• No results found

Are technology and internet firms more prone to IPO underpricing?

N/A
N/A
Protected

Academic year: 2021

Share "Are technology and internet firms more prone to IPO underpricing?"

Copied!
54
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

Are technology and internet firms more prone to IPO

underpricing?

MSc thesis

Business Economics - Finance track

University of Amsterdam

Student name: Tom de Ruiter Student number: 10671102 Supervisor: Jens Martin

December 2014

(2)

1

Abstract

This paper investigates whether firms active in the technology and internet sector are more subject to IPO underpricing than other firms. The research focuses on U.S. firms that went public during the period 1993 – 2014. Where most previous research approximated ex ante risk by industry rather than by firm-specific factors, this paper attempts to analyze the effect of ex ante risk on underpricing by

estimating ex ante risk using the Parkinson Extreme Value method. In addition, an alternative proxy based on prospectus complexity is analyzed as an additional robustness check. The paper also investigates the role of the media in explaining IPO underpricing. Ex ante risk and media coverage are both found to significantly explain IPO underpricing, but the effect of ex ante risk may be overstated due to underwriter price support. Only little evidence is found to support the hypothesis that tech firms are more prone to excessive underpricing, especially after the internet bubble. However, high-tech internet firms are found to be excessively underpriced during the entire period.

Keywords: Initial public offerings, underpricing, technology, ex ante risk, media coverage

This document is written by Tom de Ruiter who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

(3)

2

Table of Contents

Abstract ... 1 Table of Contents ... 2 1 Introduction ... 3 2 Literature review ... 5 2.1 IPO underpricing ... 5

2.2 Classical explanations of underpricing ... 6

2.3 Modern explanations of underpricing ... 9

2.4 Firm-specific determinants of underpricing ... 10

2.5 Relevance ... 13

3 Methodology & hypotheses ... 13

3.1 Objectives ... 13

3.2 Model ... 17

4 Data and descriptive statistics ... 22

4.1 Data sources ... 23 4.2 Data description ... 24 5 Results ... 26 6 Robustness checks ... 29 7 Conclusion ... 32 References ... 35 Appendix ... 38

(4)

3

1 Introduction

In late July 2014, chair of the Federal Reserve Janet Yellen warned investors for high stock-valuations in the biotech and internet sector. Her testimony follows a period of growing unease in which an increasing number of market analysts and fund managers expressed their concerns regarding a possible second tech bubble. The Nasdaq, known as the technology-heavy exchange in the US, fell 5% after the announcement, clearly demonstrating the growing anxiety surrounding internet and technology stocks. The Nasdaq has experienced its worst start since 2001, as already 30 firms have withdrawn or postponed plans to list. Nevertheless, young technology firms still seem to raise astronomical amounts of capital on the equity markets. On September 19, Alibaba’s record-breaking Initial Public Offering (IPO) raising approximately $22 billion still closed almost 38% above its offer price.

Although new issues closing above their offering price is not a surprising

phenomenon per se, issuing firms generally dislike leaving “money on the table” as a consequence of IPO underpricing. An underpriced IPO implies that the firm and its pre-IPO shareholders are getting less money for their shares than they actually deserved. Nevertheless, IPOs are underpriced on average (Lowry, Schwert, 2010). In these cases the efficient market hypothesis fails and the initial IPO offer price, which should equal the expected market price, is set too low (Ritter, Welch, 2002). As a result, the stock experiences substantial first-day returns.

Despite the prevalence of underpricing the urge to go public remains, and is driven by overall expected market returns (Pástor, Veronesi, 2005). Recent success stories of relatively young high-tech (internet) firms such as LinkedIn and Twitter further fuel the urge to go public, as IPO’s tend to cluster in periods of high market sentiment (Pástor, Veronesi, 2005). Both firms experienced extraordinary initial returns on their first trading day. Obviously, investors welcome this phenomenon. For them, the (short-term) return on investment is very high. However, for the firms raising capital it could become very costly when the amount left on the table is high, especially when it is unanticipated.

(5)

4 This paper attempts to analyze empirically whether tech-, internet and high-tech

internet firms are more likely to be subject to (excessive) IPO underpricing than firms in other industries. The research question this paper attempts to answer therefore is as follows: Are firms in the technology and internet sector subject to a greater degree of IPO underpricing?

In addition to the research question, this paper attempts to identify the specific factors contributing to this phenomenon, and investigates the role of the news media in determining IPO underpricing.

A possible explanation that firms the technology industry experience more severe underpricing is that it is simply harder to determine the true value of these firms. This implies that excess underpricing would occur as a consequence of higher uncertainty and risk (Loughran, Ritter, 2004). When the valuation of a company is more difficult due to higher complexity, the risk on the firms’ share price when it goes public is also higher (Lowry et al, 2010). An alternative explanation is that IPO underwriters deliberately underprice the IPO (Florin, Simsek, 2007). Examples of incentives to underprice deliberately are to attract more investors (Rock, K. 1986) or to make a long-run profit by satisfying investors, who are repeat customers for the underwriting bank and are expected to maintain a long-run relationship as long as there are mutual benefits (Hill, Wilson, 2006). It could be that any differences in underpricing simply arise due to qualitative differences between underwriters. This paper suggests that another possible explanation is that high-tech issues are “hyped” more by the news media. If a specific issue receives a lot of media coverage, it seems plausible that this increases the participation rate of investors the moment the issuing firms’ share goes public.

By analyzing several determinants of underpricing, this paper attempts to determine whether any possible differences in underpricing between tech, internet and other industries can be explained by valuation uncertainty, underwriter quality and the coverage in the media. By establishing whether and why drivers of underpricing are different for firms in high-tech industries would shed some light on the determinants of IPO underpricing and thus be a useful contribution to the current literature.

(6)

5 This paper analyzes a sample of 2,865 unique firm observations in the US, with new

issues ranging from 1993 to 2014. By looking at the initial returns on the 1st trading day of the offering, this paper investigates the severity and causes of underpricing for firms active in the technology, internet and high-tech internet sector. In addition, this paper offers two proxies for ex ante risk to capture the valuation uncertainty surrounding new issues.

The remainder of this thesis is organized as follows. Section 2 contains the

literature review and elaborates on previous papers in the field of IPO underpricing. Section 3 describes the methodology of the research, and lists the proposed

hypotheses and the expected coefficient signs. Section 4 describes the data sample, as well as the database construction process. Section 5 will present the main findings, and section 6 will expand on these through robustness checks. Finally, section 7 concludes.

2 Literature review

2.1 IPO underpricing

An initial public offering is a private firms’ first attempt to raise capital through the public equity market. Under the assumption of perfect capital markets it is hypothesized that when a firm goes public, its initial offering price should equal its expected market price. This assumption, however, has proven to be very unrealistic in practice. Ibbotson (1975) and Logue (1973) were among the first economists to recognize that when firms go public their respective share prices tend to jump exceptionally on the first trading day. This means that the IPO was priced below its market value. Consequently, firms tend to leave significant amounts of potential IPO proceeds on the table.

Roughly four decades have passed since the insights of Ibbotson and Logue, and their work has served as the foundation for many academics that have since aimed to explain this phenomenon. Today, we still see that IPOs tend to be underpriced on average (Lowry et al., 2010). Although the occurrence of underpricing has found extensive empirical

(7)

6 support in the current literature, a vast amount of different theories have emerged aiming to explain exactly what factors contribute to this market imperfection. This section explores the most fundamental theories regarding IPO underpricing in the current literature.

2.2 Classical explanations of underpricing

IPO underpricing has been a very popular field of research over the past several years. Even though many new insights have emerged as a result, some of the most prominent theories today have been around for quite a while.

