On the Matching of Companies and Their Financial Intermediaries: Evidence from Venture Capital *

Transcription

1 On the Matching of Companies and Their Financial Intermediaries: Evidence from Venture Capital * ALEXANDER W. BUTLER M. SINAN GOKTAN September 6, 2008 * Butler is at University of Texas at Dallas; Goktan is at California State University East Bay. Please send correspondence to the first author: Department of Finance and Managerial Economics, School of Management, SM 31, The University of Texas at Dallas, 800 W Campbell Rd., Richardson, TX butler@utdallas.edu. Acknowledgements: We thank Ekin Alakent, Andres Almazan, Lee Ann Butler, Chitru Fernando, Robert Kieschnick, Mark LaPlante, Volkan Muslu, Michael Rebello, David Robinson, and seminar participants at the University of Texas at Dallas for helpful comments. Any remaining errors are our own.

2 On the Matching of Companies and Their Financial Intermediaries: Evidence from Venture Capital ABSTRACT We use the venture capital market to examine how companies and their financial intermediaries match together, focusing on the tradeoff between the costs (due to agency problems) and benefits (due to comparative advantage in information production) of matches. Inexperienced VC firms those with the largest potential for agency problems tend to match with young and small companies within close proximity, presumably to enhance soft information production. This finding suggests a tradeoff between the benefits of better information production and costs of agency conflicts between company and financial intermediary. We then quantify an outcome, IPO initial return, from matching that demonstrates this tradeoff. We show empirically that, (only) after controlling for endogeneity in the choice of the intermediary, venture-backed companies that are close to their lead VC firm have substantially lower first day initial returns. Our findings thus rationalize why companies choose to finance through inexperienced venture capitalists that pose high agency costs (those with incentives to grandstand), despite the large expected costs of doing so. JEL Codes: G24, D80 Keywords: Venture capital, soft information, IPO, grandstanding

3 How do the characteristics of financial intermediaries affect the matching between companies and those intermediaries? The question of how the characteristics of financial intermediaries affect how they function is central to the recent literature on how banks locations (Petersen and Rajan [2002], Degryse and Ongena [2004]), incentive structures (Brickley, et al. [2003]), and organizational form (Stein [2002] and Berger, et al. [2005]) affect their lending activities. The main idea behind this literature is that the production of information about a company, particularly soft information, by an intermediary such as a bank is enhanced by the intermediary s location and other characteristics that favor strong incentives for investing in soft information gathering. Soft information here refers to information that cannot be directly verified by anyone other than the agent who produces it, whereas hard information refers to the type of information that can be measured, recorded, and transferred to others (Stein [2002]). If a financial intermediary s characteristics, such as its incentives to produce soft information, affect how it functions, do companies choose their intermediaries based on observable characteristics? Put differently, how do companies and intermediaries choose their mates? This is an important question in our framework because we argue that the very characteristics that drive the matching also lead to possible agency conflicts. Our paper addresses this issue using data on U.S. start-up companies and the venture firms that fund them. The matching process should be particularly important for start-up companies for two main reasons. First, start-ups have a clear need for production of soft information and second, they may find themselves interacting with inexperienced venture capitalists that have well-documented incentives to engage in grandstanding (Gompers [1996]). Grandstanding refers to the tendency of less experienced VC firms to take their portfolio companies public too early in order to increase the VC firm s reputation, which in turn allows the VC firm to raise more capital in the future. Grandstanding is an 1

4 agency problem that has the effect of making venture capital finance particularly expensive for companies that use inexperienced VC firms, because companies that are taken public early are penalized by greater underpricing at the IPO. Thus, grandstanding represents a cost to the start-up company. The supply side (why VCs might engage in this behavior) of grandstanding is intuitive. But on the demand side, why would companies enter into financing agreements with such VC firms if it is likely to be so expensive in terms of the going-public strategies that grandstanding VC firms encourage? We show in this paper that the answer to this demand side question is rooted in the more fundamental question of how firms and intermediaries match with each other. In this particular matching between the VC firm and the start-up company, there is a tradeoff between soft information production and grandstanding. We provide evidence that the same characteristics such as small size and lack of reputation of VC firms that create incentives for them to engage in grandstanding also create incentives for such intermediaries to engage in soft information production. Indeed, we show that it is the difficult-to-evaluate portfolio companies most in need of soft information production that choose to match with inexperienced VC firms. This matching provides a plausible explanation for why grandstanding can persist in equilibrium. To determine the benefits of this matching, we examine soft information production by VC firms and the interaction between information production and grandstanding incentives. There are many ways in which soft information production might manifest itself. We focus on the effect of distance between the VC firm and the portfolio company on the initial returns of portfolio companies that complete an initial public offering. Others show that geographical distance correlates with soft information production (Petersen and Rajan [2002], Berger et al. [2005], Malloy [2005], Butler [2008]), and so we use distance as a proxy for monitoring costs and information 2

5 production by VC firms. This research design allows us to gain direct insights on the tradeoff between information production and grandstanding by venture capitalists. We show that venture-backed companies that are close to their VC firms are, on average, younger and smaller, with relatively low asset tangibility the very companies that most need soft information production. We stress that this is not just a Silicon Valley effect because we control for both the state and the industry of the company in our study. Further, VC firms that invest in close companies are less experienced VC firms the firms that are most likely to engage in grandstanding. This result is intuitive. Monitoring a distant company may be more costly and require greater resources which might be lacking in inexperienced VC firms (Gompers [1996] and Lee and Wahal [2004]). Thus, young companies match with inexperienced VC firms, and they do so at close physical proximity, presumably to enhance soft information production. Our main result, described below, reveals how this matching on geographic proximity affects an important economic outcome for start-up companies the initial return on their IPO. Can a nearby VC firm add value in the IPO process through reduced underpricing? Simple ordinary least squares (OLS) techniques are inadequate to address this issue because it is precisely those companies in close proximity that will both face grandstanding and that will need the most soft information production. Companies with such characteristics are likely to have the greatest underpricing at the IPO stage. This selection bias potentially confounds any beneficial effect of proximity. Indeed, we find this is the case: using an OLS framework as a benchmark model shows that there is no net effect of venture capital distance on IPO underpricing. That is, our simple OLS tests indicate that the near and distant VC firms have no different impact on IPO underpricing. In our main statistical test we use a treatment effects model, a common technique in labor and health economics research, to remove the selection effect due to the endogenous matching of small companies choosing grandstanding VC firms and look at 3

