Private Equity Returns and Disclosure Around the World

Private Equity Returns and Disclosure Around the World Douglas Cumming and Uwe Walz Hofstra Conference on Private Equity May 2, 2007 1

Motivation: Worldwide Policy Debate l 2002 CALPERS disclosure lawsuit – Public pension funds must disclose venture capital and private equity returns, even on unexited investments l l Do we need mandated disclosure standards for VC and PE funds? l Biggest issue for VC/PE markets since collapse of Internet bubble l 2 Implications for understanding determinants of, and reporting of, returns Regulation of VC and PE funds one of the biggest issues in UK Financial Times last week

Research Questions 1. 2. Are unexited investment values over-reported to institutional investors? 3. Are biases in reporting unexited investments related to legal conditions? 4. 3 What are the determinants of VC and private equity returns across countries? Relative merits of alternative approaches to stimulating VC markets

Prior Research l VC / PE Returns – – – l VC Exits – – l – 4 Cumming and Mac. Intosh (2003 Journal of Banking and Finance) Cumming, Fleming and Schwienbacher (2006 Journal of Corporate Finance) VC value-added – l Cochrane (2005 Journal of Financial Economics) Cumming and Mac. Intosh (2007 Cambridge Journal of Economics) Hege, Palamino and Schwienbacher (2003 WP) Lerner, Schoar and Wong (2006 Journal of Finance) Ljungqvist and Richardson (2003 WP) Cumming (2006 Journal of Business) Gompers and Lerner (1999 MIT Press) No prior paper on disclosures of unexited VC returns

New Contributions 1. 2. Innovative application of econometric selection methods to measure VC returns 3. First look at biases in unexited returns and relations to fundraising 4. 5 First look at project-specific returns to VC and private equity across countries Policy implications: Reporting Standards needed in VC?

I. Theory and Hypotheses II. Data III. Econometric Tests IV. Policy Implications 6

Pension Plan Members (you and I) Venture Capital Cycle Why care? Distorted asset allocations, less overall fundraising Cumming & Johan (2007 JBF) CD Howe Institute, AEI Sciences Po, Brookings, PWC, EVCA, NVCA, etc. They all care a lot! Institutional and Other Investors $ Returns Venture Capital Funds $ E. g. , CALPERS California Public Pension Fund Reporting bias of unexited returns in annual reports? This Paper Returns (realized vs ‘expected’) Entrepreneurial Firms 2 -7 years before exit event (IPO, Acquisition, Write-off) 7

1. Advice, Monitoring & Returns l Monitoring/advice activities of VC are responsible for return of VC l Main focus on VC characteristic l Model with asymmetric information Advice is not contractible l l IRR must be sufficiently large to induce VC to undertake optimal level of advice/monitoring l The more productive the VC is, the higher the optimal advice/monitoring level the lower the price of shares for the VC the higher the VC returns 8

1. Advice, Monitoring & Returns (Continued) Hypotheses: l The higher the intensity of monitoring and advice the higher the expected IRR of the VC – – – l 9 Convertible securities, syndication higher expected rate of return Co-investment: lower returns Smaller portfolios (# investments) / manager lower returns Better legal environment more efficient advice and less information asymmetries upon exit the higher expected returns

2. Biases in Reporting Un-Exited Investments l Valuation take place against trade-off between Ø Ø Fundraising concerns (higher valuations potentially facilitate fundraising in next round) Reputational concerns (overvaluation damages long-run reputation) l l 10 Simple set-up: two projects, two VC types Pooling equilibria may emerge (bad projects are overstated)

2. Biases in Reporting Un-Exited Investments (Continued) Hypotheses: l Expected Fundraising Benefit > Expected Reputation Cost – – l Legal environment increases costs of overstatement – – 11 Inexperienced VCs: overstate Earlier stage and high tech: overstate Syndicated investment: less likely to overstate Co-investment: more likely to overstate Less stringent accounting rules: overstate Sarbanes Oxley: less likely to overstate

I. Theory and Hypotheses II. Data III. Econometric Tests IV. Policy Implications 12

CEPRES Dataset l l l 13 221 venture capital and private equity funds 72 venture capital and private equity firms 5117 entrepreneurial firms (3826 venture capital and 1214 private equity) 32 years (1971 – 2003) 39 countries (North and South America, Europe and Asia) Table 1 (see paper) defines the variables

