Good Dollars Chasing Bad Dollars: The Impact of Venture Capital Funding On Industry Stock Returns Tim Loughran Mendoza College of Business University of Notre Dame Notre Dame, IN 46556-5646 574.631.8432 voice Loughran.9@nd.edu Sophie Shive Mendoza College of Business University of Notre Dame Notre Dame, IN 46556-5646 574.631.8301 voice sshive1@nd.edu May 1, 2007 ABSTRACT: The aggregate amount of venture capital (VC) distributions in non-publicly traded firms since 1980 is more than $390 billion dollars. We find that the lagged level of VC distributions is negatively related to subsequent quarterly industry returns after adjusting for other factors. Thus, money invested in non-publicly traded firms subsequently hurts the valuations of publicly traded companies in the same industry. The return on assets of an industry is also negatively related to lagged VC funding. It appears that the stock market fails to quickly incorporate the information of new VC funding into the valuations of publicly traded firms. As more money pours into a hot industry, increased competition and technological innovation causes subsequent industry stock returns and operating performance to be lower. *We would like to thank Robert Battalio and Paul Schultz for helpful comments.
1. Introduction Venture capitalists play an increasingly important role in the U.S. economy. Technologically savvy firms like Apple Computer, Sun Microsystems, Microsoft, ebay, and Google all had venture capital funding prior to going public. Yet the venture capitalist (VC) provides much more than simply cash for the young firms. VCs often serve on the firm s board of directors and can provide critical insights that potentially allow the entrepreneur to turn small illiquid start-ups into large publicly-traded firms. Prior research by Lerner (1994) suggests that venture capitalists are quite good at selecting periods of time when valuations are especially high to take young firms public. Brav and Gompers (1997) provide evidence that venture capital-backed initial public offerings (IPOs) have substantially higher returns after the offering than non-vc backed IPOs. Limited research, however, has focused on the implications of VC funding on subsequent industry returns. That is, is there an effect on the valuations of publicly traded firms in the same industry as VC disbursements are flowing into it? During the time period from 1980 to 2005, the aggregate amount of venture capital disbursements (i.e., funding) to U.S. firms, according to VenturExpert, totaled $392.4 billion. This paper examines the impact that the quarterly level of venture capital funding has on the subsequent returns of publicly-traded firms. Do venture capitalists pour money into industries that subsequently have high stock returns? Or, are subsequent industry returns dampened when the inflow of capital leads to increased competition and/or better innovation? How quickly do financial markets incorporate the information contained in the venture capital funding announcement? Does the industry return on assets suffer as a result of prior VC funding? 1
Fama and French (1997) classifications are used to create quarterly industry stock returns for all publicly-traded operating companies. Our data source for 73,346 rounds of venture capital funding is VenturExpert. The funding from the venture capitalist is for non-publicly traded firms at various stages of their life cycle. Some of the capital is for Seed or Startup financing while other capital is for Expansion or Later Stage financing prior to taking the firm public. Not surprisingly, the majority of VC funding is directed to Expansions. In a panel of 3,502 quarterly industry observations, we find evidence that higher levels of quarterly venture capital funding led to lower subsequent quarterly returns in that particular industry. So, if there is a large amount of VC dollars invested in the telecommunications industry in a particular quarter, the following quarter telecommunications industry stock return will be lower, all else being equal. The patterns are generally similar whether the industry returns are value weighted or equally weighted. The results are even stronger when the amount of venture capital disbursements is scaled by the level of total assets in the industry. In all of the regressions, we report t-statistics using standard errors that are heteroskedasticity-robust and clustered by both industry and time interval (either quarter or year). Our results are generally significant for both equal-weighted and value-weighted industry returns, and both including and excluding the internet bubble period of 1998-2001. When the results are categorized by funding type (i.e., Seed, Startup, or Expansion), we find negative and statistically significant coefficients (except for Seed financing) for all of the regressions with quarterly industry returns as the dependent variable. Since Seed financing never accounts for more than 9% of all quarterly VC 2
disbursements, the lack of a strong relationship in this case should not be overly surprising. The vast majority of VC disbursements occur in five different industries. As a robustness check, we run the analysis focusing on the five most commonly funded industries. In these regressions, there are only 102 quarterly observations. Even with the small sample size, a number of the coefficients on VC dollars or VC dollars scaled by total industry assets are negative and statistically significant when either value or equally weighted returns are the dependent variable. The highest and most significant coefficients occur in the Chips Industry. Lastly, we examine if industry operating performance is affected by the influx of VC dollars. Using annual return on assets (ROA) across the different industries, we find a strong negative relationship between VC disbursements and subsequent industry operating performance. U.S. venture capitalists are the envy of the world. VCs have skills in helping bring young, unpolished firms public at high valuations. Clearly, managers of these firms and investors in the VC funds can profit from this process. We find, however, that there may be negative ramifications of high VC funding. As money pours into young, illiquid, nonpublicly traded firms seeking more innovative techniques and patients, the returns of companies already public suffer as a result. There appears to be a slow diffusion of information concerning VC disbursements into the stock market. The rest of this article is as follows. In Section 2, we describe our sample. In Section 3, we motivate our estimation procedure. Section 4 reports our empirical findings. We conclude in Section 5. 3
