if they were affected by TWEA– Inflate ITT

TABLE 3– INTENT TO TREAT: DEPENDENT VARIABLE IS PATENTS BY U.S. INVENTORS PER SUBCLASS AND YEAR

(1) (2) (3) (4)

Number of enemy patents 0.065*** 0.077*** (0.007) (0.008) Remaining lifetime of enemy patents 0.008*** 0.01*** (0.0001) (0.0001) Number of patents by foreign inventors 0.311*** 0.308*** (0.02) (0.003) Subclass fixed effects Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Observations 547,430 547,430 547,430 547,430 Number of subclasses 8,422 8,422 8,422 8,422

Robust standard errors clustered at the subclass level in parentheses *** p<0.01, ** p<0.05, * p<0.1

Note: Data from www.uspto.gov and the Lexis Nexis Chronological Patent Files (1790-1970). Data include all 165,400 patents between 1875 and 1939 in 21 USPT classes that included at least one licensed enemy-owned patent. These 21 classes are subdivided into 8,422 subclasses. Data on inventor nationality are based on a key word search for country names in Lexis Nexis. Regressions that include a two year lag drop the first two years of data.

Tariffs• 1922 Fordney-McCumber Act

– Protects indigo, alizarin and all vat dyes• 1930 Smoot-Hawley Tariff

– Eichengreen 1989, Irwin 1998, Cuccini 1994

Demand shock of WWI

British naval blockade cuts off German imports

Increased demand for domestically produced chemicals

Navy blue uniforms, tarp, explosives

Indigo is most affected• 90 percent imported from

Germany• Navy uniforms• Most persistent price effect

– 1914: < 20 cents per pound (in real prices)

– 1917: 70 cents – 1919: 40 cents

• Price increase affects all indigo invention

• TWEA affects treated subclasses

FIGURE 15 – INDIGO PRICE PER POUND

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921

real

190

9 $

per p

ound

Notes: Price data from Haynes 1945, Haber 1971, p.185.

FIGURE B2 – INDIGO PRICE PER POUND

Additional robustness checks

• Control for treatment-specific time trends• Control for interactions between broader

classes and year fixed effects• Restrict the sample to primary subclasses• Drop new subclasses classes

Control for differential growth above the subclass level

• Control for class * year fixed effect

• λm,c,t where m indexes class• Estimate on a 10% subsample of subclasses, 10%

of each class

Restrict to primary subclasses

• Secondary subclasses– Cross-reference related technologies– Affected by licensing

• Potential problem– A lot of weight on patents that belong to many

subclasses– Assume that the number of patents is independent

across subclasses• Restrict to 6,740 primary subclasses

– Compared with 8,422 primary and secondary– 0.2 patents per primary subclass and year

What was the mechanism by which the TWEA encouraged patenting?

• Firm level analysis– Compare effects of firm’ own licenses to effects of any U.S.

license on a firm– Only own licenses allow firm to produce– Other U.S. licenses include spillovers, absence of German

competitors, etc.• Examine U.S. patents by Pont de Nemours & Co.

– Licensed 234 enemy-owned patents under the TWEA– Granted 3,571 patents, 1875-1939– Du Pont patents cover 5,716 subclasses, 1,618 in our data

• Caveat of firm-level analysis– Exogeneity is less clear at firm-level– Use as robustness check to distinguish alternative mechanisms

FIGURE 14 - YEAR-SPECIFIC TREATMENT EFFECTS – DU PONT

Aggregate Effects

• U.S. chemists become more cited– In U.S. and German

journals– Relative to German

chemists• After 1929 growth of

chemical production exceeds growth of other manufacturing

Conclusions (1/2)• Compulsory licensing encourages domestic

invention– Patents increase by 0.1 per subclass and year – Average 0.7 patents per subclass and year– Significant at 1 percent

• Effects intensify over time– Strongest effects sets in 7 to 10 years – In the 1930s around 0.5 additional patents per year

• Significant aggregate effects– Production, comparative advantage– Science, citations

Conclusions (2/2)• Potential mechanisms

– Learning by doing– Effects of firm’s own licenses are about 4 times as

large• Implications for domestic policy

– Encourage production by other firms– Licensing, potentially shorter patent lives

• International policy implications– Compulsory licensing encourages domestic invention – Counteracting potentially negative incentive effects

Extra slides

Expulsion of German scientists in 1933

Patents of dyestuffs in treated subclasses, 1875 to 1945

0

50

100

150

200

250

300

1875 1880 1885 1890 1895 1900 1905 1910 1915 1920 1925 1930 1935 1940 1945

Pate

nts

by n

atio

nalit

y in

trea

ted

subc

lass

es

U.S. Inventors German Inventors Other Inverntors

Patents of dyestuffs in untreated subclasses, 1875 to 1945

0

500

1000

1500

2000

2500

3000

3500

4000

1875 1880 1885 1890 1895 1900 1905 1910 1915 1920 1925 1930 1935 1940 1945

Pate

nts

by n

atio

nalit

y in

unt

reat

ed su

bcla

sses

U.S. Inventors German Inventors Other Inventors

Control for patent quality• Some patents are more important than others• Measures of quality: License fees

– Some data from Alien Property Custodian Report• Alternative measure of quality: Citations

– Trajtenberg 1990– NBER patent data file: Hall, Jaffe, Tratjenberg 2001– All 16 million citations made between 1975 and 1999– Made to nearly 3 million U.S. patents granted

between January 1963 and December 1999– 154 out of the 699 Haynes patents have 1 citation

Licensed patents with at least one citation (by year of grant)

0

5

10

15

20

25

30

35

1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922

TABLE 4 – INSTRUMENTAL VARIABLE REGRESSION, DEPENDENT VARIABLE IS PATENTS BY U.S. INVENTORS PER USPTO SUBCLASS AND YEAR (1875-1939)

First Stage Second Stage

I II III IV

At least one German patent 1902-1918 0.127 *** (0.0005) German patents 1902-1918 0.119*** (0.0002) Subclass has at least one license 3.270*** (0.051) Number of licenses 0.644*** (0.012) Subclass fixed effects Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Observations 547,430 547,430 547,430 547,430 Number of subclasses 8,422 8,422 8,422 8,422 Note: Data from www.uspto.gov and the Lexis Nexis Chronological Patent Files (1790-1970). Our data consist of all 165,400 patents between 1875 and 1939 in 21 USPTO main classes that contained at least one licensed enemy patent. These 21 main classes are subdivided into 8,422 subclasses. Data on inventor nationality are based on a key word search for country names in Lexis Nexis.