14

15

Table 2. Summary Statistics and Difference Tests Unrealized / Partially realized Ent Firm Investments Fully Realized Ent Firm Investments Difference Tests PE Fund Characteristics # Ent Firms Average IRR Median IRR Means Medians All Funds All Funds in the Data 2619 63. 23 0. 00 2419 68. 67 16. 99 0. 22 p <= 0. 00*** Market and Legal Factors MSCI Return > 3. 5% 611 76. 88 9. 32 1908 58. 07 20. 21 -1. 14 p <= 0. 000*** MSCI Return < 3. 5% 2010 59. 07 0. 00 511 108. 24 -10. 99 0. 64 p <= 0. 000*** Risk Free Return > 3. 5% 2333 49. 36 0. 04 2021 79. 59 17. 41 1. 36 p <= 0. 000*** Risk Free Return < 3. 5% 311 213. 32 0. 00 411 12. 92 15. 74 -1. 32 p <= 0. 000*** Legality Index > 20 1874 60. 01 2. 16 1631 47. 23 19. 26 -0. 87 p <= 0. 000*** Legality Index < 20 747 71. 30 0. 00 788 113. 04 14. 21 0. 54 p <= 0. 000*** Country Earnings Aggressiveness Index > -0. 383 765 27. 43 3. 17 646 85. 50 18. 39 1. 03 p <= 0. 000*** Country Earnings Aggressiveness Index < -0. 383 1858 77. 92 0. 00 1773 62. 54 16. 22 -0. 54 p <= 0. 000*** Country Disclosure Level Index > 76 621 18. 80 5. 64 595 91. 96 19. 05 1. 20 p <= 0. 000*** Country Disclosure Level Index < 76 2000 77. 02 0. 00 1824 61. 07 15. 68 -0. 60 p <= 0. 000*** 16

Table 2. Summary Statistics and Difference Tests (Continued) Unrealized / Partially realized Ent Firm Investments Fully Realized Ent Firm Investments Difference Tests PE Fund Characteristics # Ent Firms Average IRR Median IRR Means Medians Fund Characteristics Fund Number in the PE Firm > 3 1603 69. 37 0. 00 781 88. 72 1. 51 0. 34 p <= 0. 000*** Fund Number in the PE Firm < 3 1018 53. 55 10. 30 1638 59. 11 20. 27 0. 29 p <= 0. 000*** Age of Specific PE Fund > 1795 days 1230 54. 15 9. 23 2233 57. 48 18. 73 0. 19 p <= 0. 000*** Age of Specific PE Fund < 1795 days 1391 71. 25 0. 00 186 202. 96 -91. 74 0. 67 p <= 0. 000*** Portfolio Size (# Investees) / # General Partners > 20 1035 59. 58 0. 00 988 21. 29 12. 34 -2. 52** p <= 0. 000*** Portfolio Size (# Investees) / # General Partners < 20 1586 65. 61 1. 70 1431 101. 38 22. 07 0. 87 p <= 0. 000*** Entrepreneurial Firm Characteristics Seed Stage 146 8. 88 0. 00 71 520. 37 -2. 92 1. 01 p <= 0. 097* Start-up Stage 56 126. 72 18. 97 34 48. 58 -11. 45 -1. 65* p <= 0. 127 Early Stage 672 39. 55 0. 00 424 -1. 52 -29. 14 -2. 93*** p <= 0. 000*** Expansion Stage 240 36. 40 0. 00 226 28. 91 14. 54 -0. 56 p <= 0. 000*** Unknown Seed, Early or Expansion Stage 838 91. 80 5. 09 1119 71. 69 20. 00 -0. 36 p <= 0. 000*** Late Stage 168 55. 77 0. 00 116 121. 20 25. 34 1. 50 p <= 0. 000*** MBO/MBI 309 43. 79 8. 53 266 33. 33 28. 27 -0. 35 p <= 0. 000*** LBO 30 27. 43 13. 55 17 32. 73 44. 72 0. 37 p <= 0. 052* Other Type of Private Equity 153 144. 32 17. 14 132 69. 11 25. 52 -0. 69 p <= 0. 006*** Publicly Listed Firm 9 31. 41 0. 00 14 649. 54 29. 45 1. 29 p <= 0. 680 Industry Market / Book > 5 1448 101. 95 0. 00 816 80. 27 6. 08 -0. 55 p <= 0. 000*** Industry Market / Book < 5 1173 15. 42 7. 92 1603 62. 76 20. 28 2. 01** p <= 0. 000*** 17