2. Sample 2.1. Data Our source for the venture capital data is VenturExpert from Thomson Financial Economics. We use data on all venture capital disbursements between 1980 and 2005. The VenturExpert database classifies industries in its own way, so we re-code these industries into the Fama and French (1997) 48 industry classifications by comparing at individual VenturExpert industry codes and SIC codes. Fama and French (1997) match SIC codes into their classification. Fama and French industries with no VC funding, such as tobacco products, toys, and textiles, are excluded from the analysis. Thus, a total of 34 of the 48 industries remain in our sample. Including all 48 industries in the analysis does not change the empirical results. Compustat provides the information on all industry accounting variables. Four main classifications of funding are available in VenturExpert. These are Early Stage (Seed, Startup, Other Early Stage), Expansion, Later Stage, and Other Stage (Special situation Acquisition, VC Partnership). According to Thomson Venture Economics, Seed capital supports an entrepreneur in developing her idea to the stage where it could qualify for startup capital financing. This includes creating a business plan, product development, and market research. Startup capital is for companies that are completing development, but have not sold their products yet. Other Early Stage is the first round of financing for companies who have received startup financing. This round tends to involve institutional venture funds, while the first two rounds (i.e., Seed and Startup) usually involve angel investors. 4
Expansion financing is for companies who are making and selling their products and have a growing customer base, but might not be profitable yet. Last, Later Stage financing is for companies that have achieved a stable growth rate and are more likely to be profitable or to have positive cash flow. We exclude the Other Stage category (Acquisitions, Special Situation, and VC Partnership funding) because it is not clear that these would lead to increased competition and innovation in an industry. Including these categories does not change our results, however. We briefly use Google Inc. to serve as an example of our data set. Google had two rounds of venture capital funding according to the VenturExpert database. On June 4, 1999, three venture capitalists (Angel Investors, Kleiner Perkins Caufield & Byers, and Sequoia Capital) invested a total of $25 million in an Other Early Stage financing round. On September 1, 2000, both Kleiner Perkins Caufield & Byers (KPCB) and Sequoia Capital invested a combined $15.175 million in an Expansion round of financing. When the firm went public in August of 2004, all three of the venture capitalists had seats on Google s nine member board. Figure 1 presents the time series of quarterly venture capital disbursements compared to the quarterly level of Nasdaq while Figure 2 reports the relationship between Nasdaq levels and VC disbursements scaled by industry total assets. VC disbursements had a slight decline during the 1984 to 1995 period. Disbursements and Nasdaq both report an enormous spike during the internet bubble period of 1998-2001. These two figures present the patterns on the aggregate venture capital funding level. In the rest of 5
our analysis, we will focus on the patterns of investing within a particular Fama-French Industry. Table 1 provides descriptive statistics of rounds and VC disbursements broken down by the Fama and French s (1997) industry classifications with at least some funding during our time period. During our sample period, there was an aggregate $394.2 billion in disbursements by venture capitalists in 73,346 separate rounds. Thus, there were over 73,000 separate investments by venture capitalists into non-publicly traded firms during the sample time period. In our analysis, we will focus on the dollar amount invested, not on the number of rounds that venture capitalists engaged in. Thus, in a quarter, five rounds of $1 million each will count the same as one round of $5 million. Figure 3 reports the quarterly VC disbursements scaled by total funding and categorized by financing stage. In all but 8 quarters, Expansion financing had the largest share of total funding. Startup financing s relative share generally decreased during the time period. As might be expected, Seed financing was usually the lowest percentage category, never amounting to more than 8.9% of all VC disbursements in a particular quarter. Not all industries are as likely to receive financial interest from venture capitalists. Over 83% of the total venture capital disbursements went to only five industries: Business Services (37.3%), Telecommunications (24.4%), Pharmaceutical Products (9.2%), Computers (6.6%), and Chips and Electronic Equipment (5.7%). Figure 4 shows the time-series trend in relative disbursements in the highest funded Fama- French industries. 6