6 the average treatment effect of VC distance from the IPO company. Using a sample of venture-backed IPOs from , we find that distance indeed matters. We show that after controlling for the endogeneity of how firms and VCs match geographically, venture-backed companies within close proximity to their VC firms leave less money on the table at the IPO. Holding other factors constant and controlling for the endogenous choice of distance, companies close to their VC firms have initial returns that are a remarkable 12 percentage points (for the full sample; 7 percentage points when we exclude the internet bubble) or, about 35% of the mean (for the full sample; 33% of the non-bubble period mean), lower than companies distant from their VCs. This reduction in initial return is an outcome of better soft information production arising from proximity. Through its focus on how intermediaries match with firms, our paper brings together two literatures the one on the effect of geography in financial intermediation (Petersen and Rajan [2002], Berger et al. [2005], Butler [2008]) and the one on agency problems in venture capital (Gompers [1996], Lee and Wahal [2004]). Closest to the spirit of our paper is Berger, et al. [2005]. In their two-stage least squares approach, they model both how borrowing companies choose the size of their bank and what economic outcomes result as a function of (the exogenous portion of) bank size. Similarly, we consider both how companies choose their intermediary (their VC firm) and the impact those choices have on related economic outcomes. Fernando, Gatchev, and Spindt [2005] are also interested in how firms and intermediaries (IPO lead managers) match, but they model only the company s choice of underwriter reputation, not the joint decision by both company and the intermediary. One straightforward aspect of our paper is that it is a continuation of Gompers [1996]. His paper documents the strong incentives for inexperienced VC firms to grandstand, but leaves as an open question why start-up companies would have a demand 4

7 for inexperienced VC firms that are likely to grandstand. Gompers [1996] states, While this paper does not address the reasons entrepreneurs seek financing from young venture capital firms who then rush them to the IPO market, the issue deserves greater attention[ ]. We fill this gap with our matching analysis, thereby supplying the demand side explanation for grandstanding. The remainder of this paper is organized as follows. Section 1 gives a brief description of the importance of proximity and soft information in VC industry, Section 2 describes the data, Section 3 presents the results, and Section 4 provides concluding remarks. 1. Importance of Proximity and Soft Information in Venture Capital Industry How can better information production through close proximity add value to a venture-backed company? This section discusses in detail how we anticipate soft information production to be important in the venture capital industry. When information about projects is soft and cannot be credibly transmitted, financing through intermediaries that invest in research and create relationship-based lending with their clients is a better alternative than arms-length finance. Under such circumstances, proximity is important (Coval and Moskowitz [2001], Malloy [2005]). Butler [2008] shows that high-risk bonds and non-rated bonds are more difficult to evaluate and that investment banks with a local presence are better able to assess soft information. As a result, these local investment banks charge lower fees and sell bonds at lower yields. These studies support the positive correlation between proximity and better information production that we try to examine in this paper. In addition to financing, many VC firms provide intensive oversight for the companies in their portfolios. VC firms heavily invest in research on the companies they finance and are involved in soft information production which is needed to evaluate their 5

8 investments. The fact that VC firms provide extensive monitoring and supervising services along with financing makes distance an important factor in the VC industry. Proximity enables VC firms to produce more accurate information and this information will be conveyed to the company more effectively. We argue that distance is especially important for young and small venture-backed companies that are in their early stages since they require extensive supervising. Distance is also important for young and inexperienced VC firms since they have relatively limited resources to monitor a distant portfolio company and they have a comparative advantage at soft information production due to their organizational form. A close VC firm may be able to improve the valuation of a venture-backed company through enhanced screening, monitoring, and certification. Proximity improves screening because VC firms can receive more accurate and credible information about the potential investment opportunities within their region through their network, and as a result, make better investment choices. VC firms may be able to provide close monitoring, through active involvement in the daily business of the company, serving on the board of the company, and matching the company with key customers and suppliers (Lerner [1995], Jeng and Wells [1997]). In addition to those services, location specific advantages such as established networks and industry expertise found in the close vicinity of the VC firm also benefit those companies located in close distance. Consistent with the view that proximity enhances monitoring ability, Tian [2007] documents that venture-backed companies located closer to their VC firms receive fewer financing rounds, receive more investment amount per round, have a higher chance of successful exit and have better operating performance in the IPO year. VC firms facilitate certification for underwriter firms and investors. As Megginson and Weiss [1991] show, venture capitalists tend to use the same underwriters. 6

9 Under these circumstances, a VC firm can credibly provide certification for the company under its close supervision which will increase the perceived value of the company to investors and underwriters. We analyze the relationship between proximity and valuation by looking at the effect of proximity on IPO underpricing. We hypothesize that better information production/sharing as a result of proximity will manifest itself through IPO valuation and proximity will create a decrease in the underpricing of the venture-backed companies at the IPO stage. 2. Data We gather information on an initial sample of 2,742 venture-backed IPOs from the period from the Securities Data Company (SDC) database. Following previous researchers, we eliminate offerings (i) identified as unit offerings (ii) not involving common stock, (iii) of financial firms with SIC codes between , (iv) of very small issues with offer size below 20 million dollars, and (v) for which SDC did not provide information required for our tests. Both the venture-backed IPO company and the VC firm must have their main office or branch office within the U.S. We require the company be in the Center for Research in Security Prices database (CRSP) and the Compustat database. We obtain information on the founding year of the company, IPO volume in a given year, underwriter rank, and a list of internet companies from Professor Jay Ritter s website. After applying all our filters and data requirements, the resulting sample consists of 915 IPOs. To mitigate the effect of outliers or data errors, we winsorize all variables in our models at 1% and 99% levels. To determine the distance between the VC firm and the venture-backed IPO, we first identify the lead investment firm which is the firm in the syndicate that typically undertakes the main task of monitoring and consulting (Gompers [1996]). The co- 7