Table 2. Summary Statistics and Difference Tests (Continued) Unrealized / Partially realized Ent Firm Investments Fully Realized Ent Firm Investments Difference Tests PE Fund Characteristics # Ent Firms Average IRR Median IRR Means Medians Investment Characteristics Lead Investment 864 75. 01 8. 33 633 45. 11 20. 33 -1. 21 p <= 0. 000*** Syndicated Investment 729 68. 11 0. 00 449 151. 27 15. 88 1. 01 p <= 0. 000*** Co-Investment 526 44. 51 0. 00 313 48. 02 13. 27 0. 13 p <= 0. 000*** PE Board Seat(s) 743 42. 84 0. 00 447 112. 40 0. 26 0. 84 p <= 0. 000*** Convertible Security with Actual Periodic Cash Flows 967 123. 03 12. 77 1162 73. 62 25. 99 -0. 95 p <= 0. 000*** Standard Deviation of Cash Flows to Entrepreneur / Initial $ Invested 1203 130. 12 0. 00 1364 125. 85 31. 49 -0. 04 p <= 0. 000*** Standard Deviation of Cash Flows to Entrepreneur / Initial $ Invested 1418 6. 47 2. 92 1055 -5. 26 0. 33 -3. 06*** p <= 0. 216 Initial Amount Invested > $US 2, 500, 000 1310 34. 62 5. 04 1040 75. 58 25. 22 1. 09 p <= 0. 000*** Initial Amount Invested < $US 2, 500, 000 1311 91. 80 0. 00 1379 63. 46 8. 60 -0. 75 p <= 0. 000*** 18

Table 3. [Condensed] Correlation Matrix (1) Log (1+IRR) 1. 00 (2) Log (MSCI) 0. 15 (3) Log (Interest) -0. 06 (4) Log (Legality) 0. 03 (5) Log (Committed Capital) -0. 06 (6) Log (Fund Number) 0. 04 (7) Log (Portfolio Size / Manager) 0. 03 (8) Seed -0. 10 (9) Early -0. 03 (10) Expansion 0. 03 (11) Late 0. 03 (12) Log (Industry Market / Book) 0. 01 (13) Lead Investor 0. 07 (14) Syndicated Investment 0. 06 (15) Co-Investment -0. 06 (16) Board Seats 0. 00 (17) Convertible Security 0. 05 (18) Standard Deviation of Cash Flows 0. 01 (19) Log (Initial Investment) -0. 04 19

I. Theory and Hypotheses II. Data III. Econometric Tests IV. Remarks 20

“Realized Returns Econometrics” l Multi-step Heckman correction to measure the returns to VC and private equity investment l Heckman selection corrections for 1. 2. l 21 Unexited / Exited Investments Partial / Full Exits Statistical problems associated with OLS on a subsample of fully realized IRRs

3 -Step Heckman Correction 1. 2. Selection Corrected Probit: Full / Partial Exit, accounting for the selection effects associated with an actual exit (step 1) 3. 22 Probit: Exit / No Exit Heckman Linear Regression IRR, accounting for both steps # 1 and 2 Contrast to Cochrane (2002): moves from step # 1 to step # 3 Contrast to Ljungqvist and Richardson (2003): OLS on restricted sample of realized returns