The first explanation deals with information asymmetry, and states that (excess) underpricing arises through the underwriting party (Rock, K., 1986). The proposition is loosely based on the famous ‘lemon’s theory’ (Akerlof, 1970). When an

underwriter starts the book building process to discover the price investors are willing to pay for a share of the firm going public, it relies on informed investors’ bids. When enough informed investors exist, the underwriter should be able to efficiently find the expected market value of the firm’s shares. However, when underwriters are unable to attract enough informed investors to purchase the

shares, they will also have to attract uninformed investors to fill their order book. As uninformed investors lack costly information about the issuing firm, the underwriter will have to compensate them for this lack of information by lowering the offer price. The underwriter thus knowingly underprices the firms’ shares. This implies that when the underwriter is able to attract enough informed investors, underpricing is expected to be less severe. It is argued that more prominent underwriters, who typically have access to larger networks of informed investors, are better able to attract enough informed investors (Carter, Dark, Singh, 1998) (Wang, Yung, 2011). A more prominent underwriter is thus argued to engage in a certification role by staking its reputation and by doing so is able to find enough informed investors. Carter et al. investigate this effect using a measure for underwriter reputation to benchmark the activity of informed investors. They find evidence that more prestigious underwriters are associated with IPOs that are subject to less underpricing. They do also document, however, that this negative relationship changes to a positive one in the 1990s, but do not give an explanation for this.

(8)

7 Loughran and Ritter (2004) confirm this sign change and document a positive

relationship between underpricing and underwriter reputation over the period 1990 - 2003. They argue that the certification theory as proposed by Carter, Dark and Singh (1998) is unable to explain underpricing during this period, and propose the “changing issuer objective function” hypothesis. In this hypothesis, Loughran and Ritter suggest that underwriters relaxed their underwriting criteria since the early 1990s and took a growing number of young and unprofitable firms public. Instead of aiming to maximize IPO proceeds as the underwriters did before the 1990s, they now engage in more rent-seeking behavior by aiming to become the lead

underwriter – who typically receives larger fees – by publishing bullish analyst coverage reports. Issuing firms are here assumed to choose their lead underwriter primarily on the basis of expected analyst coverage. As the issuing firms do not pay the underwriter a direct fee for the costly analyst coverage, they end up paying for this cost indirectly as underpricing.

A second concept fundamental to the underpricing literature builds upon the principal-agent problem. Baron and Holmström (1980) were among the first to argue that underwriters might have other incentives besides optimizing the IPO proceeds for the issuer. As the underwriter incurs a substantial part of the costs for finding the optimal price to maximize the IPO proceeds, it may be more cost-effective for the underwriter to set the price sub-optimal so to limit its own costs and make a greater profit. More specifically, the underwriter’s rent-seeking behavior is subject to a trade-off where it wants to set the price high enough so to satisfy the issuer, but also low enough so that finding potential investors is cost-efficient. As the issuing firm typically has less information about the true price than the underwriter, determining whether the offer price is set correctly will prove difficult. Due to this moral hazard problem between the underwriter and the issuing firm, Baron and Holmström (1980) argue that the underwriter has an incentive to exert sub-optimal effort. When the true value of the issuing firm is subject to a relatively large amount of

uncertainty, the extent of informational asymmetry between the underwriter and the issuer becomes larger. As a consequence, it becomes harder for the issuing firm to monitor the effort of the underwriter so that the degree of underpricing increases. Therefore, when it is more difficult to identify the true value of a company, underpricing is expected to be higher as a consequence of increased information asymmetry.

(9)

8 More recent explanations of underpricing that build on incentive misalignment

between the issuer and the underwriter offer slightly different explanations. Like before, the underwriter is argued to have incentives to engage in rent-seeking behavior for his own benefit. Reasons for the underwriter to engage in this rent-seeking behavior may be to maintain its reputation towards investors or to

incentivize aggressive bidding (Sherman, 2004). As the underwriter typically has a lasting relationship with investors as these are repeat customers when it comes to investing in public offerings. The underwriter thus has an incentive to keep investors satisfied to maintain business over the long run (Hill, Wilson, 2006). More

prestigious underwriters, who typically have a larger network of investors, have a greater incentive to satisfy their investors so that they remain repeat customers. They are thus argued to preserve relationships with investors by compensating them through underpricing. Complementing this theory is the research by Schenone (2004), which shows that when a firm going public has a prior relationship with the underwriter the amount of underpricing is generally lower.

Some researchers argue that underpricing does not arise through deliberate actions, but rather is the result of underwriter price support (Ruud, 1993).

Underwriter price support occurs when an underwriting firm attempts to stabilize the stock during the IPO by purchasing whenever the price falls below a certain

benchmark, typically the initial offer price. By analyzing the distribution of initial IPO returns, Ruud argues that the positive mean initial returns are the result of the partially unobserved left negative tail of the distribution. This unobserved tail is argued to be the result of underwriter price support, and biases the mean initial returns upward. Prabhala and Puri (1998), however, suggest that even though it may be true that underwriter price support affects the measured degree of

underpricing, it is by itself not able to fully explain the underpricing phenomenon. By looking at the mean and variance of initial IPO returns, they found that IPOs remain significantly underpriced after accounting for price support.

(10)

9 2.3 Modern explanations of underpricing

The previous section focused on the fundamental literature underlying IPO underpricing. However, especially in the last decade, many new theories have emerged offering

alternative, often complementing explanations for why new issues tend to be underpriced.

An important stream of literature that found considerable footing recently deals with uncertainty surrounding the valuation of a company. When a company has uncertain prospects and is thus more difficult to value, underpricing is an efficient response to the complexity of the valuation (Lowry et al, 2010). Firms such as for example Twitter Inc. are shrouded in uncertainty, as it is unclear how their business model is going to generate cash flows in the future. A company’s uncertain prospects are directly related to the ex ante risk of the company, as uncertainty related to the valuation of a firm translates into price uncertainty (Knopf, Teall, 1999). Loughran and Ritter (2004) also suggest that riskier IPO’s tend to experience more severe underpricing than less risky ones.

Another explanation for excess underpricing deals with a firm’s toleration of underpricing. Firms may anticipate some extent of underpricing to occur and choose to tolerate it, as they intend to return to the capital market to raise additional capital through a seasoned equity offering (SEO) at a later date (Su, Fleisher, 1999). By allowing excess underpricing to occur, the issuing firm wishes to make investors eager to participate in a future equity offer. This could allow a firm to raise more money at more desirable terms when they return to the equity market at a later date. This means that the higher a firms appetite for additional capital, the more it tolerates excess underpricing.

In an IPO, primary shares are the newly issued shares that are sold to investors. Any cash that is generated under this process is transferred to the company. Secondary shares, in contrast, are the pre-existing shares that already belong to shareholders that are sold to investors. Habib and Ljunqvist (2001) argue that issuing firms care about underpricing to the extent for which they participate in the offering. When an IPO consists solely of primary shares, the costs of promotion are borne by the issuing firm. When an IPO would consist solely of secondary shares, all promotion costs would be incurred by the selling

shareholders. Habib and Ljungqvist reason that the issuing firm attempts to minimize wealth losses, which consist of promotion costs and money left on the table as a

(11)

10 consequence of underpricing. As an increase in promotion costs leads to a decrease in underpricing, there is a tradeoff for the issuing firm. By analyzing a sample of US IPOs over the period 1991 to 1995, the authors find that underpricing decreases when the issuing firms stands to lose more from it. In other words, when the issuing firm has more skin in the game, it exerts more effort in reducing underpricing.