10 investors in the syndicate are involved with the business of the financed firm to a considerably lesser degree and so their proximity is, arguably, not as important (Wright and Lockett [2003]). Measuring distance only between the lead VC firm and the company (i.e., by excluding the distance effect of non-lead syndicate members) biases against our finding a distance effect. We identify the lead investor following Lee and Wahal [2004] and choose the VC firm with the largest investment in the syndicate as the lead investor. We obtain the zip codes for the lead VC firm headquarters and any branch offices and the zip codes for the venture-backed IPO from SDC. We compute the zip code to zip code distance between the IPO company and each of the VC firm s offices and select the minimum as the shortest distance in the analysis. We use the proprietary in-house program of a company that is in the mail sorting business to calculate the distances between the zip codes. To test the reliability of the distances, we randomly selected a sample of observations and calculated the distances between zip codes through and obtain very similar results. We classify IPOs which are within 25, 50, and 100 miles to VC firms with dummy variables. We prefer to use close VC dummy variables to capture the effect of proximity instead of a continuous measure of distance (such as miles) because we do not expect a linear relation between proximity and its effect on monitoring and/or certification. Figure 1 shows that a disproportionate percentage of venture-backed companies are within 25 miles of their lead VC firm. This finding is consistent with the local bias result that Dai and Cumming [2006] document. Along these lines, some observers argue that the effective geographical radius within which VC firms prefer to make investments may be restricted to one to two hours travel time from their office (Mason and Harrison [1992]) or less (Dai and Cumming [2006]). <Insert Figure 1 here> 8

11 Our main dependent variable is the initial return on the IPO, which we compute as the percentage change from the SDC offer price to the first day closing price from CRSP. Many factors besides VC firm distance affect initial returns, so we construct the following control variables: price revision, measured as [offer price-midpoint of original filing range] / midpoint of original filing range*100; the proportional IPO filing range, computed as [original high filing price original low filing price] / midpoint of original filing price*100. All of the variables mentioned above are from SDC. We use the log of total IPO proceeds (from SDC) and log of total assets (from COMPUSTAT, data item #6) as measures of the size of the IPO and size of the company, respectively, and we convert both of these measures to 2004 dollars. We calculate asset tangibility as net plant, property, and equipment (COMPUSTAT, data item #7) / total assets (COMPUSTAT, data item #6) *100. As a proxy for uncertainty about the company, we use the standard deviation of stock returns in the after-market. Because underwriters might provide price support at first, we exclude the first several days, and calculate the standard deviation of returns (from CRSP) for 10 through 180 calendar days post IPO. We control for market wide movements from the filing date to the offer date using the percent change in the CRSP equal weighted composite index during the filing period. We control for the IPO volume by number of IPOs in a given year from Professor Jay Ritter s website. The age of the company at IPO is in years and we calculate this variable by subtracting the year the company was founded from the IPO year. The top underwriter dummy takes a value of one for lead managers with a tombstone ranking greater than 8 (Carter and Manaster [1990], Carter, Dark, and Singh [1998], Loughran and Ritter [2002]). We also use an internet company dummy that takes a value of one if the company is listed as an internet IPO and a technology company dummy; both are defined in the website of Professor Jay Ritter at University of Florida. 9

12 We calculate the age of the VC firm at IPO by subtracting the founding year of the VC firm (from SDC) from the IPO year. Following Gompers [1996], we create a young firm dummy that takes a value of one for VC firms that are less than 6 years old and a value of zero otherwise. We use the total investments ($ million) made by VC firm prior to the first investment year in the IPO company as another proxy for VC firm experience. (We note that we do not use the total amount of investments that VC firm participated variable that SDC supplies. The SDC variable represents investments of VC firms to the date the data are downloaded so it is not representative of the VC firm experience when the company receives its funding.) We calculate the total prior VC firm investments variable for each year for each VC firm by summing the total investments of the VC firm in companies in all prior years within the SDC data. We then use the year the VC firms makes its first investment in the portfolio company as the reference year to assign the experience of the VC firm for each observation. We create an industrial cluster dummy variable. An industrial cluster is a geographic area that has a significant number of firms that belong to a particular industry. This variable is motivated by Almazan, et al. [2006] who show the effect of geographical industrial clusters on financial decisions of firms. In our context, the idea is that a company located in an area where industry specific expertise is present is more likely to work with a close VC firm that is specialized in that industry who can advise/monitor more effectively and can make a greater use of close distance. For example, internet companies have an industry cluster in Silicon Valley, near San Francisco, California and telecommunications companies have an industry cluster in the Telecom Corridor, near Dallas, Texas. We denote a company as being in an industrial cluster if there are at least three IPOs with the same 2-digit SIC code within a 25 mile radius in our sample. Finally, we control for the time and industry varying characteristics of the IPO by using dummy variables representing the year of the IPO and industry dummy variables 10

13 represented by two digit SIC codes, respectively. (We note that this does not create a perfect collinearity with the internet and technology company dummies because these classifications are defined with finer granularity, i.e., at the 3- and 4-digit SIC code level.) We also control for the state the IPO company is in because of the well-known concentrations of VC firms in certain parts of the country (we discuss this in more detail below). Thus, all of our multivariate tests reflect within-state effects. 3. Results This section introduces the descriptive statistics and regression results for our sample. Sub-section A gives information on VC industry characteristics. Sub-section B gives descriptive statistics of the variables used in our analysis. Sub-section C introduces the characteristics of VC firms and venture-backed companies that are close to each other. Sub-section D examines the effects of grandstanding. Sub-sections E, F and G introduce the simple OLS regressions and average treatment effect regressions. Subsection H examines the effect of close proximity on initial returns in sub-samples and Sub-section I discusses robustness checks. A. Characteristics of the venture-backed IPOs: Geographic and industrial distribution Figure 2 presents the geographic distribution of the venture-backed companies in our sample on a U.S. map. Figure 2, Panel A presents companies that are not within 25 mile proximity to their VC firms whereas Figure 2, Panel B presents companies that are within 25 mile proximity. The two figures show that companies that are within 25 mile proximity to their VC firms do not present a specific region of the U.S., such as California. They exist rather uniformly in regions where venture-backed IPO activity is generally high. <Insert Figure 2 here> 11