Table 4. Heckman Corrected IRR Regressions Panel A. Seed, Start-up, Early and Expansion Stage Investments Model (1) Model (2) 1 st Step Heckman Regression: Bivariate Probit Model OLS on Subsample of Fully Realized IRRs Predicted Sign for Realized Returns Dependent Variable = Log(1+IRR) Step 1 a: Determinants of Exit Step 1 b: Determinants of Full Exit, conditioned on step 1 a regarding an actual exit 2 nd Step Heckman Regression (Realized IRRs) Dependent Variable=1 if Exit Dependent Variable=1 if Full Exit Dependent Variable = Log(1+IRR) Coefficient t-statistic -5. 87 -1. 5 -0. 44*** Duration of VC Investment (in Days) Market and Legal Factors Constant Coefficient t-statistic Coefficient t-statistic -11. 7 -1. 80 -1. 0 -20. 27 -4. 0*** 0. 0006 24. 4*** -0. 00008 2. 5** Log (MSCI Return) + 0. 77 1. 0 1. 15 2. 1** Log (Risk Free Rate) ? -13. 32 -2. 7*** -27. 08 -4. 6*** Log (Legality Index) + 4. 25 4. 0*** 0. 52 0. 8 3. 64 2. 5** Log (Committed Capital Overall Market at Inv Date) - -0. 94 -6. 5*** 1. 31 5. 4*** Fund Characteristics Log (Fund Number in the VC Firm) + -0. 01 -0. 1 -0. 14 -1. 1 Log (Portfolio Size (# Investees) / General Partner) - -0. 41 -2. 5** -0. 51 -3. 2*** Industry Dummy Variables? Yes No Yes Country Dummy Variables? Yes No Yes Exit Year Dummies? Yes No Yes Continued… 23

Table 4. Heckman Corrected IRR Regressions (continued) Model (1) Model (2) 1 st Step Heckman Regression: Bivariate Probit Model OLS on Subsample of Fully Realized IRRs Predicted Sign for Realized Returns Dependent Variable = Log(1+IRR) Step 1 a: Determinants of Exit Step 1 b: Determinants of Full Exit, conditioned on step 1 a regarding an actual exit 2 nd Step Heckman Regression (Realized IRRs) Dependent Variable=1 if Exit Dependent Variable=1 if Full Exit Dependent Variable = Log(1+IRR) t-statistic Investment Characteristics Coefficient t-statistic Coefficient t-statistic Lead Investment ? 0. 18 0. 6 0. 19 0. 7 Syndicated Investment + 0. 34 1. 3 -0. 42 -3. 8*** 0. 51 1. 7* Co-Investment - -0. 30 -1. 2 -0. 40 -1. 7* VC Board Seat(s) + -0. 46 -1. 2 -0. 70 -2. 2** Convertible Security with Actual Periodic Cash Flows + 2. 43 13. 6*** 1. 97 9. 9*** Standard Deviation of Cash Flows to Entrepreneur ? 1. 200 E-02 2. 0** 0. 01 2. 8*** Log (Amount Invested) ? 9. 052 E-02 1. 5 0. 11 1. 9* Heckman Lambda A - -2. 18 -2. 3** Heckman Lambda B - -7. 22 -11. 3*** Model Diagnostics Number of Observations 1358 3213 1358 Adjusted R 2 0. 29 0. 35 F Statistic 15. 74*** 19. 65*** Loglikelihood Function -3456. 74 -2756. 44 -3373. 45 5. 15 5. 03 Akaike Information Statistic Prior work: explains 1% (Cochrane, 2001) to 12% (Ljungqvist and Rihardson, 2003) of variation in VC returns 24

Unexited Reported IRRs (2000 – 2003) versus Predicted IRRs l l Log(1+IRR Reported)-Log(1+IRR Expected) = Log((1+Reported IRR)/(1+Predicted IRR) = 143% l 25 Contrast reported unexited IRRs (as reported to the institutional investors) with predicted IRRs for unexited investments Regression evidence: quite remarkably(!) consistent with the proposition that more informational asymmetry is associated with more ‘lying’!

