1. The Economic Meaning of Patent Citations: Value and Organizational Form in Patenting Start-ups Oral Examination (Ph.D. in Business Administration) Edward J. Egan Monday May 9 th, 2011

2. Research Question What is the economic meaning of patent citations?

3. Approach advance theoriesUse the property right and informational aspects of patents to advance theories about what a relationship between patents could mean Test the predictionsTest the predictions these theories make about the value of successful start-ups that own patents, and their choice of organizational form

4. Why Successful Start-ups? Patents should matter to these firms and these should be ‘important’ patents –If the relationships between patents don’t have economic meaning here, they probably don’t anywhere We have a good measure of the long-term value of these firms, and the outcome of an organizational form choice as to how value is best realized.

5. What is a Successful Start-up? Must have one (29%) or more patents before its liquidity event Could be VC backed (30%), a spin-off, from a university lab, or privately funded. Must be privately held and experience success in the form of an acquisition (54%) of an IPO (46%) Could be in IT (35%) or Biotech (8%*). Or classified as Manufacturing (75%), Information & Consulting (17%), or Trade & Transportation (10%) *Difficult to determine using NAIC codes. Biotech accounts for 30% of VC backed startups in our sample.

6. What is a Patent? A property rightA property right –A right to exclude and to a monopoly, but not a right to use InformationInformation –A complete, useful, novel and non-obvious application of knowledge Useful facts from the literature: –Patents are used only 50% of the time (1) –The average value of a 1957 patent was $0.5m in 1988 prices (1) –Patents cost an average of $25,000 each to acquire (2) (1) Reported in Griliches (1998) (2) According to Lemley (2001)

7. Patent Characteristics Dates of application and granting –Term lengths of 17 yrs from grant pre-1995 or 20 yrs from application post-1995 Claims –May be a measure of breadth Classes –USPTO: 400 classes, over 100,000 sub-classes –IPC: 8 sections, 61,387 sub groups –HJT (3) : 6 categories, 36 sub-categories Assignees –37% of our firms have a patent jointly assigned with one or more outside corporation (2) Hall, Jaffe and Trajtenberg (2001), “The NBER Patent Citation Data File”, NBER Working Paper

8. Patent Characteristics CitationsCitations –Who adds citations? On an average patent, 65% of citations are added by the examiner (4) For 40% of patents, all citations are added by the examiner (4) –What do citations mean? 35% of cited patents are complements (5) 3% of cited patents are substitutes (5) What about the other 64%?What about the other 64%? (4) Data on who assigned citations is available from the USPTO for patents filed after 2000. Alcacer & Gittelman (2006) report these results. (2) From survey data presented in Harhoff, Scherer & Vopel (2003)

9. An Example: Citations Received More citations a patent received => the patent is of greater valueCommon claim in the literature: More citations a patent received => the patent is of greater value Evidence that this is true: Harhoff et al. (1999) – Survey Data Hall et al. (2001) – Weak evidence from Manufacturing Firms Indirect evidence: Trajtenberg (1990), Bessen (2008) Contradicting evidence: Sampat & Ziedonis (2004), Hall et al. (2007) Still on open question: Moser (2011).

10. Two Theories and a View of Citations Property Rights TheoryProperty Rights Theory –Citations represent a complement or a substitute and patents are inputs or outputs from firm’s productions functions Signaling TheorySignaling Theory –Patents act as signals of firm value. Citations may influence the nature of these signals Who gets the best signal? How strong is the signal? Landscape ViewLandscape View –If citations characteristics influence value or organizational form in systematic ways that can not be explained by the above theories, are these effects consistent with a Landscape view?