14 Table 1, Panel A describes the composition of our sample of venture-backed IPO companies by state. Consistent with conventional wisdom, California has by far the largest number of venture-backed IPOs, with 324 of the 915 sample companies. Massachusetts, Texas, and New York also have large numbers of venture-backed IPOs. Venture-backed IPO companies are most likely to be within 25 miles of their VC firm in New York (41% of companies), California (38% of companies) and New Jersey (37% of companies). <Insert Table 1 here> Table I, Panel B reports the distribution of venture-backed IPO companies by industry. The most represented industries are business services (SIC code 73) with 34% of our sample of venture-backed IPOs and electronic equipment (SIC code 36) with 12% of our sample. These two industries, along with industrial machinery (SIC code 35), are also the ones that are most likely to be within 25 miles of their VC firm (33% of electronic equipment companies, 32% of business services companies and 28% of industrial machinery companies). <Insert Figure 3 here> Figure 3, reports the distribution of venture-backed IPOs by year. The bubble period of contains a disproportionate number of our sample companies (255 of 915, or 28%). Because this was an unusual period for initial return levels and IPO numbers, we will discuss below the sensitivity of our results to the inclusion or exclusion of this period. B. Characteristics of venture-backed IPOs: Summary statistics Table 2, Panel A presents descriptive statistics of some of the variables in our sample. Note that the mean and median initial returns are quite high in the overall sample, 34.9% and 14.7% respectively. This result is driven by both the fact that our 12

15 sample consists of venture-backed IPOs (which are known to have relatively high initial returns compared to the universe of IPOs) and by the huge initial returns during the bubble period of 1999 and As we show in Table 2, Panel B, excluding the bubble period, the average initial return for IPOs is around 18%. During the bubble period, however, the initial return is 79% on average. These numbers are consistent with the results in Lee and Wahal [2004] and reflect the non-stationary in the underpricing of IPOs (Loughran and Ritter [2004]). <Insert Table 2 here> C. Characteristics of venture-backed IPOs: Close versus distant venture-backing Simple univariate comparisons indicate that initial returns for venture-backed IPO companies close to their VC firms are higher than those companies that are distant from their VC firms. However, as we will explain later in detail, this result is due to omitted variable bias and endogeneity in the choice of proximity. <Insert Table 3 here> The choice of whether to match with a close or a distant VC firm is endogenous. In Table 3 we present some company and VC firm characteristics that may be related to this choice. Table 3, Panel A has univariate tests for differences of characteristics for close versus distant matches. Companies that match with close VC firms are smaller and younger, have lower asset tangibility, and are more likely to be in industrial clusters than companies that match with distant VC firms. The average age of a portfolio company within 25 miles of its VC firm is 8 years at the IPO stage, whereas those that are more distant have an average age of 13 years. This difference and, except as indicated, the others we mention in the remainder of this sub-section are statistically significant at the 1% level. Similarly, companies within 25 mile proximity of their VC firm have average total assets of $129 million whereas those that are not close have average total assets of 13

16 $249 million. The asset tangibility ratio is 17% for companies within 25 mile proximity, compared to 27% for those companies that are not. Companies that are located in an industrial cluster location are more likely to be close to their VC firms: 82% of the IPO companies in industrial clusters are close to their VC firms whereas 56% of off-cluster IPO companies are close to their VC firms. We also examine whether relatively inexperienced VC firms tend to invest in portfolio companies within close proximity, which would be consistent with inexperienced VC firms having limited resources and a comparative advantage at soft information production. We use the dollar amount of all previous investments in which a VC firm participated as a proxy for VC firm experience. The companies that are within 25 mile proximity have less experienced VC firms on average. D. Is there grandstanding? Grandstanding occurs when inexperienced VC firms, in an effort to establish a reputation for bringing portfolio companies public, push their portfolio companies public too early. The idea is that the VC firm will recoup any opportunity costs by being able to raise larger subsequent investment funds sooner than they otherwise would. More established VC firms have less need to try to build reputation in this way. In this section, we document activities consistent with the grandstanding hypothesis of Gompers [1996]. Table 3, Panel B compares descriptive statistics for initial return and price revision for the IPOs of young VC firms versus experienced VC firms. We follow Gompers [1996] and define a young/inexperienced firm as one that is less than 6 years old. Table 3, Panel B shows that IPO companies backed by experienced VC firms have average initial returns of 33% whereas those backed by young firms have average initial return of 58% and the difference in means is significant. This result is consistent with 14

17 Gompers [1996] and shows that, on average, companies backed by young VC firms leave more money on the table at the IPO. Another implication of the grandstanding hypothesis is that young VC firms take younger companies to IPO compared to experienced VC firms. In Table 4, we report regressions where age of the company at the IPO (in years) is the dependent variable, following Gompers [1996]. In the first regression we include both the 25 mile proximity dummy together with the young VC dummy. The 25 mile proximity dummy is significant at 5% level and the young VC dummy has a p-value of 10.7%. Interestingly, the magnitudes of the coefficients of these two dummies are quite similar, the 25 mile proximity dummy has a coefficient of and the young VC firm dummy has a coefficient of <Insert Table 4 here> Overall, our data are consistent with grandstanding activities by relatively inexperienced VC firms. Moreover, close VC firms are more apt to have the characteristics of grandstanders than distant VC firms. Furthermore, it is the IPO companies that are close to their VC firms that are youngest and smallest. Because these characteristics may also affect IPO initial returns, we need to control for the endogenous choice of proximity in our analysis. E. Modeling the choice of proximity We further analyze the matching between the VC firm and the venture-backed company in a multivariate setting. We run a probit regression using the 25 miles proximity dummy as our dependent variable to analyze the characteristics of both VC firms and companies that choose to be close to one another. (We note that other proximity cutoffs give similar results.) Table 5 gives the results for this regression. Consistent with the univariate tests, the results here suggest that venture-backed 15