Barber and Odean (2003) showed that individual investors have a tendency to buy stocks that are covered in the media. Building upon this reasoning, one would expect that media coverage prior to an IPO would have an impact on the first-day returns. Liu et al. (2007) investigate the effects of qualitative media coverage during the filing period of an IPO and find that increased coverage relates to a greater degree of underpricing after an upward price-revision during the filing period. Bhattacharya et al. (2004) explored the role of the media on internet stock during the internet bubble period. By looking at news items and classifying them as good, neutral or bad, they found that media coverage was more

intense for internet IPOs during the bubble period. In addition, they looked at the effects on daily adjusted returns based on these daily news items over the entire bubble period. However, Bhattacharya et al. only found a weak relation between good-quality news coverage and adjusted returns. They did not, however, look at underpricing specifically. It would be interesting to investigate the role the media has in predicting underpricing. In addition, it would be interesting to see how quantitative rather than qualitative news coverage would impact underpricing both before, during and after the bubble.

2.4 Firm-specific determinants of underpricing

A growing number of academics look for alternate explanations of underpricing, arguing that underpricing is not only the result of underlying forces related to information

asymmetries and incentive misalignment. Inherent to this reasoning is the fact that many companies are fundamentally different from each other, and are therefore to some degree expected to behave differently under specific circumstances. One of the best examples in the context of IPO underpricing is the internet bubble during 1999 – 2000. During this period, the equity values of many young internet-based companies rose to astronomical heights due to rapidly increasing stock prices, optimistic investor sentiment and stock-market speculation. Ljungqvist and Wilhelm (2003) showed that the excessive first-day

(12)

11 returns associated with internet IPOs can be (at least partially) contributed to certain firm-characteristics such as pre-IPO ownership structure and a firm’s transparency and riskiness prior to the IPO (ex ante risk). More specifically, when ownership is more dispersed underpricing was found to be more severe. In addition, firms with greater valuation uncertainty and thus higher ex ante risk also significantly explained the higher degree of underpricing.

However, Ljungqvist and Wilhelm measured the degree of ex ante risk by looking at the effects of underpricing on internet- and tech-firm dummies, arguing that ex ante risk is higher for these firms. I believe that this measure of risk is subject to some degree of self-selection, as internet- and tech-firms are the specific firms of which the initial returns skyrocketed during the bubble period. It is also subject to endogeneity issues due to

omitted variables that contribute to ex-ante risk. The effects of ex ante risk on underpricing could therefore be investigated more specifically by using a different measure for ex ante risk.

Another theory on underpricing that builds upon firm-specific dynamics deals with types of capital backing. Lee and Wahal (2004) documented that IPOs with venture capital backing experienced larger underpricing relative to their non-venture backed counterparts. Under their “grandstanding” hypothesis, they argue that especially younger venture capitalists aiming to strengthen their reputation are more willing to take smaller and possibly riskier firms public. Gompers and Lerner (1997) argue that any differences are attributable to conflicts of interest between the issuer and the VC backing firm, which they claim

diminishes the more prominent the backing VC firm is. Their reasoning is comparable to Carter, Dark and Singh’s underwriter reputation hypothesis. Coakley et al. (2009) offer another different explanation. They find that underpricing is larger for venture capital

backed firms during the bubble but lower afterwards. They reason that the certification role of venture capitalists allows issuing firms to overcome agency problems, as the backing firms “stake their reputation” on the issue, resulting in lower underpricing.

Levis (2011) documented that private equity-backed IPOs performed better than in the long run relative to their non-private equity-backed counterparts over the period 1992 - 2005. However, he also recognized that the amount of underpricing for PE-backed IPOs is

(13)

12 significantly lower compared to non-private equity-backed IPOs. One explanation for this lower underpricing is that IPOs backed by PE investors have lower adverse selection problems (Ferretti, Meles, 2011). In these cases, it is argued that private equity investors are signaling the quality of the issuing firm to the market. This is partly contributable to their continuous role in the IPO market: successful PE investors build up a reputation which enhances their signaling strength. Although the certification theory as argued by Coakley et al. (2009), Levis (2011) and Ferretti and Meles (2011) offers a different explanation than the grandstanding theory as proposed by Lee and Wahal (2004) and Gompers and Lerner (1997), they agree upon the fact that VC and PE-backing appears to play a significant role in explaining underpricing.

It is, however, very important to realize that some researchers seem to use different definitions regarding venture capital backing and PE backing. It may be possible that their vast differences in results inherent to the debate are contributable to their definitions of a venture capitalist. Some argue that venture capitalists and private equity backers are identical, whereas others acknowledge the subtle differences (primarily dealing with leveraged buyouts). The official definition of a venture capitalists states that VC financing is part of the broader private equity activities that for instance also include leveraged

buyouts (Cumming, Johan, 2013). Under this definition, the difference between VC and PE backing primarily deals with the size and age of the backed firms: venture capitalists tend to focus on smaller, younger firms than PE investors (Feretti, Meles, 2011). However, this paper argues that in terms of operating strategy PE and VC investors are more like

opposites: PE investors identify the firms’ fundamentals and aim to maximize profitability through operational efficiency, whereas VC investors build upon the existing employees’ ideas and try to figure out how profitable the firm can become through added knowledge and expertise. However, this paper recognizes that both VC and PE investors fulfill similar roles in optimizing and signaling profitability as reasoned under the certification theory, and are thus expected to behave similarly in their roles as IPO backers.

(14)

13 2.5 Relevance

As discussed before, several academics such as Ljunqvist and Wilhelm (2003) and Loughran and Ritter (2004) have noted that underpricing is more severe for tech and internet firms as these are subject to higher valuation uncertainty (ex ante risk). However, due to the lack of a clear measure of ex ante risk, this was tested through industry

dummies to account for differences in risk. This paper argues that this method is subject to endogeneity problems, as not every firm within an industry is expected to be subject to comparable levels of ex ante risk. In addition, it is unreasonable to believe that every tech firm for instance is riskier than every non-tech firm. This paper aims to investigate this gap in the literature by identifying ex ante risk each firm individually, rather than aggregating it on the industry level. It then becomes possible to identify whether tech and internet firms remain underpriced after accounting for risk.

This paper also aims to contribute to the debate in the literature surrounding the role of the media and its relation to underpricing, as well as to establish whether the likelihood of a seasoned equity offering could explain underpricing. Finally, by analyzing these issues this paper aims to answer whether tech firms are more prone to IPO underpricing.

3 Methodology & hypotheses

3.1 Objectives

The previous section elaborated on the academic background inherent to this research. This section elaborates on the methodology used to answer the research question.

This paper aims to investigate whether tech- and internet firms are more prone to IPO underpricing. In addition, it attempts to identify the determinants for IPO underpricing, and whether these determinants are able to explain any differences between tech and non-tech firms. The first step in answering the research question is to confirm that firms active in the technology and internet sector are indeed subject to more underpricing relative to other firms as expected.

Hypothesis 1: Firms active in the internet and technology sectors are subject to more underpricing.

(15)

14 The next step is identifying the variables that play a role in explaining underpricing, and whether these variables behave differently for firms active in the technology and internet sector. As outlined in the previous section, Carter, Dark and Singh (1998) argued that more prestigious underwriters are better able to find enough informed investors and therefore are better able to limit the extent of underpricing through. However, this theory was only found to hold in the years prior to 1990. Under the changing issuer objective function hypothesis, this paper expects to confirm a positive relationship between

underpricing and underwriter reputation for the period 1990 – 2003 similar to Loughran and Ritter (2004), and expects this to hold until 2013. More prestigious underwriters are thus expected to be associated with a larger degree of underpricing for the period 1990 - 2013. It is therefore expected that the sample will find that firms with larger first day returns are generally associated with more prestigious underwriters for the period after 1990. This brings us to hypothesis 2.

Hypothesis 2: Underpricing is positively related to underwriter reputation.

Rejecting hypothesis 2 would reject the changing issuer objective function hypothesis in favor of the underwriter certification theory.

Like many academics have argued before, the uncertainty around the valuation of a

company due to higher ex ante risk is expected to increase the degree of underpricing of a new issue. This paper therefore expects to find larger first day returns for firms with a higher ex ante risk. Underpricing is therefore expected to be positively related to the ex ante risk of a firm.