In the early 1980s, the Computer Industry had large relative funding, accounting for over 45% of all disbursements in the second quarter of 1982. In the late 1980s, Business Services surpassed Computers as the most funded industry. During the bubble period, Business Services accounted for almost half of VC dollars while Telecommunications received approximately 30% of all dollars. 2.2. Summary Statistics Table 2 reports the summary statistics for our sample. There are a total of 3,502 industry-quarter observations (34 Fama-French Industries multiplied by 103 quarters), during the sample period. Since we are using lagged quarterly VC disbursements, the analysis starts at the end of the first quarter in 1980. The average venture capital disbursement per industry-quarter is $110.1 million. The maximum is $15,174 million (first quarter of 2000 in the Business Services industry) while the median amount of quarterly funding per industry is $7.3 million. In a number of industries, there were no VC disbursements in a particular quarter. When VC dollars are scaled by industry total assets, the mean value over the 103 quarters is 0.21% compared to a median of 0.01%. We obtain both value and equal weighted quarterly industry-level stock returns and book-to-market ratios from Kenneth French s website. The average equally weighted industry stock return is slightly higher than the corresponding valued weighted returns (4.1% versus 3.8%). The average value weighted quarterly book-to-market ratio is 0.58 compared to 0.54 for the equally weighted time series. As control variables in our regressions, we include the natural log of the number of IPOs in an industry-quarter plus one and a measure of industry competition (the 7
Herfindahl Index). Our data source for 6,620 IPOs over 1980-2005 is the Thomson Financial Securities Data (also known as Securities Data Co.) new issues database. To remove very small IPOs, we require IPOs to have an offer price of at least $5.00 per share. In all of our analysis, we exclude best efforts offers; American Depository Receipts (ADRs); closed-end funds; real estate investment trusts; limited partnerships; and firms not listed on CRSP within six quarters of the offering. Lowry (2003) links the level of IPO volume with investor sentiment. The Herfindahl Index is computed by dividing the sum of the squares of the sales (Compustat item #12) in a Fama-French industry by the squared total sales of that industry. A high value for the Herfindahl Index indicates that the industry has a low level of competition. In our sample, the industry with the smallest quarterly Herfindahl Index (0.01) is Computers while the industry with the highest quarterly index (0.73) is Mines. In the top five VC funded industries, the average quarterly Herfindahl Indexes range from 0.01 to 0.23. The last row of Table 2 reports that the mean Herfindahl Index across all industries is 0.14 compared to a median quarterly value of 0.10. Hou and Robinson (2007) show that the Herfindahl index is related to subsequent industry returns. 3. Estimation We estimate our models using panel regressions with both quarterly and industry dummies. We do not run Fama-MacBeth (1973) regressions because our panel is only 34 industries wide in a given quarter, and at the beginning of the sample time period, many of the quarter-industry data points (i.e., VC disbursements) are zero. Since the first stage 8
of the Fama-Macbeth procedure would involve quarterly cross sectional regressions, we would have too few degrees of freedom to obtain any power. If we do run Fama-MacBeth regressions, all variables are insignificant, including the book-to-market ratio. Petersen (2007) states that the Fama-MacBeth methodology is equivalent to a fixed effects regression if one properly controls for cross-sectional correlation, as we describe below. Furthermore, fixed effects regressions have the advantage of permitting us to control for industry effects, which is not possible with the Fama-MacBeth procedure. As Petersen (2007) notes, a failure to control for industry effects might bias the standard errors. Since many of the industry fixed effects are statistically significant, we think they should be controlled for. We include both time-period and industry fixed effects wherever possible in our analysis. Clustering can be present, however, even after including cross-sectional and time effects according to Bertrand, Duflo, and Mullainathan (2004). Thus, we also cluster the standard errors by both quarter and industry. To cluster standard errors using non-nested clusters, we use the methodology proposed by Thompson (2006) and Cameron, Gelbach, and Miller (2006). As Thompson (2006) explains, clustering the standard errors by just one variable or not clustering them at all can bias the standard errors when both cross-sectional and time-series effects are present. Clustered standard errors are always heteroskedasticity-consistent. In cases where industry-level variables are used, we cannot also use time dummies due to collinearity, so we omit the time dummies. We find that the choice of quarterly dummies or industry-level variables does not change our results, however. 9
4. Empirical Results We want to test whether VC funding in a particular industry is related to industry returns in subsequent quarters, controlling for variables which have been shown to have some predictive power in explaining future returns. In one specification, we also control for the contemporaneous Fama and French (1993) factors plus momentum. As control variables, we use Fama-French industry book-to-market ratios, number of initial public offerings (IPOs), two quarters of past industry returns, and an industry Herfindahl Index. In the fixed-effect regressions, we generally also include industry and quarterly dummies. There are 3,502 quarterly observations (34 different industries multiplied by 103 quarters) when industry returns has only one lag. In regressions when two quarterly industry return lags are explanatory variables, the sample size drops to 3,468 quarterly observations. The main results are presented in Tables 3A and 3B. In all the regressions (except for the ones using the Fama-French factors), the dependent variable is either the value weighted or equally weighted quarterly industry return. In the regressions with the Fama- French factors, the dependent variable is excess returns over the Treasury risk-free rate. This choice of dependent variable is only to keep with common practice; using raw or excess returns for any of the models does not change the significance of our results. Table 3A uses Venture Capital Dollars as the independent variable of interest and presents different specifications ranging from this variable alone to a purely predictive model with control variables. The specification is as follows: VW or EW Industry Return i,t = a 0 + a 1 VC Dollars i,t-1 + a 2 Book/Market i,t-1 + a 3 Industry Return i,t-1 + a 4 Industry Return i,t-2 + a 5 Log(N IPOs +1) i,t-1 +a 6 Herfindahl i,t-1 + a 7 (Mkt-Rf) t + a 8 SMB t + a 9 HML t + a 10 MOM t + Industry Dummies + Quarterly Dummies + e i,t 10