18 companies within close proximity to their VC firms are smaller, younger, more opaque, and are more likely to be in an industrial cluster. In addition, close companies match with VC firms that are less experienced, as evidenced by the dollar amount of previous investments, which is negative and statistically significant. <Insert Table 5 here> This result is interesting because the characteristics of the VC firms and the venture-backed companies that are in close proximity to each other coincide with those most likely to be involved in grandstanding. Previous evidence on grandstanding suggests that young and inexperienced VC firms take companies earlier to IPO in order to establish reputation and successfully raise capital for new funds (Gompers [1996], Lee and Wahal [2004]). Moreover, Lee and Wahal [2004] show that the companies backed by relatively young/inexperienced VC firms are younger, smaller, and more underpriced at their IPO than those of established VC firms. F. Matching based on soft information production The evidence in the previous section suggests smaller and younger venturebacked companies and inexperienced VC firms tend to match within close proximity. This evidence is consistent with matching based on the start-up company s soft information production needs. However, the matching between the VC firm and the start-up company might simply be a case of lower quality companies not having access to high quality intermediaries. If so, such an outcome would be consistent with the findings of Fernando et al. [2005]. They document that IPO issuers and IPO underwriters associate by mutual choice: lower quality companies tend to match with underwriters with lower reputation. It is difficult to distinguish between these two (not mutually exclusive) explanations that less established companies pair with less established VC firms (a) 16

19 because they benefit from doing so due to those VC firms comparative advantage at soft information production, or (b) because they have no better opportunity available to them. To distinguish between the two, we examine the determinants of company size at the IPO, a proxy for company quality and transparency. The idea here is that under case (b) above (the no better opportunity available story), the reasons for matching between company and VC firm comes primarily from similarities in quality and the ability of a VC firm to produce soft information is secondary at most. In contrast, under case (a) above (the comparative advantage story), soft information production is of primary importance. To test this argument, we model the size of the venture-backed company with independent variables that we use in Table 5, and include an interaction term between geographic proximity and VC firm experience. The intuition for our approach is as follows. Imagine a small venture-backed company located in, say, California. Under the comparative advantage story, this company is unlikely to match with an inexperienced VC firm that is distant, such as one that has its nearest office in, say, New York. Under the alternative story, geographic proximity is a secondary concern, and matches between companies and VC firms arise on the basis of their having comparable positions on their respective quality spectra. We regress start-up company size on the interaction of proximity and VC firm experience, variables for the direct effects of proximity and VC firm experience, and control variables. If the comparative advantage story holds in the data, we expect the interaction term to be statistically significant. That is, the effect of geographic proximity (our proxy for soft information production) on start-up company size at the IPO should be stronger for inexperienced VC firms because these are the ones with an advantage at soft information production. 17

20 We present the results in Table 6. The dependent variable is the natural logarithm of total assets of the venture-backed company. The first column in Table 6 shows that distance between the VC firm and the venture-backed company increases with company size. Also, VC firm experience, measured by the total dollar amount of investments previously completed, is positively related to company size. Both of these results are consistent with our previous findings in Table 5 and suggest that, on average, smaller venture-backed companies are closer to their VC firms and are more likely to work with inexperienced VC firms. As discussed above, our primary interest is in the interaction between distance and VC firm experience. If smaller companies choose to match with inexperienced VC firms within close distance to enhance soft information production, then the interaction variable should be significantly and negatively related with company size. We define a VC firm with less than $100 million in previous investments as inexperienced. (We note that our results are not sensitive to this cutoff: using $200 million or $300 million as our cutoff gives similar results.) Consistent with the comparative advantage story, the size of the venture-backed company is significantly and negatively related with this interaction variable. When we include some additional control variables (second column of Table 6), we reach similar conclusion. <Insert Table 6 here> This result confirms our argument that smaller companies and inexperienced VC firms choose to match with one another within close proximity and this finding is not driven by only a good matches with good, bad matches with bad equilibrium. If soft information production played no part, then the interaction term in this analysis would be insignificant. The fact that the interaction is significant is consistent with our hypothesis that firm and intermediary choose to match with one another based on soft information production abilities and needs. 18

21 The economic magnitude of the effect of the matching within close distance is difficult to interpret with the analyses in Table 6 because the interaction term is a dummy variable whereas the components of the interaction term are continuous variables in the model. To ease interpretation, we substitute the continuous distance measure with a dummy variable that represents being close within a certain radius (25 miles, 50 miles or 100 miles). The magnitude of the interaction is economically large: in unreported regressions, the results suggest that the effect of being close and working with an inexperienced VC firm has a 52% to 127% (depending on the cutoff for proximity) greater effect on the size of the venture-backed company compared to the effect of just being close to the VC firm. G. Endogeneity: Average treatment effect model Our results in the previous section show that the choice of proximity between venture-backed company and VC firm is endogenous. Indeed, it is precisely those companies, which are likely to have the greatest underpricing at the IPO stage that will need the most soft information production. This soft information production can be done most easily by a close and small VC firm. However, because those small VC firms are also likely to be young and/or inexperienced, they are more likely to have incentives to grandstand and this creates a confounding effect. Under these circumstances, OLS estimates are inconsistent because the proximity dummy is correlated with the error term (Greene [2003]). We deal with the endogeneity of proximity with a treatment effects model (Heckman [1976], Heckman [1978], Manning [2004]). Treatment effects models are common in labor and health economics research and are gaining popularity in finance applications (Ciamarra [2006], Li and Prabhala [2005]). In our model, the treatment is having a proximate VC firm. If we have Y 1 and Y 0 as the outcome (IPO initial return) 19