Hypothesis 3: Underpricing is positively related to the ex ante risk of a firm.

Many academics have approximated ex ante risk by simply looking at the industry of a firm and hypothesizing that this accurately depicts any differences between ex ante risk.

Although a rough measure, this makes some intuitive sense as tech firms are expected to be subject to more ex ante risk than other firms (Engelen, van Essen, 2010). On average, technology firms tend to have faster growth opportunities, but are also much harder to valuate: technology firms tend to have relatively large amounts of intangible assets. It is

(16)

15 therefore expected that the results of this paper will show that technology firms indeed have higher ex ante risk relative to the other firms, partially explaining the possible differences in underpricing between technology and non-technology firms’ IPOs. When an issuing firm knows it needs to revisit the capital markets in the future, it is hypothesized that it is willing to accept some expected degree of underpricing to please investors, who are repeat customers. By doing so it stimulates the willingness of these investors to participate in an SEO so that it may raise additional equity in the future against possibly better terms. This paper attempts to approximate the likelihood of a seasoned equity offering by looking at the future capital appetite of a firm. When a firm burns up more of its shareholder capital today, its need for future capital and therefore refinancing increases. This future capital appetite is approximated by looking at the burn rate of a firm: the rate at which the firm depletes its cash balance. This burn rate functions as a proxy for the issuing firms’ likelihood of revisiting the capital market through an SEO to raise

additional financing.

Hypothesis 4: Underpricing is negatively related to a firm’s future capital appetite.

As firms active in the tech and internet sectors are expected to be subject to more uncertain cash flows, they burn up more of their shareholder capital than their non-technology peers. These firms are thus expected to be subject to larger burn rates, and expected to have a larger future capital appetite. Technology firms are thus expected to allow more underpricing to occur, so to stimulate investors to reinvest in the future. The next step in the process is to determine whether media coverage does indeed affect the degree of underpricing. However, precisely measuring the media-hype surrounding a firm’s IPO is virtually impossible. As with many aspects of economics, to accurately measure how a group of random individuals interprets and responds to information relies on the critical assumption that every individual behaves rationally. Even though this assumption is rather naïve in practice, it is usually justified by interpreting results as an average response to the unit of information. There is, however, a second problem. An individuals’ interpretation of a specific unit of information can be vastly different to that of its peers’, as interpretation not only relies on rationality but also on prior informational asymmetries between individuals. This hurdle is typically crossed by simply assuming all individuals to have identical prior knowledge, again knowing this assumption is rather

(17)

16 naïve in practice. These two facets of behavioral economics, rational behavior and

information asymmetry respectively, both arise when trying to measure the media-hype surrounding an IPO. Like before, an investor’s interpretation and reaction to a specific unit of information regarding an IPO can be vastly different than that of one of its peers’. It is for this reason that an investors’ reaction to the media and other forms of information rely on the assumption that investors are all rational and comparable to begin with.

By looking at quantity rather than quality of information surrounding a firms’ IPO, it

becomes possible to negate the side effects of irrational behavior and different individuals’ informational advantages prior to the news. After all, a specific investors’ return given his interpretation of information is not of interest. Instead, we need to know whether

widespread access to any information in the first place impacts all investors’ incentives to participate in an IPO event. For this reason, instead of identifying news as either positive or negative, the absolute amount of coverage in the media is used to measure the extent of a potential media hype. After all, as many marketing professionals have reasoned

before: there’s no such thing as bad publicity. It is therefore expected that the initial returns are larger the greater the extent of media coverage.

Hypothesis 5: Underpricing is positively related to the absolute amount of media coverage.

Before a venture capitalist or private equity firm invests in a firm, it typically engages in comprehensive due diligence. Naturally, the VC or PE investor only wants to invest if the investment exhibits optimistic prospects. Once the investor is engaged with a firm it provides further knowledge and resources to maximize the profitability of the firm, and takes part in extensive monitoring. These roles that VC and PE investors partake in signal to other investors that the backed issue has a positive outlook: it reduces the adverse selection problem, making it easier for the underwriter to find enough informed investors. For this reason, this paper expects to find that VC or PE backed issues are less prone to underpricing.

(18)

17 Rejecting hypothesis 6 would mean rejecting the signaling theory in favor of Lee and

Wahal’s (2004) the grandstanding theory, which states that venture capitalists aiming to strengthen their reputation are willing to take relatively more risky firms public.

3.2 Model

This section discusses the underlying econometric model for the research.

The research focuses on US-listed firms during the period 1993 – 2014. This period is interesting as it covers 4 periods in which IPO markets were subject to different market conditions. In the first period, 1993 – 1997, covering the years prior to the internet bubble, markets behaved relatively normal. During the second period, 1998 – 2000, the number of new issues increased dramatically, as well as their returns. When the bubble burst, the IPO market turned relatively pessimistic and was characterized by a low amount of issues and especially tech firms refrained from going public. The aftermath of the bubble lasted until approximately 2004, after which the IPO market started to show clear signs of optimism again. These periods will therefore be analyzed separately.

The underlying econometric model to analyze the research question is defined as follows.

(1). 𝐼𝑛𝑖𝑡𝑖𝑎𝑙  𝑅𝑒𝑡𝑢𝑟𝑛𝑠 =   𝛽!+ 𝛽!∗ 𝑇𝑒𝑐ℎ  𝑑𝑢𝑚𝑚𝑦 + 𝛽!∗ 𝐵𝑢𝑟𝑛  𝑟𝑎𝑡𝑒 + 𝛽!∗ 𝐸𝑥  𝑎𝑛𝑡𝑒  𝑟𝑖𝑠𝑘 + 𝛽!∗  𝑈𝑊  𝑟𝑎𝑛𝑘 + 𝛽!∗ 𝐹𝑖𝑟𝑚  𝑎𝑔𝑒  𝑎𝑡  𝐼𝑃𝑂   +  𝛽!∗ 𝐹𝑖𝑟𝑚  𝑠𝑖𝑧𝑒 + 𝛽!∗ 𝐵𝑎𝑐𝑘𝑖𝑛𝑔 + 𝛽!  ∗  𝑀𝑒𝑑𝑖𝑎  𝑐𝑜𝑣𝑒𝑟𝑎𝑔𝑒 + 𝛽!∗ 𝑃𝑎𝑟𝑡𝑖𝑐𝑖𝑝𝑎𝑡𝑖𝑜𝑛  𝑟𝑎𝑡𝑒 +  𝜀 (2). 𝐼𝑛𝑖𝑡𝑖𝑎𝑙  𝑅𝑒𝑡𝑢𝑟𝑛𝑠 =   𝛽!+ 𝛽!∗ 𝑇𝑒𝑐ℎ − 𝐼𝑛𝑡𝑒𝑟𝑛𝑒𝑡  𝑖𝑛𝑡𝑒𝑟𝑎𝑐𝑡𝑖𝑜𝑛  𝑡𝑒𝑟𝑚 + 𝛽!∗ 𝐵𝑢𝑟𝑛  𝑟𝑎𝑡𝑒 + 𝛽!∗ 𝐸𝑥  𝑎𝑛𝑡𝑒  𝑟𝑖𝑠𝑘 + 𝛽!∗  𝑈𝑊  𝑟𝑎𝑛𝑘 + 𝛽!∗ 𝐹𝑖𝑟𝑚  𝑎𝑔𝑒  𝑎𝑡  𝐼𝑃𝑂   +  𝛽!∗ 𝐹𝑖𝑟𝑚  𝑠𝑖𝑧𝑒 + 𝛽!∗ 𝐵𝑎𝑐𝑘𝑖𝑛𝑔 + 𝛽!  ∗  𝑀𝑒𝑑𝑖𝑎  𝑐𝑜𝑣𝑒𝑟𝑎𝑔𝑒 + 𝛽!∗ 𝑃𝑎𝑟𝑡𝑖𝑐𝑖𝑝𝑎𝑡𝑖𝑜𝑛  𝑟𝑎𝑡𝑒 +  𝜀

(19)

18 Where:

Initial Returns = 1st day returns (%)

Tech dummy = technology firm dummy variable

Tech-internet

interaction term = High-tech internet firm interaction term

Burn rate = rate at which the firm depletes its shareholder capital

Ex ante risk = uncertainty prior to the IPO

UW rank = high underwriter reputation dummy variable

Firm age at IPO = age of the firm at issue date

Firm size = size of the firm (in terms of assets)

Backing = venture capital or private equity backed issue dummy

variable

Media coverage = degree of media coverage in the week before the IPO

Participation rate = number of primary shares offered relative to total shares

Two separate multiple regressions are used to analyze the research question. The first regression includes a technology dummy to identify differences in underpricing between firms active in the technology sector and those who are not. In the second regression, this technology dummy is substituted for the internet dummy, which shows firms active in the high tech internet industry.