T-statistics generated from the double-clustered errors (i.e., both industry and quarter) are in parentheses under the coefficients. In the models with the Fama-French factors, we could not incorporate quarterly dummies because of collinearity with the economy-wide Fama-French and momentum factors. The coefficient on VC Dollars is stable across specifications and consistently negative. The t-statistics on the VC Dollar coefficient in Table 3A range from -2.04 to - 3.75. Looking at the coefficient from specification (2) which is a purely predictive regression with all of the controls and dummies, the economic significance of this variable is that $500 million in additional quarterly venture capital investment in one industry would decrease the subsequent quarterly value-weighted return by approximately (1.21 / 1000) * 500 = 0.605 of a percent, or 61 basis points. The result for equal-weighted returns is slightly larger with a VC Dollar coefficient of -1.36 in model (5). The negative coefficients on VC Dollars are consistent with a slow diffusion of the information content of the venture capital investment by the stock market. Our result is similar to Hong, Torous, and Valkanov (2007) who find that the stock market reacts with a delay to new information contained in industry returns. Information on fundamentals appears to be diffused only gradually across markets. Other independent variables that are often significant in the specification include industry book-to-market and the first quarterly lag of industry returns. The positive and significant coefficient on book-to-market for the value weighted industry returns implies that when the industry book-to-market ratio is high (i.e., tilted towards value), subsequent returns are higher, all else being equal. The coefficient on the log number of IPOs is 11
consistently negative as might be predicted from Baker and Wurgler (2000), but not significant. The Herfindahl index is never significant predictor of quarterly industry returns. This differs from the findings in Hou and Robinson (2007), but their study uses 2-digit SIC industries which are much more numerous, and they are able to obtain a longer time series. Comparing the coefficients using the raw industry returns as the dependent variable (columns (2) and (5)) with the coefficients controlling for contemporaneous Fama-French factors (columns (3) and (6)) reports minor differences. When industry quarterly returns are valued weighted, the coefficient on VC Dollars is -1.21 (t-statistic of -2.84). When excess quarterly industry returns over Treasuries are the dependent variable, the coefficient on VC Dollars is -0.83 (t-statistic of -3.75). Table 3B replicates the results using lagged VC Dollars divided by industry assets instead of simply lagged VC Dollars. In this table, the dependent variable is raw returns for all regressions except the ones which have the contemporaneous Fama-French factors, in which case excess returns are used. The t-statistics are much larger for this variable than they were for the unscaled variable in Table 3A. In Table 3B, the t-statistics range from -4.01 to -6.25. Scaling the VC disbursements may provide a better gauge of the true magnitude of the funding amount. That is, $1 billion of disbursements in a very large industry may not be expected to have the same impact as if the $1 billion disbursement was made in a relatively small industry. As before, book-to-market is significant only when the value weighted industry returns are used while lagged industry returns are significant in the regressions only with equally weighted industry returns as the dependent variable. Using column 2 as before to 12
gauge the economic significance of the coefficient values, VC investment totaling one percent of an industry s total assets is associated with a subsequent quarterly return that is 31 basis points lower for value weighted return and 34 basis points lower for equal weighted return. 4.1. Time-period sub-sample regressions One might think that solely the observations from the time period referred to as the bubble period of 1998-2001 might have been driving this result. Recall that Figures 1 and 2 reported a spike in both the level of Nasdaq and in the level of venture capital disbursements during the internet bubble period. To test this hypothesis, we break down the sample into bubble and non-bubble periods and re-run the tests from columns (2) and (6) of Tables 3A and 3B on each sub-sample. The results appear in Tables 4A and 4B. Table 4A uses VC Dollars as the explanatory variable of interest while Table 4B uses VC Dollars scaled by industry assets. Both industry and quarterly dummies are included in all the regressions. As before, both value and equal weighted industry quarterly returns are the dependent variables. While the t-statistics tend to be stronger for the bubble period, the coefficients on VC Dollars and VC Dollars/Assets are smaller in the bubble period than they are for the non-bubble period. In only one of the 12 regressions contained in Tables 4A and 4B is the VC disbursement coefficient not significant at conventional levels. In Table 4A, the non-bubble period regression with the equally weighted industry returns as a t-statistic of 13
only -1.60. Interestingly, both bubble and non-bubble periods have larger coefficients than the combined period. 4.2. Types of Venture Capital Funding We are interested to know if only one type of venture capital funding is driving these results. That is, are our results being driven by venture capital Expansion financing? If our premise is correct, all types of VC funding should be related to returns, because they all can lead to increased competition and innovation in an industry by funding new firms. To test this which types of VC funding lead to lower returns, we replicate regressions (2) and (5) in Tables 3A and 3B by type of venture capital funding, and the results appear in Tables 5A and 5B. The model in Table 5A is the following: VW or EW Industry Return i,t = a 0 + a 1 VC Dollars i,t-1 + a 2 Book/Market i,t-1 + a 3 Industry Return i,t-1 + a 4 Industry Return i,t-2 +a 5 Log(N IPOs +1) i,t-1 +a 6 Herfindahl i,t-1 + Industry Dummies + Quarterly Dummies + e i,t In this model, VC Dollars is now defined as the dollars from each stage of funding, as specified in the columns of the table. As before, the errors are clustered by quarter and industry. The control variables retain fairly stable coefficients, while all of the VC funding measures are negatively related to returns. Across the 20 regressions in Tables 5A and 5B, all of the coefficients (except for Seed financing using value weighted returns) are negative and statistically significant at conventional levels. The magnitude of the coefficients is inversely related to the amount of funding dollars in that particular category in our sample, which is detailed in Table 1. The coefficients from the seed 14