Table 6. Unexited Reported IRRs versus Predicted IRRs Dep Var = Unexited IRR – Predicted IRR from Respective Model # Panel A. Seed, Start-up, Early and Expansion Stage Investments Model (1 a) Dependent Variable: Unrealized Log(1+IRR) - Fitted Values from Predicted Log (1+IRR) in Model (1) of Table IV Panel A Dependent Variable: Unrealized Log(1+IRR) - Fitted Values from Predicted Log (1+IRR) in Model (2) of Table IV Panel A Coefficient t-statistic 8. 8*** 11. 08 13. 9*** 20. 77 3. 6*** 22. 18 5. 4*** Constant Model (2 b) 35. 61 Predicted Sign Model (2 a) Dependent Variable: Unrealized Log(1+IRR) - Fitted Values from Predicted Log (1+IRR) in Model (1) of Table IV Panel A Model (1 b) Market and Legal Factors Log (MSCI Return) - -1. 21 -6. 4*** -3. 11 -9. 1*** Log (Risk Free Rate) ? 36. 57 16. 2*** -14. 17 -3. 8*** Country Earnings Aggressiveness Index + 42. 46 5. 9*** 32. 38 9. 1*** 37. 95 3. 5*** 28. 95 4. 7*** Country Disclosure Level Index - -6. 59 -6. 8*** -0. 05 -5. 4*** -5. 29 -3. 8*** -3. 18 -3. 3*** Sarbanes Oxley - -0. 56 -6. 2*** -0. 34 -5. 8*** -1. 07 -8. 1*** -1. 38 -15. 3*** Fund Characteristics Log (Age of VC Fund within the VC Firm) - -0. 42 -6. 7*** -1. 75 -13. 5*** Log (Portfolio Size (# Investees) / General Partner) + 0. 49 10. 9*** 0. 82 11. 1*** Industry Dummy Variables? Yes Yes Country Dummy Variables? Yes Yes Continued… 26

Panel A. Seed, Start-up, Early and Expansion Stage Investments Model (1 a) Predicted Sign Model (2 a) Model (2 b) Dependent Variable: Unrealized Log(1+IRR) - Fitted Values from Predicted Log (1+IRR) in Model (1) of Table IV Panel A Model (1 b) Dependent Variable: Unrealized Log(1+IRR) - Fitted Values from Predicted Log (1+IRR) in Model (1) of Table IV Panel A Dependent Variable: Unrealized Log(1+IRR) - Fitted Values from Predicted Log (1+IRR) in Model (2) of Table IV Panel A Coefficient t-statistic Coefficient t-statistic Investment Characteristics Coefficient Lead Investment ? -0. 05 -0. 7 0. 14 1. 3 Syndicated Investment - -0. 310 -3. 2*** -0. 28 -4. 2*** -0. 76 -5. 5*** -0. 75 -8. 2*** Co-Investment + 0. 23 3. 8*** 0. 17 1. 9* VC Board Seat(s) ? 0. 51 5. 7*** 0. 68 5. 5*** Convertible Security with Actual Periodic Cash Flows - -2. 46 -19. 5*** -2. 64 -10. 5*** Standard Deviation of Cash Flows to Entrepreneur ? -0. 01 -9. 7*** -0. 01 -2. 4** Log (Amount Invested) ? -0. 11 -6. 5*** -0. 05 -1. 5 Model Diagnostics Number of Observations 1122 Adjusted R 2 0. 36 0. 74 0. 25 0. 70 37. 50*** 102. 20*** 28. 10*** 91. 66*** -1830. 18 -1307. 71 -2262. 49 -1740. 97 3. 29 2. 39 4. 06 3. 16 F Statistic Loglikelihood Function Akaike Information Statistic 27

Appendix: Compare Actual IRR to Prior Reported Unexited IRR (This is possible now in 2006!) l l Subsample of 80 observations (investee firms) from 11 countries for which both the realized and unrealized reported IRR are known (Canada, Finland, France, Germany, Israel, Norway, Spain, Sweden, the Netherlands, the UK, and the US) The correlation between out-of-sample average realized IRRs and our predicted IRRs is 0. 45 Average Duration Report Exit 2. 6 years Unrealized reported IRR 219. 71% 2. 56% Subsequently Realized Reported IRR 28 Median 98. 46% 8. 70% Predicted IRR (Based on Table IV Model) 15. 22% 7. 75%