22 with and without treatment, respectively, then we are interested in the effect of proximity on Y. Thus, the effect of treatment is ATE = E(Y 1 - Y 0 ). If we know a set of variables X that affect the decision to receive treatment, then our average treatment effect will be ATE = E(Y 1 - Y 0 X). This specification gives us an unbiased estimate only if the decision to receive treatment is randomized across firms that receive treatment and those that do not. However, in our model, the decision to receive the treatment (i.e., choice of proximity) is not random, which creates a selection bias. For example, as we have seen from the previous analysis, companies that choose to receive the treatment are younger and smaller. To overcome the bias that we mention in the above discussion, we use the endogenous dummy variable approach developed by Heckman [1976, 1978]. We model the probability of receiving the treatment, equation (2) below, with the structural outcome, equation (1) below: Y i = α + β 1 X i + β 2 D i + ε i (1) D i * = δz i + u i (2) D i = 1 if D i * > 0 D i = 0 if D i * < 0 Equation (2) reflects the decision to receive treatment, where Z i is a set of characteristics that affect the choice of receiving treatment. D i in the first equation is an endogenous dummy variable, indicating whether the treatment is received and it is determined with D i * from the second equation which is an unobserved latent variable. The variables in X can overlap with variables in Z, but it is assumed that at least one component of Z is unique. The individual error terms ε i and u i are assumed to have bivariate normal distribution. We use full information maximum likelihood to solve the model (Heckman [1976], Heckman [1978]). 20

23 H. IPO Initial returns: The effect of proximity To apply the average treatment effects model, we first estimate a model to predict the choice to receive the treatment: being close to the VC firm. This step is directly analogous to the probit model estimated in Table 5, though here the treatment equation is estimated jointly with the structural equation of initial returns. The dependent variable is a proximity dummy, with explanatory variables identical to those in Table 5 (log of total assets, total previous investments of the VC firm, company age at the IPO date, industrial cluster dummy, and asset tangibility ratio). In our treatment analysis, we use as an instrument a dummy variable for whether the start-up company is geographically located in an industrial cluster. Companies and venture firms tend to coexist within industry clusters, such as Silicon Valley, so companies are more likely to be close to their VC firms when they are located within industry clusters. On the other hand, we do not see a direct connection between IPO underpricing and being in an industry cluster, other than its indirect effect through proximity. The treatment equation results are very similar to those reported in Table 5, so we do not report them in a separate table. (The treatment equation results are available from the authors upon request.) Table 7 gives the results for the initial return regressions for both OLS and the average treatment effects model. In these regressions the dependent variable is the IPO initial return in percentage terms. We examine how proximity to the VC firm effects underpricing of the venture-backed IPO by regressing initial return, defined as percentage change from the offer price to the first day closing price, on a proximity dummy and several control variables. We control for the filing range, after-market variability of the stock, market return during the filing period (CRSP-Equal weighted index), a dummy for internet companies, IPO volume, price revision, offer size, age of company, young VC firm dummy and a dummy for top underwriter. We also include dummy variables for 21

24 each state, two digit SIC code, and year of the IPO company to capture industry, state, and time-specific effects. It is important to control for state effects in the regression because this control ensures us that the results are not driven by venture-heavy states such as California. <Insert Table 7 here> The first regression model is a baseline OLS regression that does not control for the potential endogeneity problem. In this regression, the coefficient of the 25 mile proximity dummy is not significantly different from zero. Consistent with other studies, the coefficients on the control variables show that price revision, the internet company dummy, standard deviation of return and the young VC firm dummy, are all positively and significantly related to initial return whereas offer size, IPO filing range, top underwriter dummy and market return are negatively and significantly related. All of the coefficients of the variables have the expected sign and are consistent with prior studies on IPO underpricing. The R-squareds for these regressions are 60%. The second regression in Table 7 gives the results for the average treatment effects model that explicitly corrects for the endogenous choice of proximity. This regression is jointly estimated with the treatment equation described above. We reject the hypothesis of independent equations (exogeneity) at the 5% level. Thus, the treatment effects model is a more correct specification than OLS. All control variables have the predicted signs and do not differ much from the OLS regression results. Our main variable of interest is the coefficient on the proximity dummy in the average treatment effects model and the young VC firm dummy. The regression results show that company proximity has a negative and significant effect on the initial return of a venture-backed company at the IPO stage. Overall, we see that once we account for the characteristics of the companies that choose to be close to their VC firms and correct for the selection bias in the model, we are able to show that IPOs that 22

25 are close to their VC firms are significantly less underpriced. (We note that other specifications produce similar results.) The average treatment effect on the treated is -12.3% for the 25 mile proximity specification. This means that, absent the offsetting endogeneity issues, IPOs backed by close VC firms have underpricing of 12.3 percentage points less, on average, than those backed by distant VC firms. Compared to the whole sample average initial returns of 44.7% for those companies within 25 mile proximity, this treatment effect is about one fourth of the overall initial return. The coefficient on the young VC firm dummy in the treatment regression is about 9.8% and is significant, similar in magnitude to the coefficient in Gompers [1996]. Overall, our results show that being close to the VC firm can decrease the underpricing at the IPO stage in a magnitude that can offset the negative effects of an inexperienced VC firm that is likely to grandstand. This result is important because it suggests that proximity can partially explain why companies choose to work with young VC firms despite the risk of grandstanding. We then estimate the robustness of our results to alternative definitions of close. We change the definition from within 25 miles to within 50 miles and then within 100 miles. Models (3) and (4) are average treatment effects models just like model (2), except using different distance cutoffs for what is denoted as close. The results are similar, with each producing a statistically significant distance effect of about the same order of magnitude: 15 percentage points for a 50 mile cutoff, 16 percentage points for a broader 100 mile cutoff. We do not report the OLS regressions for these robustness tests because no matter what distance cutoff we use, proximity is always insignificant due to the offsetting endogeneity problem. To assess whether this distance effect is unique to the internet bubble period, we re-run the tests omitting the IPOs that occurred during The coefficient for the 25 mile proximity dummy remains significant at the 10% level. The magnitude of the 23