financing are largest, followed by the Other Early Stage, Later Stage, Startup and Expansion stages. This pattern is repeated when using VC dollars divided by industry assets, in Table 5B. Again, the results are slightly stronger when dividing by total industry assets. 4.3. Industry-Level OLS In order to make sure that one industry or an omitted industry-level variable is not driving the results, we test the relation with individual industry-level OLS tests on the 5 largest industries, which together make up 83.2% of the VC funding in our sample period. These industries are Business Services (37.3%), Telecom (24.4%), Drugs (9.2%), Computers (6.6%), and Chips (5.7%). The results appear in Tables 6A and 6B. All of the coefficients on VC funding are negative, but the coefficient for VC funding in the Drugs industry is not significant when the dependent variable is valueweighted returns. When equal-weighted returns are used, Drugs is significant but Business Services and Computers are not. We obtain similar results when we scale the VC funding dollar amount by industry total assets, in Table 6B. In both Table 6A and Table 6B, we observe the pattern found in Table 5 that in the industries that have the greatest amount of total funding, additional funding has a smaller effect on subsequent returns. The five industries are presented in decreasing order of the amount of VC funding in the sample, and the coefficients for the VC funding variable in the value-weighted returns section Table 6A are -1.49, -1.59, -4.54, -20.89, and -36.63, respectively. The same pattern is observed in the equal weighted section of the table, as well as in Table 6B. 15
This pattern suggests that the industries with less total VC funding are more sensitive to VC funding when it does come. This is possibly because in industries with less VC funding, fewer firms benefit from the advantages that venture capitalists provide, and so firms that do secure the funding and backing are at a greater advantage. 4.4. Additional robustness tests We have included several additional variables in the model for robustness. These were not significant and did not change our results. One of these variables is the prior 12- month s dividend yield, which is a typical predictor of returns in the asset pricing literature (see Boudoukh, Michaely, Richardson, and Roberts (2007)). This variable is highly correlated with industry book-to-market, and so was not included in the model. Including it does not change our results. Other variables which did not change our results included IPO first day returns, additional lags of the log number of IPOs, and industry market capitalization. 4.5. Industry Operating Performance To determine why there is a negative relationship between VC funding and stock returns, one could examine the subsequent operating performance of the particular industry. Prior research has examined operating performance after IPOs (Jain and Kini (1994) and Mikkelson, Partch, and Shah (1997)), seasoned equity offerings (Loughran and Ritter (1997)), dismissal of top managers (Denis and Denis, 1995), and stock market liberalization (Mitton (2006)). If quarterly stock returns are indeed lower with higher levels of VC disbursements, one might expect to see the industry realize lower levels of 16
future operating performance. That is, poor stock and operating performance should be expected to go hand and hand. In Table 7, we examine this relationship using annual accounting data. Each year, for each industry, during the 1981-2005 time period, we run the following fixed-effect regression: Industry ROA i,t = a 0 + a 1 VC Dollars i,t-1 + a 2 (VC Dollars/Assets) i,t-1 + a 3 CAPEX/Assets i,t-1 + a 4 R&D/Asserts i,t-1 + Industry Dummies + Yearly Dummies + e i,t In the regression, the dependent variable is industry ROA, defined as the aggregate industry net income before extraordinary items (Compustat item #18) scaled by aggregate industry total assets (Compustat item #6). That is, for each industry in a given year, ROA is defined as the summation of all the net income before extraordinary items in an industry and dividing by summation of all the total assets in an industry. The independent variables are all lagged by one year. Four different explanatory variables are used: industry VC dollars (in millions), industry VC dollars/industry total assets, aggregate industry capital expenditures (Compustat item #30) scaled by industry total assets, and aggregate industry research and development (Compustat item #46) scaled by industry total assets. In all six of the regressions, there are 850 observations (850 = 34 different Fama- French Industries * 25 years of data). Since we are summing the accounting variables across all firms in each industry for every ratio, all the accounting ratios are effectively value weighted. In column 1 of Table 5, the only explanatory variable is the prior year s industry VC dollars. The coefficient value of -1.11 on VC dollars is highly significant (tstatistic of -4.08). As before, all t-statistics use errors that are double clustered by 17