The variable Initial Returns, which measures the amount of underpricing, is defined as the percentage change in returns on the first trading day of the IPO, and is constructed as follows.

𝐼𝑛𝑖𝑡𝑖𝑎𝑙  𝑅𝑒𝑡𝑢𝑟𝑛 = (𝑓𝑖𝑟𝑠𝑡  𝑑𝑎𝑦  𝑐𝑙𝑜𝑠𝑖𝑛𝑔  𝑝𝑟𝑖𝑐𝑒 − 𝑜𝑓𝑓𝑒𝑟  𝑝𝑟𝑖𝑐𝑒) 𝑜𝑓𝑓𝑒𝑟  𝑝𝑟𝑖𝑐𝑒

In the above equation, the first day closing price is the official closing price at the end of the day on the first trading day. The offer price is the issue price at which the shares went public.

(20)

19 To approximate the likelihood of a seasoned equity offering through a firms’ appetite for future capital, the burn rate is calculated. This variable is constructed as follows.

𝐵𝑢𝑟𝑛  𝑟𝑎𝑡𝑒 =(𝑛𝑒𝑡  𝑐𝑎𝑠ℎ  𝑓𝑙𝑜𝑤𝑠  𝑓𝑟𝑜𝑚  𝑂𝐴   +  𝑛𝑒𝑡  𝑐𝑎𝑠ℎ  𝑓𝑙𝑜𝑤𝑠  𝑓𝑟𝑜𝑚  𝐼𝐴) 𝐼𝑃𝑂  𝑝𝑟𝑜𝑐𝑒𝑒𝑑𝑠

𝐼𝑃𝑂  𝑝𝑟𝑜𝑐𝑒𝑒𝑑𝑠 = 𝑜𝑓𝑓𝑒𝑟  𝑝𝑟𝑖𝑐𝑒 ∗ 𝑛𝑢𝑚𝑏𝑒𝑟  𝑜𝑓  𝑠ℎ𝑎𝑟𝑒𝑠  𝑜𝑓𝑓𝑒𝑟𝑒𝑑 Where:

Net Cash Flows from OA = net cash flows from operating activities Net Cash Flows from IA = net cash flows from investing activities

It is important to note that net cash flows from investing activities are almost always negative. A greater negative burn rate implies that the firm depletes its IPO proceeds faster. It is therefore expected that β2 has a negative sign.

Under normal circumstances the preferred method to estimate a firms’ exposure to risk is by looking at its stock beta, as projected by the Capital Asset Pricing model (CAPM). When a firm goes public, however, it is impossible to use the standard methods to estimate a firms’ ex ante risk. Naturally, there is no access to the historic stock-data necessary for CAPM estimation as it does not exist yet. Therefore, stock betas cannot be used as a proxy for ex ante risk.

A common alternative in the existing literature is to use the standard deviations of the first 20 trading days, but this has been shown to be a poor measure of ex ante risk (Johnson, Miller, 1988). Other academics, such as Ljungqvist and Wilhelm (2003) and Roosenboom and Schramade (2006) have used high-tech industry dummies as an indicator of risk, but this method is to a large degree subject to endogeneity problems. It is not very reasonable to assume that all tech firms are subject to higher risk on average. Therefore, an

alternative measure for valuation uncertainty is needed. This research estimates ex ante risk by using the Parkinson Extreme Value method, as first proposed by Parkinson (1980). When the natural logarithm of a stock price appears to follow a normally distributed

(21)

20 standard deviation (Knopf, Teall, 1999). Parkinson Extreme Values are estimated as the natural logarithm of the highest price of the first trading day divided by the lowest price of the first trading day; ln(H/L).

𝐸𝑥  𝑎𝑛𝑡𝑒  𝑟𝑖𝑠𝑘 = ln 𝑓𝑖𝑟𝑠𝑡  𝑑𝑎𝑦  ℎ𝑖𝑔ℎ𝑒𝑠𝑡  𝑡𝑟𝑎𝑑𝑖𝑛𝑔  𝑝𝑟𝑖𝑐𝑒 𝑓𝑖𝑟𝑠𝑡  𝑑𝑎𝑦  𝑙𝑜𝑤𝑒𝑠𝑡  𝑡𝑟𝑎𝑑𝑖𝑛𝑔  𝑝𝑟𝑖𝑐𝑒

As ex ante risk is expected to be positively related to underpricing, β3 is expected to have a positive sign.

To analyze the reputation and quality of the underwriters, the 2012-updated ranking of underwriting parties’ reputation along the lines of the framework as initially brought forth by Carter, Dark and Singh (1998) and further enhanced by Loughran and Ritter (2004) is used. This ranking gives each underwriter a rank on a scale ranging from 0 to 9.0 for specific time periods. More specifically, a rank of 0 – 2.0 typically implies that the

underwriter is primarily associated with penny stocks, whereas a ranking of 8.0 or higher identifies a very prominent underwriter. This ranking contains underwriter reputation rankings for different periods and thus allows relative comparability between underwriters in terms of reputation and quality over time.

To distinguish between high- and low-quality underwriters, a rank of 8.0 is set as the cutoff point, which is consistent with the proposed cutoff point under Loughran and Ritter (2004). Underwriters below this cutoff point are classified as low-quality. The difference between high- and low-quality underwriters is then captured by a dummy variable based upon this benchmark.

𝐻𝑖𝑔ℎ  𝑞𝑢𝑎𝑙𝑖𝑡𝑦  𝑢𝑛𝑑𝑒𝑟𝑤𝑟𝑖𝑡𝑒𝑟  𝑑𝑢𝑚𝑚𝑦 =   (0  |  𝑢𝑛𝑑𝑒𝑟𝑤𝑟𝑖𝑡𝑒𝑟  𝑟𝑎𝑛𝑘 < 8.0)1    𝑢𝑛𝑑𝑒𝑟𝑤𝑟𝑖𝑡𝑒𝑟  𝑟𝑎𝑛𝑘 ≥ 8.0)

As high quality underwriters are expected to be positively related to underpricing, it is expected that β4 has a positive sign.

An important control variable is the age of a firm at the time of issuing. Older firms

(22)

21 is available to the public. Both these factors are expected to influence investor’s decisions in valuing an IPO, and are therefore expected to play a significant role in the underpricing process. The variable is constructed as follows.

ln  (𝐹𝑖𝑟𝑚  𝑎𝑔𝑒  𝑎𝑡  𝐼𝑃𝑂) =  ln  (𝑖𝑠𝑠𝑢𝑒  𝑑𝑎𝑡𝑒 − 𝑓𝑜𝑢𝑛𝑑𝑖𝑛𝑔  𝑑𝑎𝑡𝑒 + 1)

As regression analysis assumes variables to be normally distributed, the natural logarithm is taken to adjust firm age at IPO for normality. 1 year is added so that the natural

logarithm of firms that go public in their founding year does not become ln(0), as ex = 0 is a mathematically undefined expression.