26 effect is much smaller, down to 7 percentage points from 12. Of course, the non-bubble period initial returns are also much smaller (21% mean outside the bubble period for IPOs with close VC firms), and 7 percentage points corresponds to approximately one third less underpricing compared to the overall sample mean. I. Sub-sample tests: Who benefits the most from proximity? One of our main arguments in this paper is that young venture-backed companies are those that will benefit most from soft information production and thus, are likely to see the largest benefit from being close to their VC firms. To test this hypothesis, we repeat our regression tests from Table 7 with two sub-samples of the data: one subsample comprises only those companies that are 6 years or younger at the IPO, and the other sub-sample comprises the complement. (We note that, due to our treatment effects model approach, simply including an interaction term would produce inconsistent estimates; see Wooldridge [2002, p. 236].) The results are reported in Table 8. We do not report the treatment regression results since these results are similar to those in Table 5. <Insert Table 8 here> The first column in Table 8 represents the regression for the sub-sample of young companies. The coefficient on the 25 mile proximity is -28.4% and statistically significant (p = 6.2%). Compared to the coefficient of -12.2% reported in the whole sample, we see that the gain from being close to the VC firm is more than doubled for companies that are younger. Looking at column 2 in Table 8, we see that the regression for the sub-sample comprising older companies has a coefficient of -6.2% (not statistically significant) for the 25 mile proximity dummy. A visual comparison across these two sub-sample regressions indicates that the effect of VC firm proximity on underpricing is substantially greater for the young portfolio companies, which are the companies that most need soft information production. 24

27 Overall, these results are consistent with the arguments in this paper. Younger companies benefit from being close to their VC firms. For that purpose, they tend to match within close proximity with VC firms that are relatively less experienced who have greater incentives to produce information. Overall we see that, especially for younger companies, benefits from close proximity outweigh the cost of grandstanding. J. Robustness Our results are robust to different specifications of close proximity. Instead of using a close dummy variable, we also try actual distance in miles (logged and nonlogged) as the independent variable to measure the effect of proximity on underpricing. Under a linear specification, the difference between 1 and 101 miles distance is treated the same as a difference between 2400 and 2500 miles distance. Not surprisingly, the distance variable is not statistically significant when we replace our proximity dummy with distance in miles. This result is not surprising because even a casual visual inspection of the distribution of distance in Figure 1 suggests that the effect of distance is likely to be non-linear. Companies that are located where VC firms develop industry-specific expertise are also more likely to be closer to their VC firms. We control for this factor with the cluster location dummy in our treatment equation. We also control for 1-digit SIC industry code in our treatment equation and find consistent results. (We note that the maximum likelihood estimation procedure does not converge if we use 2-digit SIC codes.) Finally, including IPO year dummies in the treatment equation does not change our results. 4. Conclusion 25

28 Why do companies choose to work with inexperienced financial intermediaries that have high potential for agency conflicts? We argue that the answer partially lies in the need for information production. Financial intermediaries produce information about the companies that they work with and information is valuable, especially for smaller companies that are more opaque and rely on soft information. Due to their organizational form [Stein (2002)], inexperienced financial intermediaries have greater advantage at producing soft information. As a result, smaller companies and less experienced financial intermediaries match together to achieve better information production. In equilibrium, there is a tradeoff between the benefits of better information production and costs due to agency conflicts between the company and the financial intermediary. We look into the venture capital industry where benefits from soft information production are likely to be high and where high agency costs due to grandstanding exist. Our results provide a demand side argument for grandstanding (Gompers [1996]) by relatively inexperienced VC firms. Grandstanding is costly for the portfolio companies because their IPOs are underpriced substantially more than their counterparts that are brought to market by more seasoned VC firms. Why would portfolio companies be drawn to venture capitalists that are likely to grandstand? We argue that companies most in need of soft information production choose to pair with VC firms that have characteristics most conducive to producing such soft information relatively small and inexperienced VC firms that are physically close to the company. But inexperienced VC firms also have the strongest incentives to grandstand. Thus, there is a tradeoff the companies most in need of soft information production will naturally pair with VC firms most likely to engage in costly grandstanding. We directly examine the nature of the matching between VC firms and portfolio companies and our conclusions strongly support the argument above. We find that when soft information production is most in demand (i.e., companies that are young, small, and 26

29 have low asset tangibility) VC firms and portfolio companies match in close proximity. These geographically close matches, though, are with relatively inexperienced VC firms. This is not surprising, as these firms may be best suited for soft information production (e.g., they may have organizational forms that are less hierarchical than well-established VC firms (Stein [2002]), but it also bears out our tradeoff argument. We also note that the matching between the VC firm and the start-up company is not simply a case of lower quality companies not having access to high quality intermediaries, like what Fernando et al. [2005] find. We document evidence that matching between the inexperienced VC firm and the small company is taking place within close distance, supporting our argument that the matching is rather driven by soft information needs. We document an effect of this tradeoff between information benefits and agency costs by examining the effect of proximity between venture-backed IPO companies and their lead VC firm. In OLS results, we find no difference in the initial returns of the venture-backed companies which are close to their VC firms compared to those that are distant. However, we show that this (non-)result is driven by the fact that portfolio companies and VC firms choose to be close to one another based on soft information production needs. After controlling for the endogeneity in the proximity choice, we show that venture-backed IPOs that are within a short commute distance to their VC firms have substantially lower underpricing. One interpretation of the result is that the lower underpricing of the IPO is a result of the enhanced monitoring and/or certification role of the close VC firm. Our results also suggest that the benefits from being close are higher for smaller companies, which strengthen our argument that matching within close distance is driven by soft information needs. These findings combine to provide a demand side rationale for grandstanding. Gompers [1996] documents strong incentives for inexperienced VC firms to grandstand, 27

30 but leaves as an open question why start-up companies would have a demand for inexperienced VC firms that are likely to grandstand. Gompers [1996] states, While this paper does not address the reasons entrepreneurs seek financing from young venture capital firms who then rush them to the IPO market, the issue deserves greater attention[ ]. We fill this gap with our matching analysis, thereby supplying the demand side explanation for grandstanding. The organizational characteristics that encourage grandstanding by VC firms are the same characteristics that encourage soft information production. 28