industry and year. The negative coefficient implies that the higher are industry VC dollars in the prior year; the lower is next year s ROA. Column 2 reports that the VC dollar coefficient remains significant when industry and calendar year dummies are included in the fixed-effect regressions. The next column reports that the VC dollar coefficient is still negative and statistically significant when CAPEX/asset, R&D/assets, as well as industry and calendar year dummies are included in the regressions. The last three columns of the table report that when VC dollars scaled by industry total assets are used instead of VC dollars, the main results stay strong. This industry operating performance evidence is consistent with our stock return findings. All else being equal, higher levels of VC disbursements subsequently leads to lower stock returns as well as worse operating performance in the industry. 5. Summary and Conclusion Venture capitalists clearly play an important role in the U.S. economy. During the 1980-2005 time period, venture capital disbursements totaled $394.2 billion in nonpublicly traded companies over 73,346 unique rounds of investing. Quite a number of these investments enabled young firms to issue public equity at relatively high valuations. For some investors and managers, enormous wealth was created. We find, however, that there may be a downside to high levels of VC funding. VC investing in young nonpublicly traded firms appears to lead to lower valuations and worse operating performance for industry firms that are already trading on a major stock exchange. 18
Using quarterly Fama and French industry stock returns, we find that there is a negative relationship between VC funding and subsequent industry stock returns. Our results are generally robust to whether the industry returns are equal or value weighted. Interestingly, the value weighted returns are usually more significant than the equally weighted returns. We report that scaling VC investing by the aggregate total assets of the industry produces stronger results than when raw VC dollars are used in the fixed-effect regressions. Evidence is offered that the effect is found in both the bubble and nonbubble periods of our sample. When the fixed-effect regressions focus on the type of VC funding, the coefficients on VC dollars and dollars scaled by industry assets are consistently negative. When regressions are independently run on the five most funded Fama-French Industries, the general patterns remain robust. Using annual Compustat data, we find that higher VC funding is also related to lower subsequent operating performance for the particular Fama and French industry. The poor subsequent industry operating performance is consistent with the evidence of lower industry stock returns. As money pours into potential competitors who are not public yet, both the operating and stock performance of an industry suffers. The results are consistent with a slow dissemination of information into the valuations of publicly traded firms. 19
References Baker, M., Wurgler, J., 2000. The equity share in new issues and aggregate stock returns. Journal of Finance 55, 2219-2257. Bertrand, M., Duflo, E., Mullainathan, S., 2004. How much should we trust differences-in-differences estimates? Quarterly Journal of Economics 119, 249-75. Boudoukh, J., Michaely, R., Richardson, M., and Roberts, M., 2007. On the importance of measuring payout yield: Implications for asset pricing. Journal of Finance 62, 877-915. Brav, A., Gompers, P., 1997. Myth or reality? The long-run underperformance of initial public offerings: Evidence from venture and nonventure capital-backed companies. Journal of Finance 52, 1791-1821. Cameron, C., Gelbach, J., Miller, D., 2006. Robust inference with multi-way clustering. University of California-Davis working paper. Denis, D., Denis D., 1995. Performance changes following top management dismissals. Journal of Finance 50, 1029-1057. Fama, E., French, K., 1993. Common risk factors in the returns on stocks and bonds. Journal of Financial Economics 33, 3-56. Fama, E., French, K., 1997. Industry costs of equity. Journal of Financial Economics 43, 153-193. Fama, E., MacBeth, J., 1973. Risk, return and equilibrium: Empirical tests. Journal of Political Economy 71, 607-636. Gompers, P., Lerner, J., 2000. Money chasing deals? The impact of fund inflows on private equity valuations. Journal of Financial Economics 55, 281-325. Hong, H., Torous W., Valkanov, R., 2007. Do industries lead stock markets? Journal of Financial Economics 83, 367-396. Hou, K., and D. Robinson., 2007. Industry concentration and average stock returns. Forthcoming, Journal of Finance. Jain, B., Kini, O., 1994. The post-ipo operating performance of IPO firms. Journal of Finance 49, 1699-1726. Lerner, J., 1994. Venture capitalists and the decision to go public. Journal of Financial Economics 35, 293-316. 20
Loughran, T., Ritter, J., 1997. The operating performance of firms conducting seasoned equity offerings. Journal of Finance 52, 1823-1850. Lowry, M., 2003. Why does IPO volume fluctuate so much? Journal of Financial Economics 67, 3-40. Mikkelson, W., Partch, M., Shah, K., 1997. Ownership and operating performance of companies that go public. Journal of Financial Economics 44, 281-3073 Mitton, T., 2006. Stock market liberalization and operating performance at the firm level. Journal of Financial Economics 81, 625-647. Petersen, M., 2007. Estimating standard errors in finance panel data sets: comparing approaches. Northwestern University working paper. Thompson, Samuel B., 2006. Simple formulas for standard errors that cluster by both firm and time. Harvard University working paper. White, H., 1980. A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity. Econometrica 48, 817-838. 21
Fig. 1. Quarterly Level of Nasdaq and Venture Capital Disbursements, 1980-2005 Nasdaq Level 0 2000 4000 6000 0 10 20 30 Level of VC Funding in Billions$ 1980 1985 1990 1995 2000 2005 Nasdaq VC Funding Fig. 2. Quarterly Level of Nasdaq and Venture Capital Disbursements Scaled by Industry Total Assets, 1980-2005 Nasdaq Level 0 2000 4000 6000 0.5 1 1.5 VC Funding/Total Assets, % 1980 1985 1990 1995 2000 2005 Nasdaq VC Funding/Total Assets 22
Fig. 3. Venture Capital Disbursements as a Percent of Total Funding by Stage, 1980-2005 Proportion of Total Funding 0.2.4.6.8 1980 1985 1990 1995 2000 2005 Year Seed Other Early Later Stage Startup Expansion Fig. 4. Venture Capital Disbursements as a Percent of Total Funding for the Five Most- Funded Industries, 1980-2005 Proportion of Total Funding 0.2.4.6 1980 1985 1990 1995 2000 2005 Year BusSv Drugs Chips Telcm Comps 23