As it is expected that younger firm are harder to valuate and typically have a less solid outlook than more established firms, it is expected that the degree of underpricing is

greater for younger issuing firms. It is for this reason that β5 is expected to have a negative sign.

In the previous section this paper hypothesized that VC and PE investors partake in a signaling role to investors, reducing underpricing. For this reason, underpricing is expected to be negatively related to VC or PE backing, and β7 is expected to have a negative sign.

The variable media coverage signals the amount of publicity the firm receives in the 2 weeks prior to the issue date. The amount of coverage is measured as an absolute number of news articles that were published in the 2 week interval. As it is expected that more coverage means more underpricing, β8 is expected to have a positive sign.

Participation rate is the final variable in the regression model. As the degree to which the issuing firm is expected to exert effort in minimizing the wealth losses of underpricing is dependent on the degree to which the issuing firm is exposed to the offering, it can be approximated by the variable Participation rate as defined below.

𝑃𝑎𝑟𝑡𝑖𝑐𝑖𝑝𝑎𝑡𝑖𝑜𝑛  𝑟𝑎𝑡𝑒 =𝑃𝑟𝑖𝑚𝑎𝑟𝑦  𝑠ℎ𝑎𝑟𝑒𝑠  𝑜𝑓𝑓𝑒𝑟𝑒𝑑 𝑇𝑜𝑡𝑎𝑙  𝑠ℎ𝑎𝑟𝑒𝑠  𝑜𝑓𝑓𝑒𝑟𝑒𝑑

A higher participation rate implies that the issuing firm has relatively more exposure to underpricing, and will thus exert more effort in minimizing it. As a higher participation rate

(23)

22 thus implies lower underpricing, the sign of β9 is therefore expected to be negative, as suggested by Habib and Ljungqvist (2001).

To further analyze the performance of initial public offerings beyond simply looking at initial returns, this paper briefly investigates the effects of the before-mentioned variables on the long-run performance of the offering. By using 60-day returns as the dependent variable in a similar multiple regression model, long run effects can be compared to the degree of underpricing as found in the original regression model. The regression model looks as follows. (3). 𝑃𝑜𝑠𝑡  𝑟𝑒𝑡𝑢𝑟𝑛𝑠 =   𝛽!+ 𝛽!∗ 𝑇𝑒𝑐ℎ  𝑑𝑢𝑚𝑚𝑦 + 𝛽!∗ 𝐵𝑢𝑟𝑛  𝑟𝑎𝑡𝑒 + 𝛽!∗ 𝐸𝑥  𝑎𝑛𝑡𝑒  𝑟𝑖𝑠𝑘 + 𝛽! ∗  𝑈𝑊  𝑟𝑎𝑛𝑘 + 𝛽!∗ 𝐹𝑖𝑟𝑚  𝑎𝑔𝑒  𝑎𝑡  𝐼𝑃𝑂   +  𝛽!∗ 𝐹𝑖𝑟𝑚  𝑠𝑖𝑧𝑒 + 𝛽!∗ 𝐵𝑎𝑐𝑘𝑖𝑛𝑔 + 𝛽!  ∗  𝑀𝑒𝑑𝑖𝑎  𝑐𝑜𝑣𝑒𝑟𝑎𝑔𝑒 + 𝛽!∗ 𝑃𝑎𝑟𝑡𝑖𝑐𝑖𝑝𝑎𝑡𝑖𝑜𝑛  𝑟𝑎𝑡𝑒 +  𝜀 (4). 𝑃𝑜𝑠𝑡  𝑟𝑒𝑡𝑢𝑟𝑛𝑠 =   𝛽!+ 𝛽!∗ 𝑇𝑒𝑐ℎ − 𝐼𝑛𝑡𝑒𝑟𝑛𝑒𝑡  𝑖𝑛𝑡𝑒𝑟𝑎𝑐𝑡𝑖𝑜𝑛  𝑡𝑒𝑟𝑚 + 𝛽!∗ 𝐵𝑢𝑟𝑛  𝑟𝑎𝑡𝑒 + 𝛽! ∗ 𝐸𝑥  𝑎𝑛𝑡𝑒  𝑟𝑖𝑠𝑘 + 𝛽!∗  𝑈𝑊  𝑟𝑎𝑛𝑘 + 𝛽!∗ 𝐹𝑖𝑟𝑚  𝑎𝑔𝑒  𝑎𝑡  𝐼𝑃𝑂   +  𝛽!∗ 𝐹𝑖𝑟𝑚  𝑠𝑖𝑧𝑒 + 𝛽!∗ 𝐵𝑎𝑐𝑘𝑖𝑛𝑔 + 𝛽!  ∗  𝑀𝑒𝑑𝑖𝑎  𝑐𝑜𝑣𝑒𝑟𝑎𝑔𝑒 + 𝛽!∗ 𝑃𝑎𝑟𝑡𝑖𝑐𝑖𝑝𝑎𝑡𝑖𝑜𝑛  𝑟𝑎𝑡𝑒 +  𝜀

It is hypothesized that the 2-month performance of the new issues is comparable between tech firms, high-tech internet firms and other firms. It is therefore expected that regressions (3) and (4) won’t show signs of over- or underperformance between industries.

4 Data and descriptive statistics

The previous section provided the methodological framework that will be used to answer the research question. This section focuses on the data collection and variable

(24)

23 4.1 Data sources

Various data sources are needed to gather all the necessary data to gather and construct all the variables as outlined in the previous section. The majority of the necessary data items can be obtained through ThomsonOne (previously SDC Platinum), a financial research tool provided by Thomson Reuters. ThomsonOne is the preferred financial database when it comes to M&A and equity events, and allows the collection of specific data on specific IPO events over a specified time period. From this platform the necessary offer dates, offer prices, number of primary and secondary shares offered, IPO indicators, and venture capital- and private equity-backed indicators are gathered for the period 1993 – 2014. As ThomsonOne only contains data on the specific IPO events, aftermarket data is obtained from the CRSP database, accessible through the Wharton Research Data Services (WRDS) platform. From CRSP daily closing prices, daily high- and low trading prices, and other stock information are obtained for the period 1993 - 2014. By matching the daily closing prices from CRSP to the IPO event-dates from ThomsonOne initial returns can be constructed. The 60-day returns are constructed using a similar method.

Firm-specific indicators are obtained through the Compustat database, which are also accessible through WRDS. From Compustat, total assets, net cash flows from operating and investing activities, cash and short term investments are extracted. These variables are then merged to the ThomsonOne and CRSP data.

None of the three previously discussed databases contained (complete) information on the age of a firm at IPO. To collect this data the founding years for each firm in the sample are needed. A website by Jay Ritter1, a prominent researcher in the field of IPO underpricing, contains a dataset with firms’ founding years, which can be used to match founding year data to the previously collected data and allows the construction of the firm age at IPO variable. This website also contains a list of Standard Industrial Classification (SIC)-codes that classifies whether a firm is in the technology industry, as well as a list that classifies when a firm is considered an internet firm. As this research is only interested in high-tech internet firms, I hand-match the two lists to construct a list of internet firms active in the technology industry. By doing so I exclude firms such as online retailers and focus on high-tech internet firms. A list of SIC codes for each category can be found in Appendix B.

(25)

24 The underwriter reputation data along the lines of Carter, Dark, Singh (1998) is also

obtained from Jay Ritter’s website. This dataset, updated in 2012, contains underwriter reputation rankings for different time periods for a very large set of underwriters. For each IPO event in ThomsonOne, the corresponding underwriter rank for the lead underwriter during the year of the underwritten IPO is obtained and matched to the sample. It is

assumed that underwriter reputation did not change between 2012 and 2013, as the most recent update for the dataset was in 2012.