11. Property Rights Example Patents as complementary inputs –The more citations received: the greater the rents accumulated Patents as substitute outputs –The more citations received: the lower the rents accumulated Mixes of complements and substitutes –Possibly no prediction at all

12. Property Rights Theory Patents as complements –The more citations made: the lower share of rents retained –The more citations received: the greater the rents accumulated Patents as substitutes –Reverse predictions Mixes of complements and substitutes –Possibly no prediction at all

13. The Landscape View Examples of ideas that would fit under the landscape view are: –Geographic spillovers of knowledge –Basic vs. applied research –Incremental vs. Radical Innovation –Component vs. Architectural Innovation –The modularity of innovation The landscape view considers the importance of the nature of the underlying technology space

14. The ‘Experiment’ How do IPOs and acquisitions differ in terms of citation patterns? Controlling for organizational form, what influences firm value in terms of citation patterns? The importance of time-based considerations

15. Some Results on Firm Value An additional patent increases firm value by 2%, or about $340,000. The number of citations received does not matter.Citation measures do not predict value above and beyond patents themselves. The number of citations received does not matter. Broader patents are not more valuable Jointly assigned patents are not more valuable The dependent variable is the natural log of firm value. The coefficients are estimated using Ordinary Least Squares regressions with a Huber-White sandwich adjustment to correct for heteroskedasticity. All columns estimate results using the full sample. Column 1Column 2Column 3Column 4Column 5 Log No. of Patents 0.207*** (0.026) 0.208*** (0.026) 0.240*** (0.036) 0.218*** (0.027) 0.212*** (0.027) Log Avg. No. Claims Made -0.012 (0.044) Multiple Corporate Assignees -0.119 (0.078) Log Avg. Citations Made -0.067 (0.043) Zero Citations Made Indicator 0.140 (0.160) Log Avg. Citations Recd. 0.004 (0.034) Zero Citations Recd. Indicator 0.289 (0.212) Avg. (Exit Year - Patent Year) 0.002 (0.007) 0.001 (0.007) 0.002 (0.007) -0.006 (0.008) 0.002 (0.007) Acquisition Indicator -1.010*** (0.063) -1.011*** (0.063) -1.012*** (0.063) -1.008*** (0.063) Exit Year Fixed Effectsyes Industry Fixed Effectsyes State Fixed Effectsyes Sub Cat. Fixed Effectsyes Constant 2.767*** (0.798) 2.794*** (0.799) 2.779*** (0.791) 2.872*** (0.791) 2.750*** (0.809) R-Squared0.3150010.31503640.31601350.31681660.315889 No. Observations1709 Standard errors are reported in parentheses. ***, **, and * indicate significance at the p=0.01, p=0.05 and p=0.1 levels respectively.

16. Some Results on Predicting Acquisitions Having more patents is associated with an initial public offering whether or not we control for value Firm’s that make more citations are less like to be acquired and more likely to IPO. This undermines a hold-up story. The number of citations received doesn’t matter. The dependent variable is a binary indicator that takes the value one if the firm experienced an acquisition and the value zero if the firm undertook an initial public offering. The coefficients are estimated with logit regressions that correct for heteroskedasticity. All columns estimate results using the full sample. Column 1Column 2Column 3Column 4Column 5 Log No. of Patents -0.250*** (0.064) -0.252*** (0.065) -0.201** (0.088) -0.239*** (0.066) -0.251*** (0.065) Log Avg. No. Claims Made 0.039 (0.107) Multiple Corporate Assignees -0.169 (0.197) Log Avg. Citations Made -0.280*** (0.101) Zero Citations Made Indicator -0.993* (0.549) Log Avg. Citations Recd. -0.063 (0.085) Zero Citations Recd. Indicator -0.436 (0.486) Avg. (Exit Year - Patent Year) 0.143*** (0.022) 0.144*** (0.022) 0.143*** (0.022) 0.130*** (0.025) 0.143*** (0.022) Firm Value -0.008*** (0.001) Firm Value Squared 0.000*** (0.000) Exit Year Fixed Effectsyes Industry Fixed Effectsyes State Fixed Effectsyes Sub Cat. Fixed Effectsyes Constant -4.011** (1.885) -4.094** (1.903) -4.109** (1.881) -3.848** (1.953) -3.781** (1.926) R-Squared0.36068380.36073630.36099950.36467420.3610728 No. Observations1673 Standard errors are reported in parentheses. ***, **, and * indicate significance at the p=0.01, p=0.05 and p=0.1 levels respectively.