Table 1 Aggregate Venture Capital Disbursements Categorized by the Fama and French Industries, 1980-2005 Our data source for the venture capital data is VenturExpert from Thomson Financial Economics. All firms are classified into Fama and French (1997) industries. All disbursements are in one of five stages: Seed, Startup, Other Early Stage, Expansion, or Later Stage. Thirty four different industries have at least one round of VC disbursements. Total % Other in $ Total Total Early Later FF Industry Billions $ Rounds Seed Startup Stage Expan. Stage 1 Agric 1.7 0.40% 699 2.75% 23.27% 10.01% 47.01% 16.97% 2 Food 2.2 0.60% 674 2.17% 10.69% 9.47% 57.40% 20.27% 3 Soda 0.1 0.00% 63 0.00% 33.86% 6.14% 46.21% 13.79% 4 Beer 0.2 0.00% 69 0.18% 14.26% 2.40% 75.48% 7.68% 7 Fun 3.3 0.80% 818 1.40% 9.91% 3.46% 71.20% 14.04% 8 Books 0.4 0.10% 233 1.64% 25.90% 9.35% 53.55% 9.56% 9 Hshld 1.7 0.40% 636 5.69% 12.74% 4.72% 49.38% 27.46% 10 Clths 0.4 0.10% 259 1.50% 9.98% 8.44% 59.99% 20.08% 11 Hlth 6.2 1.60% 1,600 2.74% 20.28% 6.65% 51.56% 18.78% 12 MedEq 14.9 3.80% 3,553 3.43% 18.71% 5.59% 47.16% 25.11% 13 Drugs 36.1 9.20% 6,248 2.92% 19.84% 7.63% 48.42% 21.19% 14 Chems 1.3 0.30% 477 2.41% 27.43% 9.11% 45.57% 15.48% 17 BldMt 2 0.50% 812 4.60% 15.47% 6.75% 51.72% 21.46% 18 Cnstr 0.8 0.20% 360 1.30% 4.22% 7.42% 75.04% 12.01% 21 Mach 3.7 0.90% 1,965 2.27% 16.44% 10.24% 49.23% 21.81% 22 ElcEq 0.8 0.20% 195 1.67% 16.64% 5.41% 63.61% 12.67% 28 Mines 0.5 0.10% 31 0.09% 2.40% 0.64% 96.83% 0.03% 29 Coal 0.1 0.00% 28 0.53% 9.87% 7.80% 81.79% 0.00% 30 Oil 1.6 0.40% 334 0.81% 13.49% 15.28% 64.40% 6.03% 31 Util 1.5 0.40% 398 1.11% 18.39% 7.26% 55.96% 17.27% 32 Telcm 96.2 24.40% 11,926 1.63% 18.51% 5.22% 54.55% 20.08% 33 PerSv 0.7 0.20% 188 0.94% 9.87% 2.27% 67.27% 19.65% 34 BusSv 147 37.30% 26,993 2.18% 18.54% 6.73% 53.92% 18.63% 35 Comps 26.2 6.60% 5,556 2.14% 21.51% 6.20% 51.55% 18.59% 36 Chips 22.3 5.70% 4,393 2.47% 20.26% 8.16% 48.40% 20.71% 37 LabEq 5.4 1.40% 1,236 2.10% 15.40% 9.56% 50.39% 22.55% 38 Paper 0.6 0.20% 209 4.85% 28.67% 2.10% 41.29% 23.09% 39 Boxes 3.8 1.00% 888 0.69% 14.25% 4.34% 63.90% 16.83% 42 Rtail 3.2 0.80% 840 1.70% 14.92% 5.94% 61.42% 16.02% 43 Meals 2 0.50% 525 2.18% 12.62% 3.77% 64.72% 16.71% 44 Banks 3.1 0.80% 581 2.00% 14.11% 12.68% 62.99% 8.23% 45 Insur 1 0.20% 148 0.98% 42.54% 2.59% 42.34% 11.54% 46 RlEst 0.6 0.20% 183 0.37% 20.27% 12.06% 35.36% 31.93% 47 Fin 2.3 0.60% 228 0.20% 8.43% 8.71% 59.22% 23.44% Total 394.2 100.00% 73,346 1.87% 17.17% 6.89% 57.32% 16.76% 24
Table 2 Summary Statistics by industry and quarter, 1980-2005 VC Dollars are the lagged aggregate quarterly VC disbursements, in millions of dollars, within a particular Fama-French industry. VC Dollars/Assets are the aggregate quarterly VC disbursements, divided by total assets (Compustat item #6) within an industry. Both the value weighted and equally weighted industry returns are from Kenneth French s website. Book/Market is the lagged industry-level average book-to-market ratio provided on Kenneth French s website. Log(N IPOs +1) is the lagged log of the number of IPOs in that quarter plus one. The Herfindahl index is computed by dividing the sum of the squares of the sales (Compustat item #12) in each Fama-French industry by the squared total sales of that industry. There are 3,502 observations for each variable (34 Fama- French industries multiplied by 103 quarters). Variable N Mean Min Max Median VC Dollars (M) 3,502 110.1 0 15,174 7.3 VC Dollars/Assets 3,502 0.21% 0.00% 45.38% 0.01% Industry Return - VW 3,502 3.8% -42.2% 68.6% 4.1% Industry Return - EW 3,502 4.1% -41.8% 78.8% 3.5% Book/Market - VW 3,502 0.58 0.09 1.81 0.51 Book/Market - EW 3,502 0.54 0.08 1.74 0.47 Log(N IPOs +1) 3,502 0.28 0 3.43 0 Herfindahl 3,502 0.14 0.01 0.73 0.10 25
Table 3A Fama-French industry fixed effects regressions, 1980-2005 The regression dependent variable is the value weighted (VW) or equally weighted (EW) Fama and French industry raw quarterly return. Columns (3) and (6) use excess returns over the risk-free rate provided on Kenneth French s website. VC Dollars i,t-1 are the lagged aggregate quarterly VC disbursements within a particular Fama-French industry. Book/Market is the lagged industry-level average book-to-market ratio provided on Kenneth French s website. Log(N IPOs +1) is the lagged log of the number of IPOs in that quarter plus one. Book-to-market and lagged returns are value weighted when the dependent variable is value weighted, and equal weighted otherwise. Mkt-Rf, SMB, HML and MOM are the contemporaneous market return minus the risk free rate, Small- Minus-Big, High-Minus-Low book-to-market and momentum factors from Kenneth French s website. T-statistics (in parentheses) use errors that are heteroskedasticity-robust and clustered by industry and quarter. VW or EW Industry Return i,t = a 0 + a 1 VC Dollars i,t-1 + a 2 Book/Market i,t-1 + a 3 Industry Return i,t-1 + a 4 Industry Return i,t-2 +a 5 Log(N IPOs +1) i,t-1 +a 6 Herfindahl i,t-1 + a 7 (Mkt-Rf) t + a 8 SMB t + a 9 HML t + a 10 MOM t + Industry Dummies + Quarterly Dummies + e i,t Value Weighted Returns Equal Weighted Returns Raw Excess Raw Excess Variable (1) (2) (3) (4) (5) (6) VC Dollars i,t-1-1.33-1.21-0.83-1.46-1.36-1.14 (-2.68) (-2.84) (-3.75) (-2.17) (-2.29) (-2.04) Book/Market i,t-1 3.68 4.33-1.19 0.74 (2.43) (5.98) (-0.65) (0.55) Industry Return i,t-1 0.07 0.04 0.17 0.09 (1.67) (1.89) (3.14) (3.18) Industry Return i,t-2 0.05 0.00 0.00-0.01 (1.35) (0.20) (0.11) (-0.71) Log(N IPOs +1) i,t-1-0.08-0.51-0.35-0.68 (-0.16) (-1.05) (-0.84) (-1.52) Herfindahl i,t-1 3.14 2.22 1.54 0.99 (1.26) (0.82) (0.53) (0.31) (Mkt-Rf) t 0.95 0.88 (25.13) (16.05) SMB t 0.15 1.04 (2.17) (7.73) HML t 0.14 0.29 (1.80) (2.57) MOM t 2.98 6.94 (0.57) (1.01) Constant 4.63 7.32-3.49 19.08 7.63-5.23 (7.84) (3.72) (-0.65) (36.25) (1.73) (-0.72) Industry Dummies Yes Yes Yes Yes Yes Yes Quarterly Dummies Yes Yes No Yes Yes No N 3,502 3,468 3,468 3,502 3,468 3,468 R 2 0.56 0.56 0.52 0.65 0.65 0.59 26