After collecting and merging all the necessary data from ThomsonOne, CRSP and Compustat, the dataset is modified to remove unnecessary data such as non-US listed firms. Companies of which the offer price, issue date or founding year is missing are also dropped. The data sample contains 2,865 unique companies after these corrections.

The variable Media Coverage is constructed by hand-collecting and counting the number of published articles covering each firm in the week leading up to the IPO event, including the first trading day as explained in the previous section. This data is hand-collected from Factiva, a research tool that among other things allows searching for explicit news articles from specific sources for particular time periods. Using Factiva, the amount of articles covering each specific firm in the 1-week period leading up to and including the first trading day of the IPO event are counted and hand-matched to the existing research sample for each of the 2,865 observations. To be as consistent as possible, only articles published through Dow Jones Newswires are incorporated in the count as the target audience for our research consists solely of investors.

4.2 Data description

As explained in the previous section, the sample contains 2,865 unique firm observations, covering IPO events over the period 1993 – 2013. Figure 1 of Appendix A shows the dispersion of these unique events over time, as well as the average underpricing for each year. The graph clearly shows that the amount of firms going public each year was

significantly higher in the years prior to the internet bubble. During the bubble the number of IPOs was still high but showed first signs of decline, whereas the number of IPOs was

(26)

25 much lower shortly after the bubble. As expected, the average underpricing peaked

tremendously following the run-up of new issues during the internet bubble. After 2003 the sentiment for IPOs seemed to return, and the absolute number of annual IPOs goes up again. Again, this is as expected, as most of the overvalued issues have corrected to their normal levels and the market is becoming optimistic again. After 2007, the first year of the financial crisis, the number of new issues that year drops to the lowest point in the sample. In the most recent year sentiment seems to return again, and likely is the beginning of another IPO wave. In addition, it appears that average underpricing is also climbing up again.

Figure 2 of Appendix A shows the same graph, but differentiates between technology and non-technology IPOs. The ratio of technology-IPOs to non-technology IPOs is higher during the internet bubble than any other period in our sample. At the same time, it is evident that the underpricing of technology IPOs is higher for the majority of the years relative to non-technology IPOs. Only shortly after the bubble period is underpricing lower for technology IPOs, which can be explained by investors’ anxiety of technology stocks that suffered tremendous losses after the bubble burst. Interestingly, the gap in

underpricing between technology and non-technology IPO widens again after 2006.

As the methodology of this paper relies on multiple regression analysis, it is important to test whether the different variables are normally distributed. Non-normally distributed variables could distort relationships and significance tests, and therefore render the regression results unreliable. Appendix C contains histograms of the independent

variables so to identify the normality or skewness of the variable distributions, as well as possible outliers. Figure 1 of Appendix C shows the distribution of the variable Burn Rate. The variable Burn Rate has been winsorized at the 1% level. Winsorizing is a statistical method of dealing with extreme values, so to deal with suspicious outliers. Instead of dropping the outliers, they are replaced by the average of the highest or lowest percentile after the cutoff point. The sample thus consists of the same amount of observations as before the Winsorization. As can be seen from the graph, the distribution of the variable Burn Rate is bell-shaped and therefore seems to be normally distributed. Figure 2 shows the distribution of the natural logarithm of Firm age at IPO and also appears to show signs of normality as observed by its bell-shape. Figure 3 shows the spread of the variable

(27)

26 Media Coverage. This variable also shows signs of normality, but as this variable is non-negative there is a clear boundary at y = 0. Finally, figure 4 shows the distribution of the variable firm size, which clearly appears to be normally distributed. It is therefore

concluded that the variables as depicted in Appendix C are approximately normally distributed.

Table 1 of Appendix D shows the descriptive statistics of the variables in the regression model. The sample is split into two categories here, tech firms and non-tech firms. This table gives some valuable information on the distribution of the different variables, and how these behave differently for firms active in the tech industry. One of the first things that meets the eye is that tech firms experience more underpricing on average when we ignore other variables and just look at underpricing by industry. In addition, on average tech firms appear to be of smaller size, have higher cash burn rates, and are subject to higher ex ante risk. Media coverage and firm age are comparable between the two categories.

5 Results

The previous section described the data sources, acquired data and sample distribution. This section describes the regression results and attempts to answer the proposed hypotheses.

Figure 1 of Appendix E shows the first regression output, where both the tech-firm dummy, internet-firm dummy and the interaction term are included. The results indicate that Tech firms are less underpriced relative to non-tech firms, but no evidence is found that

indicates that internet firms are more prone to underpricing. Although this may appear surprising, as internet firms have been widely documented to be more prone to

underpricing during at least the internet bubble, it is important to note that internet firms are very broadly defined in this paper (as can be seen from Appendix B) and are primarily included to identify high-tech internet firms as under the interaction term (see regression equation 2). It is therefore not surprising that Figure 2 of Appendix E shows slightly different results. According to these results, tech firms are found to be more underpriced relative to non-tech firms in the pre-bubble and bubble period. The results are significant at

(28)

27 the 1% and 5% level respectively. In the bubble-aftermath period, however, the sign

changes and tech firms are found to be less underpriced. This makes intuitive sense, as investors typically refrained from investing in tech- and internet related IPOs shortly after the internet bubble burst. More specifically, only 42 tech firms had an IPO in the Bubble Aftermath period. Not only is this considerably less than other periods, these IPOs were likely subject to some degree of self-selection (due to very positive prospects) as investors were generally cautious of internet and tech firms during the bubble-aftermath period. However, tech firms remain negatively related to initial returns in the final period of the model, significant at the 5% level. The results therefore suggest that after controlling for the variables in the model, tech firms are less underpriced than their non-tech peers.

The results of this regression analysis confirm the sign change as documented by Loughran and Ritter (2004): underwriter reputation is positively related to initial returns after 1990. In addition, the results also indicate that the sign-change persists in the period after 2000, the final year analyzed by Loughran and Ritter. Underwriter reputation remains significantly positively related to underpricing for the entire sample period. This confirms hypothesis 2.

Ex ante risk is positively related to underpricing as expected, and is significant for the entire period at the 1% level. The regression results therefore strongly suggest towards confirming hypothesis 3. However, the coefficient of the ex ante risk variable is

extraordinarily strong. It seems as though the Parkinson Extreme Value method of

estimating ex ante risk is not a reliable proxy, as the way the variable is constructed leads is too dependent on the offer price, leading to co-movement with initial returns. A possible explanation is that underwriter price support prohibits the stock of reaching its actual lowest first day trading price, so that the lowest day trading price is “synthetic” and roughly equal to the original offer price of the issuing firm. For this reason, the ex ante risk proxy relies on roughly the same values as the variable initial returns, and therefore likely

overstates the effect of ex ante risk on underpricing. As a robustness check an alternative measure of ex ante risk will be tested to see whether this relationship persists.

The variable firm age at IPO shows a negative relationship with initial returns as expected, showing that younger firms are to a greater extent subject to uncertainty. However, firm

(29)

28 age is only found to be significant during the internet bubble. The variable firm size

indicates that smaller firms experience greater first day returns.

The regression output indicates that the cash burn rate of a firm is significantly positively related to underpricing, rejecting hypothesis 4. A possible explanation is that firms with higher cash burn rates exert more effort in reducing underpricing, as leaving money on the table is more costly for these firms: they know they likely need the capital in the future.

Media coverage is significantly positively related to first day returns, but is not significant in the bubble-aftermath period. This is not surprising, as the bubble aftermath period was expected to be unique as it was subject to low investor sentiment and far less IPOs. These findings therefore confirm hypothesis 5.