17. More Results on Predicting Acquisitions The fragmentation of ownership made does not predict acquisition. This also undermines a hold-up story. The only significant measures have no meaning in either a property rights or a signaling theory. The dependent variable is a binary indicator that takes the value one if the firm experienced an acquisition and the value zero if the firm undertook an initial public offering. The coefficients are estimated with logit regressions that correct for heteroskedasticity. Column 6 estimates results using the full sample. Columns 1 through 3 and column 5 estimate results on the sub sample of firms that made at least one citation. Column 4 estimates results on the sub sample of firms that make at least one citation to an identified owner. maColumn 1Column 2Column 3Column 4Column 5Column 6 Log No. of Patents-0.211*** (0.066)-0.213*** (0.065)-0.212*** (0.066)-0.223*** (0.067)-0.182*** (0.067)-0.185*** (0.065) Avg. Diff. In No. Claims Made0.004 (0.005) Avg. Frag. Made (Sub Cat.) -0.255 (0.281) Avg. Tech Diff. Made-0.325 (0.323) Avg. Frag. Made (Ownership) -0.228 (0.313) Avg. Time to Cited0.064* (0.035) Pc. Of Cited w/ No Cites 0.892* (0.489) Zero Citations Made Indicator0.382 (0.530) Log Avg. Citations Made -0.447*** (0.103)-0.373*** (0.096) Avg. (Exit Year - Patent Year)0.161*** (0.030)0.160*** (0.030)0.162*** (0.030)0.169*** (0.033)0.143*** (0.031)0.115*** (0.033) Firm Value-0.008*** (0.001) Firm Value Squared0.000*** (0.000) Exit Year Fixed Effectsyes Industry Fixed Effectsyes State Fixed Effectsyes Sub Cat. Fixed Effectsyes Constant-4.366** (1.891)-4.270** (1.931)-4.275** (1.929)-2.088 (1.952)-4.291** (1.998)-4.261** (1.952) R-Squared0.36093580.36117260.3612270.3585040.36935260.3620909 No. Observations16101607 157216101673 Standard errors are reported in parentheses. ***, **, and * indicate significance at the p=0.01, p=0.05 and p=0.1 levels respectively.

18. Debunking Hold-up and Targeted Signaling Only half of acquirers have any patents! There is no relationship between the degree of the hold-up problem and the extent of an acquirer’s patent portfolio Acquirer has Patents IndicatorLog of No. of Acquirer Patents Column 1Column 2Column 3Column 4Column 5Column 6 Log No. of Patents 0.036 (0.071) 0.189 (0.123) Log Avg. Citations Made 0.150 (0.130) 0.060 (0.165) Zero Citations Made Indicator -0.342 (0.533) -0.460 (0.667) Avg. Frag. Made (Ownership) 0.224 (0.378) -0.841 (0.524) Avg. (Exit Year - Patent Year) -0.053*** (0.019) -0.033 (0.028) -0.044 (0.028) -0.036 (0.029) -0.020 (0.042) -0.050 (0.045) Exit Year Fixed Effectsyes Industry Fixed Effectsyes State Fixed Effectsyes Sub Cat. Fixed Effectsyes Constant 3.666** (1.618) 3.341** (1.647) 2.398 (1.824) -6.452*** (1.580) -6.254*** (1.602) -5.026*** (1.720) R-Squared0.16258390.1599861 0.156664 2 0.3142290.30855810.3103374 No. Observations891 836453 434 Standard errors are reported in parentheses. ***, **, and * indicate significance at the p=0.01, p=0.05 and p=0.1 levels respectively.

19. Next Steps Consider in-term vs. out-of-term citation effects Consider self-citation effects Robustness checks on firm values Report results for: – VC backed firms only –Firms with a single patent only –Firms in specific industries: IT, Biotech, Semiconductor, etc. (This may shed light on the importance of IP strength)

20. Next Steps Refine and perhaps formalize the two theories Find good predictions that cut between the theories and test them Consider some exogenous shocks –Introduction of new patent classes –Within industry, court decisions or legislation that affect IP strength Attempt to turn the Landscape view into an actual theory

21. A Research Agenda Try to understand the nature of citations and code them –Read patent filings myself –Consider building Natural Language Processing software to read filings and code the relationship between them (e.g. complement, substitute, etc.) Consider developing network based theories of patent citations –This could provide a holistic approach –This might capture aspects of the landscape that are important

22. Network Measures Patent cite patents, which cite patents, and so forth. Therefore patents participate in networks. We need network measures that are not convoluted by measures of citations made or received. We need measures with clear meanings