31 References Almazan, Andres, Adolfo De Motta, Sheridan Titman, and Vahap Uysal, 2006, Financial structure, liquidity, and firm locations, Working paper, University of Texas at Austin. Berger, Allen N., Nathan H. Miller, Mitchell A. Petersen, Raghuram G. Rajan, and Jeremy C. Stein, 2004, Does function follow organizational form? Evidence from the lending practices of large and small banks, Journal of Financial Economics 76, Brickley, James A., James S. Linck, and Clifford W. Smith, 2003, Boundaries of the firm: Evidence from the banking industry, Journal of Financial Economics 70, Butler, W. Alexander, 2008, Distance still matters: Evidence from municipal bond underwriting, Review of Financial Studies, Vol. 21, Carter, Richard, Frederick H. Dark and, Ajai K. Singh, 1998, Underwriter reputation, initial returns, and the long-run performance of IPO stocks, Journal of Finance 53, Carter, Richard and, Steven Manaster, 1990, Initial public offerings and underwriter reputation, Journal of Finance 45, Ciamarra, Elif, 2006, Monitoring by affiliated bankers on board of directors: Evidence from corporate financing outcomes, Working paper, New York University. Coval, Joshua D., and Tobias J. Moskowitz, 2001, The geography of investment: Informed trading and asset prices, Journal of Political Economy 109, Dai, Na, and Douglas J. Cumming, 2006, Where you are matters! Local bias in venture capital investment, Working paper, University of New Mexico. Degryse, Hans, and Steven Ongena, 2004, Distance, lending relationships, and competition, Journal of Finance 60, Fernando, Chitru S., Vladimir A. Gatchev, Paul A. Spindt, 2005, Wanna dance? How firms and underwriters choose each other, Journal of Finance 60, Gompers, Paul, and Josh Lerner, 2001, The venture capital revolution, Journal of Economic Perspectives 15, Gompers, Paul A., 1995, Optimal investment, monitoring, and the staging of venture capital, Journal of Finance 6, Gompers, Paul A., 1996, Grandstanding in the venture capital industry, Journal of Financial Economics 42, Greene, William H., Econometric Analysis, (Prentice Hall, Upper Saddle River, N.J.). 29

32 Gupta, Anil K., and Harry J. Sapienza, 1992, Determinants of venture capital firms preferences regarding the industry diversity and geographic scope of their investments, Journal of Business Venturing 7, Heckman, James J., 1976, The common structure of statistical models of truncation, sample selection and limited dependent variables and a simple estimator for such models, Annals of Economic and Social Measurement 5, Heckman, James J., 1978, Dummy endogenous variables in a simultaneous equation system, Econometrica 46, Hsu, David H., 2004, What do entrepreneurs pay for venture capital affiliation?, Journal of Finance 59, Jeng, Leslie A., and Philippe C. Wells, 1997, The determinant of venture capital funding, (Harvard Business School, Cambridge, MA). Lee, Peggy M., and Sunil Wahal, 2004, Grandstanding, certification and the underpricing of venture capital backed IPOs, Journal of Financial Economics 73, Lerner, Joshua, 1995, Venture capitalists and the oversight of private firms, Journal of Finance 50, Li, Kai, and N. R. Prabhala, 2005, Self-selection models in corporate finance, Working paper, University of Maryland. Loughran, Tim, and Jay Ritter, 2002, Why don't issuers get upset about leaving money on the table in IPOs?, Review of Financial Studies, Loughran, Tim, and Jay Ritter, 2004, Why has IPO underpricing changed over time?, Financial Management 33, Malloy, Christopher, 2005, The geography of equity analysis, Journal of Finance 60, Manning, Allan, 2004, Instrumental variables for binary treatments with heterogeneous treatment effects: A simple exposition, The B.E. Journal of Economic Analysis & Policy 3, issue 1, article 9. Martin, Ron, Christian Berndt, Britta Klagge, and Peter Sunley, 2005, Spatial proximity effects and regional equity gaps in the venture capital market: Evidence from Germany and the United Kingdom, Environment and Planning 37, Mason, Colin, and Richard Harrison, 1992, The supply of equity finance in the U.K.: A strategy for closing the equity gap, Entrepreneurship and Regional Development, 4, Megginson, William L., and Kathleen H. Weiss, 1991, Venture capitalist certification in initial public offerings, Journal of Finance 46,

33 Petersen, Mitchell A., and Raghuram G. Rajan, 2002, Does distance still matter? The information revolution in small business lending, Journal of Finance 57, Ritter, Jay, 1984, The hot issue market of 1980, Journal of Business 57, Stein, Jeremy, 2002, Information production and capital allocation: Decentralized versus hierarchical firms, Journal of Finance 57, Tian, Xuan, 2007, Geography, staging and venture capital financing, Working paper, Boston College. White, Halbert, 1980, A heteroskedasticity-consistent covariance estimator and a direct test for heteroskedasticity, Econometrica 48, Wooldridge, Jeffrey M., 2002, Econometric Analysis of Cross Section and Panel Data, Cambridge, MA: MIT Press. Wright, Mike, and Lockett Andy, 2003, The structure and management of alliances: Syndication in the venture capital industry, Journal of Management Studies 40,

34 Figure 1 Distribution of venture capital firm and the venture-backed company Panel A shows the distribution of the distance between the venture capital firm and the venture-backed company in our whole sample; each bin represents 25 miles. Frequency is the number of companies within each bin. Panel B shows the distribution of the distance between the VC firm and the venture-backed company in our sample for observations where distance between the venture capital firm and venture-backed company is less than 500 miles. Each bin in panel B represents 5 miles. Panel A Frequency Distance between VC firm and company (miles) Panel B Frequency Distance between VC firm and company (miles) 32

35 Figure 2 Geographic distribution of venture capital-backed companies This figure shows the geographic distribution of venture capital-backed companies on the U.S. map (Hawaii and Alaska are omitted). The midpoint of each circle represents the zip code that the venture capital-backed company is located in and the circle around it has a radius of 25 miles. Panel A gives the distribution of the venture capitalbacked companies whose lead venture capital firm is distant (i.e., not within the 25 mile radius). Panel B gives the distribution of companies that have their lead venture capital firm close (i.e., within the 25 mile radius). Panel A Panel B 33