Table VIII. Determinants of the Difference between Reported Unrealized IRRs Disclosed to Institutional Investors and Subsequently Realized IRRs Model (A 1) Constant Model (A 4) Dependent Variable: Unrealized Reported Log(1+IRR) - Fitted Values from Predicted Log (1+IRR) in Model (1) of Table IV Panel B Unrealized Reported Log(1+IRR) - Subsequently Realized Log (1+IRR) Coefficient t-statistic 52. 581 Predicted Sign Model (A 3) Dependent Variable: Model (A 2) 3. 035*** 106. 795 1. 517 7. 729 0. 907 17. 299 2. 784*** -0. 884 -0. 801 0. 566 0. 124 -1. 925 -0. 434 -0. 168 -0. 054 0. 362 0. 987 387. 377 2. 530** 375. 457 2. 453** Market and Legal Factors Log (MSCI Return Reporting Time) - Log (MSCI Return Exit Time) ? Duration from Reporting to Realization ? Country Earnings Aggressiveness Index + 53. 544 1. 166 361. 277 1. 711* Country Disclosure Level Index - -9. 577 -2. 629*** -21. 229 -1. 420 Log (Age of PE Fund within the PE Firm) - -0. 441 -1. 529 -0. 351 -0. 320 0. 045 0. 043 Log (Portfolio Size (# Investees) / General Partner) + 0. 070 0. 309 -0. 085 -0. 049 1. 058 0. 595 Fund Characteristics Continued… 29

Model (A 1) Predicte d Sign Model (A 3) Model (A 4) Dependent Variable: Model (A 2) Dependent Variable: Unrealized Reported Log(1+IRR) - Fitted Values from Predicted Log (1+IRR) in Model (1) of Table IV Panel B Unrealized Reported Log(1+IRR) - Subsequently Realized Log (1+IRR) Coefficient t-statistic 0. 673 2. 620*** 0. 162 0. 147 -0. 189 -0. 168 0. 051 0. 042 Entrepreneurial Firm Characteristics Log (Industry Market / Book) + Industry Dummy Variables? Yes Yes Country Dummy Variables? No No Yes Investment Characteristics Syndicated Investment - -0. 639 -2. 600*** 0. 691 0. 518 0. 657 0. 511 Convertible Security with Actual Periodic Cash Flows - -2. 703 -10. 880*** -3. 201 -3. 145*** -3. 215 -3. 317*** Standard Deviation of Cash Flows to Entrepreneur ? 0. 081 1. 102 0. 005 0. 032 Log (Amount Invested) ? -0. 112 -1. 035 0. 341 0. 775 0. 372 1. 071 -3. 129 -3. 437*** Model Diagnostics Number of Observations 80 80 0. 766 0. 130 0. 131 0. 159 F Statistic 20. 97*** 1. 90** 1. 74* 2. 35** Loglikelihood Function -105. 336 -211. 641 -209. 713 -211. 481 2. 983 5. 641 5. 668 5. 587 Adjusted R 2 Akaike Information Statistic 30

Overstatement of Unexited IRRs and Fundraising l l 31 Not possible to assess causality but there is evidence of positive correlations between overstatement of unexited reported IRRs and fundraising

Correlations: Overstatement of Unexited IRRs and Fundraising Fitted Value From Difference Regression Actual Difference 1. 00 0. 23 Fitted Values from Difference Regression 0. 23 1. 00 Fund Size 0. 18 0. 27 VC Firm Age 0. 24 0. 39 Capital Under Management 32 Actual Difference (Reported Predicted IRR) 0. 21 0. 33 Capital Under Management to Date of Fundraising 0. 26 0. 37

I. Theory and Hypotheses II. Data III. Econometric Tests IV. Policy Implications 33

Measuring VC Returns l Heckman selection effects are crucial Ø Ø Ø l VC value-added is crucial Ø Ø Ø 34 • Misspecification of model without selection effects Like Cochrane (2002), unlike Ljungqvist & Richardson (2003), unlike Brander et al. (2002) Multidimensional selection effects are a useful new component introduced in this paper E. g. , portfolio size / manager Enables us to explain up to 36% of the variation in returns Cochrane explains at most 1% using market variables only; Ljungqvist & Richardson explain up to 13% with some fund variables, but no proxies for value-added Legality is crucial for cross-country differences

Unexited IRRs Reported to Institutional Investors l Our findings are quite remarkably(!) consistent with the proposition that more informational asymmetry is associated with more ‘lying’! Ø l 35 for smaller ENTs, tech companies, higher earnings aggressiveness index, lower disclosure index Positive correlation between fundraising and lying