Firms that have venture capital or private equity backing are found to experience more underpricing relative to non-backed firms for all years except the bubble-aftermath period. However, as argued before, the bubble aftermath period is expected to be unique as it was dominated by low investor sentiment and far less IPOs. In addition, VC or PE backing is not found to be significant for this period. For this reason, hypothesis 6 is rejected. The results therefore argue against the signaling theory, in favor of Lee and Wahal’s (2004) grandstanding theory.

The participation rate of the firm shows a negative sign for all but the bubble period, and is only found to be significant in the first period. The findings of Habib and Ljungqvist (2001), who noted that firms with more “skin in the game” exert more effort in reducing

underpricing for the period 1991 – 1995, are thus confirmed by the regression results. However, no significant evidence is found that this effect persists in later periods. A

possible explanation is that increased regulation on IPO filings reduced the control issuing firms have over the promotion-cost tradeoff. However, further research is needed to

identify the specific causes.

Figure 3 of Appendix E shows the second regression output. The variables show signs and strengths comparable to the first regression results. High tech internet firms, as depicted by the interaction term between Tech- and internet firms, are highly underpriced

(30)

29 during and prior to the bubble as expected. However, no evidence is found for excessive underpricing of high tech internet firms after the bubble. As tech firms were also only found to be underpriced more during the first two periods under the first regression, Hypothesis 1 is only confirmed for the first two periods.

As figure 1 of Appendix A showed that it appears that 2013 is the beginning of a new IPO wave, the regression is run again for that year specifically. Figure 4 of Appendix E shows the regression output. Again, no evidence of underpricing is detected for either tech firms or high tech internet firms, further adding to the evidence that hypothesis 1 only holds for the first two periods.

It must be noted that the explanatory power of all regression models is relatively low, ranging between R-squared values of 0.16 and 0.30. This means that the model explains approximately between 16% and 30% of the variation of the response variable.

Nevertheless, as most of the regression results are very significant, some useful conclusions may still be drawn regarding the signs of the different variables.

In addition to initial returns, columns (5) through (8) of figures 1, 2 and 3 of Appendix E show the regression results with 60-day returns as the dependent variable. Except for the bubble period, no evidence is found that tech or internet firms perform significantly better in the 2 months following the IPO. In addition, most variables that play a role in explaining exceptional first day returns are not found to be significant in explaining long run

performance as expected. An interesting exception is that new issues underwritten by high quality underwriters do appear to perform better in the long run.

6 Robustness checks

The previous section discussed the regression results and how this affects the hypotheses. This section will investigate the robustness of the research by further analyzing the standard regression assumptions and investigating some alternative regression analyses.

(31)

30 Multiple regression analyses relies on several critical assumptions such as normal

distributed independent variables and uncorrelated error terms. When running a multiple regression analysis, it is also critical to test the variables for multicollinearity.

Multicollinearity arises when two or more independent variables are highly correlated, inflating coefficient estimates and reducing overall model reliability. One possible to

method to detect multicollinearity is to look at Variance Inflation Factors (VIFs). These VIFs tell us the extent to which the standard error (and so the variance) of the coefficient of interest has been inflated upwards. A general rule of thumb is that when the VIF exceeds 5, multicollinearity is likely a problem. Figure 1 and 2 of Appendix G summarize the VIFs corresponding to the regression results of the previous section. All values are found to be very close to 1, implying that a very low degree of multicollinearity is present in the model. As an alternative robustness check, figure 3 and 4 of Appendix G show the correlation matrices of the regression analyses. Correlation is found to be very low between all

variables except between the variable Firm Size and Cash Burn Rate, but as it is below 0.5 it is still within acceptable boundaries. These tables therefore further support the earlier findings that there is low correlation among the independent variables in the model.

One possible reason for concern in the regression analysis of the previous section is that the data sample contains several IPOs with very small offer prices. These IPOs, which are typically classified as penny stocks, are generally subject to high degrees of liquidity risk and information asymmetry due to their limited reporting and disclosing requirements (Bradley et al., 2006). Consequently, the stock price is expected to behave differently compared to ordinary IPOs. It therefore makes sense to adjust the data sample to only contain ordinary IPOs by dropping issues with an offer price of less than $5.00. By doing so, 31 of the 2865 observations are deleted. The vast majority of these penny stocks went public in the pre-bubble period. This makes intuitive sense, as restrictions for issuing penny stocks increased significantly since the manipulation and fraud surrounding these firms prior to 1990. Figure 1, 2 and 3 of Appendix F contain regression results adjusted for penny stocks. Due to the limited amount of penny stocks in the sample, signs and

strengths of the variables remain comparable to the previous results.

As briefly discussed in the previous section, it appears that the Parkinson Extreme Value method of estimating ex ante risk overstates the effect of ex ante risk on underpricing,

(32)

31 likely due to underwriter price support. To remedy this problem, this paper suggest an alternative proxy for ex ante risk. Arnold et al. (2010) argue that when a prospectus is more complex, investors show greater reluctance to invest in the IPO due to higher perceived ex ante risk. By looking at the length of an issuing firms’ prospectus, this paper attempts to approximate valuation complexity as an alternative proxy for ex ante risk.

A hand collected sample of prospectus lengths is used to compute this alternative valuation complexity proxy. The sample is derived from the SEC’s Edgar database, and contains the absolute raw word count of an issuing firms’ prospectus. This variable is corrected for outliers by winsorizing the 10 highest outliers. The distribution of the word count data can be found in figure 5 of Appendix C, and appears to be normally distributed.

The regression results after substituting the Parkinson Extreme Value method of

estimating ex ante risk for the prospectus length variable and dropping penny stocks is depicted in figure 4 of Appendix F for tech firms, and in figure 5 for high tech internet firms. In both regressions, the word count variable is only significant during the bubble period. In addition, it shows a negative sign, indicating that more complex prospectuses are

associated with lower initial returns. A possible explanation here could be that more thorough prospectuses reduce asymmetric information and thus lead to lower

underpricing. However, not much can be inferred from these results, as the variable is not found to be significant for 3 out of the 4 periods. The other variables remain consistent with the previous regression results: tech firms remain significantly underpriced for the pre-bubble and pre-bubble period, but there is still no evidence for excessive underpricing after that. High-tech internet firms on the other hand are also again significantly associated with underpricing for the first two periods, but are now also found to experience higher initial returns in the final period. Long run performance, as depicted by the regressions of column (5) through (8), is again poorly explained by the variables in the model. This is in line with expectations. Using prospectus complexity as a new approximation of ex ante risk has no further interesting effects on the 60-day performance results.

The regression results up to this point all indicate a positive sign for the variable cash burn rate. A positive sign implies that firms with higher burn rates are less prone to

Referenties

GERELATEERDE DOCUMENTEN

One of the equilibria born in the saddle node bifurcation turns stable and an unstable limit cycle emerges through a subcritical Hopf bifurcation.. When we enter region (4), we are

Ook situationele kenmerken kunnen hier aan bijdragen, zoals een lage financiële behoefte (iemand gaat bijna met pensioen, kan op zijn partner bouwen of heeft geen kinderen) of de

in terms of energy, memory and processing, temporal, spatial and spatio-temporal correlation among sensor data can be exploited by adaptive sampling approaches to find out an

Therefore, because the the p-value are too small , with 95% confident interval , the amount of Market value on Equity and inverse price level also the turnover rate

A significant postive beta for the contrarian portfolio indicates that losers have more systematic risk than winners (Locke and Gupta, 2009). Six significant and positive betas

The first model uses the present value of abnormal earnings of the three years after going public, the second model only the two subsequent years and the third model only one year..

Five cluster groups of inland valleys were identified: (i) semi-perma- nently flooded with high soil organic carbon (4.2%) and moderate available phosphorus (10.2 ppm), mostly

If another user is detected speaking at the same time as the current turn owner, the interruption management function is initiated and the robot briefly directs an angry gaze