Prospectus forecast publication and forecast errors: the role of venture capitalist certification

Abstract

Using a dataset of French IPOs over the period 1996–2000 we provide a test of the certification ability of venture capitalists (VCs). We do this in two ways. First we compare forecast publication between VC- and non-VC-backed IPOs. Secondly, conditional on publication, we investigate the accuracy of VC-backed versus non-VC-backed IPOs' prospectus forecasts. We also test for the association between VC reputation and both forecast issuance and forecast accuracy. VC concerns for reputation should result in lower chances of issuing a forecast and in more accurate forecasts for VC-backed IPOs. Our main findings are that (i) the prospectuses of IPOs backed by VCs are indeed less likely to include financial forecasts; (ii) high-reputation VC backers are associated with lower forecast errors. These findings are consistent with the Certification Hypothesis and are robust to controlling for both sample selection bias, due to the discretionary nature of prospectus forecast issuance, and for earnings management practices.

Keywords:

• venture capitalist certification
• prospectus forecasts
• French IPOs

Notes

1. These two markets merged in February 2005.

2. The Autorité des Marchés Financiers.

3. As Bernstein and Lemaitre (2004) point out, the French IPO market was virtually dormant between 2002 and early 2004, which was in effect the first time the new rules were put into effect.

4. There is not a high concentration of VCs in either the high- or low-reputation lead VC categories. Seventy-five per cent of the IPOs in the high-reputation VC group have different lead VCs, as do 77% of those in the low-reputation group, so the results are not biased by repeated inclusion of the same VC firm. All lead VCs in the sample are based in France, although a very small number are subsidiaries of foreign groups. However controlling for the presence of a foreign VC parent does not change the results for the association with forecast issuance or profit forecast error.

5. The robustness of our results to the use of other deflators was investigated. We used the actual value of earnings as a deflator as well as IPOs' market value estimated at the offer price. Results are not reported as they remain qualitatively the same.

6. To cite the most extreme results of previous studies, Firth and Smith's (1992) New Zealand study and Lee, Taylor, and Taylor's (2006) Australian research find FEs in excess of 100%, while Keasey and McGuinness (1991) for the UK and Firth et al. (1995) for Singapore find FEs of only 11% or less. An example of a study with mid-range FEs is Jelic et al.' s (1998) findings for Malaysia with FEs of 55%.

7. Given the non-normality of the data, in addition to the parametric t-test we use a non-parametric bootstrap test from Efron and Tibshirani (1993, 224) to test for differences in means. However, our conclusions are not significantly affected by the type of test used. Differences in medians are tested for using the Mann-Whitney U test.

8. Other IPOs are defined as all IPOs not backed by a high-reputation lead VC. This group therefore includes both IPOs backed by non-high-reputation VCs and all non-VC-backed IPOs.

9. Here, other IPOs are all IPOs that are not backed by a low-reputation lead VC.

10. However, using a linear model does not change our findings for the impact of VC reputation on forecast error.

11. The residuals of all the regression runs of Equation (3) are highly non-normal. We tested the robustness of our parametric inferences by computing bootstrapped p-values for each coefficient estimate (see Davison and Hinkley 1999, 264–81). However, inferences do not change significantly whether one uses bootstrapped or parametric p-values, so only the latter are reported.

12. All variable definitions are set out in Table 2.

13. In unreported results we also controlled for industrial affiliations, using dummy variables, but did not find any of these variables to have a significant influence on the forecast error above that of the market on which the firm is floated.

14. When testing our hypotheses we first use all the above explanatory variables. However, we ultimately want to select the best explanatory model with the fewest number of variables. To achieve this we use a backward variable elimination procedure with removal criteria p-value > 0.2. By reducing the number of variables in our models we also address concerns that collinearity (see Table 6) may adversely affect the accuracy of our coefficient estimates.

15. In model 2 a backward variable elimination procedure is used (with removal criterion p > 0.20). When all variables are allowed to remain in the model the results are very similar with the coefficient for High-Reputation-Lead-VC negative and significant at better than the 5% level. We also test for the inclusion of year 1999 and 2000 dummies in model 2 to control for the hot issue market period. However the coefficient on neither year dummy is significant while the significantly negative coefficient on the High-Reputation-Lead-VC variable remains.

16. We find some evidence (although only weakly significant, at the 10% level on a one-sided test) that low-reputation lead VC IPOs are younger (average IPO age = 17 years) than high-reputation lead VC IPOs (23 years). This is not due to differences in the stage (early versus later) at which VCs invested in the firm.

17. We examine the influence of each observation using the Cook's D statistic. A worrisome case would be one where the Cook's D statistic is greater than one. Although no such case was identified we excluded a number of observations with relatively large Cook's D. The exclusion of these observations does not change our results for the lower forecast error of high-reputation VCs.

18. For the sake of clarity, we only report results after the use of a backward variable elimination procedure with removal criteria p > 0.20.

19. Again, we use the Cook's D statistic to identify potential problematic observations. No worrisome case was identified. We removed, however, a number of observations with relatively large Cook's D, with this having no significant effect on our main results regarding the impact of VC reputation on forecast error. A further run using the log value of a weighted average of the age of all VC investors, where the weights are the shareholding of each VC divided by the total VC shareholding, produces similar results to model 4, with the coefficient on the reputation proxy negative and significant at better than 5%.

20. A further robustness test, based on Dolvin and Pyles (2006), is undertaken using a binary variable equal to one when the lead VC served as a lead in any prior IPO in the sample, and zero otherwise. The coefficient is negative and significant at the 10% level on a two-tailed test.

21. Following Dolvin (2005) and others we also tested the association between venture capitalist quality and underpricing but found no significant relationship.

22. We also investigate signed discretionary current accruals as well as discretionary ‘total’ accruals. In both cases our findings remaining qualitatively the same.

23. Teoh, Welch, and Wong (1998b) find that firms engage in earnings management prior to seasoned equity offerings. However less than 10% of our benchmark firms had SEOs over the period of our study, and their impact on our measures of IPOs' earnings management is negligible.

24. As a further check of the potential impact of earnings management we ran our regression models for FE using the ADCA of each IPO as an additional explanatory variable. The coefficient on the ADCA variable is negative but not significantly different from zero, so that our findings remain unchanged.