Table 3B Fama-French industry fixed effects regressions, 1980-2005 The regression dependent variable is the value weighted (VW) or equally weighted (EW) Fama and French industry raw quarterly return. Columns (3) and (6) use excess returns over the risk-free rate provided on Kenneth French s website. (VC Dollars/Assets) i,t-1 are the lagged aggregate quarterly VC disbursements, divided by total assets (Compustat item #6) within a particular Fama-French industry. Book/Market is the lagged industrylevel average book-to-market ratio provided on Kenneth French s website. Log(N IPOs +1) is the lagged log of the number of IPOs in that quarter plus one. Book-to-market and lagged returns are value weighted when the dependent variable is value weighted, and equal weighted otherwise. Mkt-Rf, SMB, HML and MOM are the contemporaneous market return minus the risk free rate, Small-Minus-Big, High-Minus-Low book-tomarket and momentum factors from Kenneth French s website. T-statistics (in parentheses) use errors that are heteroskedasticity-robust and clustered by industry and quarter. VW or EW Industry Return i,t = a 0 + a 1 (VC Dollars/Assets) i,t-1 + a 2 Book/Market i,t-1 + a 3 Industry Return i,t-1 + a 4 Industry Return i,t-2 +a 5 Log(N IPOs +1) i,t-1 +a 6 Herfindahl i,t-1 + a 7 (Mkt-Rf) t + a 8 SMB t + a 9 HML t + a 10 MOM t + Industry Dummies + Quarterly Dummies + e i,t Value Weighted Returns Equal Weighted Returns Raw Excess Raw Excess Variable (1) (2) (3) (4) (5) (6) (VC Dollars/Assets) i,t-1-33.10-30.90-22.22-36.00-34.09-27.64 (-4.92) (-5.64) (-6.25) (-4.01) (-4.55) (-4.39) Book/Market i,t-1 3.86 4.54-1.03 1.03 (2.46) (6.57) (-0.54) (0.77) Industry Return i,t-1 0.07 0.04 0.17 0.09 (1.72) (1.97) (3.19) (3.30) Industry Return i,t-2 0.05 0.01 0.00-0.01 (1.43) (0.26) (0.11) (-0.68) Log(N IPOs +1) i,t-1-0.08-0.50-0.34-0.65 (-0.15) (-1.02) (-0.82) (-1.45) Herfindahl i,t-1 3.24 2.43 1.64 1.08 (1.29) (0.84) (0.57) (0.33) (Mkt-Rf) t 0.95 0.88 (25.55) (16.12) SMB t 0.14 1.03 (2.07) (7.67) HML t 0.13 0.28 (1.77) (2.54) MOM t 2.67 6.42 (0.51) (0.95) Constant -1.98 7.31-3.35 0.52 7.31-4.90 (-4.52) (3.70) (-0.63) (0.33) (1.62) (-0.68) Industry Dummies Yes Yes Yes Yes Yes Yes Quarterly Dummies Yes Yes No Yes Yes No N 3,502 3,468 3,468 3,502 3,468 3,468 R 2 0.56 0.56 0.52 0.56 0.65 0.59 27
Table 4A Fama-French industry fixed effects regressions by time periods The regression dependent variable is the value weighted (VW) or equally weighted (EW) Fama and French industry quarterly return. VC Dollars are the lagged aggregate quarterly VC disbursements, in millions of dollars, within a particular Fama-French industry. Book/Market is the lagged industry-level average book-to-market ratio provided on Kenneth French s website. Log(N IPOs +1) is the lagged log of the number of IPOs in that quarter plus one. Book-to-market and lagged returns are value weighted when the dependent variable is value weighted, and equal weighted otherwise. The coefficient on VC Dollars is multiplied by 1,000. The bubble period is years 1998-2001 while the non-bubble period is years 1980-1997 and 2002-2005. T-statistics (in parentheses) use errors that are heteroskedasticity-robust and clustered by industry and quarter. VW or EW Industry Return i,t = a 0 + a 1 VC Dollars i,t-1 + a 2 Book/Market i,t-1 + a 3 Industry Return i,t-1 + a 4 Industry Return i,t-2 +a 5 Log(N IPOs +1) i,t-1 +a 6 Herfindahl i,t-1 + Industry Dummies + Quarterly Dummies + e i,t Variable Value Weighted Industry Returns All Non- Bubble Bubble All Equal Weighted Industry Returns Non- Bubble Bubble VC Dollars i,t-1-1.21-3.46-2.43-1.36-3.34-2.84 (-2.84) (-2.62) (-4.45) (-2.29) (-1.60) (-3.78) Book/Market i,t-1 3.68 2.80 10.96-1.19-2.35 8.51 (2.43) (2.05) (1.50) (-0.65) (-1.72) (0.66) Industry Return i,t-1 0.07 0.05 0.06 0.17 0.11 0.24 (1.67) (1.68) (0.58) (3.14) (3.53) (1.90) Industry Return i,t-2 0.05 0.04 0.01 0.00 0.00-0.01 (1.35) (0.95) (0.11) (0.11) (0.06) (-0.10) Log(N IPOs +1) i,t-1-0.08 0.01-0.53-0.35-0.20-2.11 (-0.16) (0.04) (-0.23) (-0.84) (-0.46) (-1.04) Herfindahl i,t-1 3.14 2.71 4.65 1.54 1.98-10.10 (1.26) (0.96) (0.45) (0.53) (0.64) (-0.72) Constant 7.32 4.44 3.45 7.63 12.90-22.99 (3.72) (2.75) (0.85) (1.73) (4.87) (-1.40) Industry Dummies Yes Yes Yes Yes Yes Yes Quarterly Dummies Yes Yes Yes Yes Yes Yes N 3,468 2,924 544 3,468 2,924 544 R 2 0.56 0.61 0.48 0.65 0.70 0.57 28
Table 4B Fama-French industry fixed effects regressions by time periods The regression dependent variable is the value weighted (VW) or equally weighted (EW) Fama and French industry quarterly return. (VC Dollars/Assets) i,t-1 are the lagged aggregate quarterly VC disbursements, divided by total assets (Compustat item #6) within a particular Fama-French industry. Book/Market is the lagged industry-level average book-to-market ratio provided on Kenneth French s website. Log(N IPOs +1) is the lagged log of the number of IPOs in that quarter plus one. Book-to-market and lagged returns are value weighted when the dependent variable is value weighted, and equal weighted otherwise. The bubble period is years 1998-2001 while the non-bubble period is years 1980-1997 and 2002-2005. T-statistics (in parentheses) use errors that are heteroskedasticity-robust and clustered by industry and quarter. VW or EW Industry Return i,t = a 0 + a 1 (VC Dollars/Assets) i,t-1 + a 2 Book/Market i,t-1 + a 3 Industry Return i,t-1 + a 4 Industry Return i,t-2 +a 5 Log(N IPOs +1) i,t-1 +a 6 Herfindahl i,t-1 + Industry Dummies + Quarterly Dummies + e i,t Value Weighted Industry Returns Equal Weighted Industry Returns Variable All Non- Bubble Bubble All Non- Bubble Bubble (VC Dollars/Assets) i,t-1-30.90-97.10-68.21-34.09-97.55-73.59 (-5.64) (-4.85) (-10.23) (4.55) (-2.00) (9.09) Book/Market i,t-1 3.86 2.66 11.68-1.03-2.62 9.40 (2.46) (1.99) (1.55) (-0.54) (-1.90) (0.71) Industry Return i,t-1 0.07 0.05 0.08 0.17 0.11 0.27 (1.72) (1.81) (0.74) (3.19) (3.47) (2.02) Industry Return i,t-2 0.05 0.04 0.02 0.00 0.00-0.01 (1.43) (1.12) (0.19) (0.11) (0.00) (-0.10) Log(N IPOs +1) i,t-1-0.08 0.03-0.59-0.34-0.19-2.10 (-0.15) -0.08 (-0.25) (-0.82) (-0.43) (-0.99) Herfindahl i,t-1 3.24 2.89 5.26 1.64 2.14-8.99 (1.29) (1.01) (0.50) (0.57) (0.65) (-0.68) Constant 7.31 4.71 3.19 7.31 13.07-26.12 (3.70) (2.93) (0.77) (1.62) (4.93) (-1.52) Industry Dummies Yes Yes Yes Yes Yes Yes Quarterly Dummies Yes Yes Yes Yes Yes Yes N 3,468 2,924 544 3,468 2,924 544 R 2 0.56 0.61 0.48 0.65 0.70 0.56 29