How Do Investment Banks Value Initial Public Offerings (IPOs)?
Address for correspondence: Marc Deloof, University of Antwerp, Prinsstraat 13, 2000 Antwerp, Belgium. e-mail: email@example.com
Abstract: We investigate the valuation and the pricing of initial public offerings (IPOs) by investment banks for a unique dataset of 49 IPOs on Euronext Brussels in the 1993–2001 period. We find that for each IPO several valuation methods are used, of which Discounted Free Cash Flow (DFCF) is the most popular. The offer price is mainly based on DFCF valuation, to which a discount is applied. Our results suggest that DDM tends to underestimate value, while DFCF produces unbiased value estimates. When using multiples, investment banks rely mostly on future earnings and cash flows. Multiples based on post-IPO forecasted earnings and cash flows result in more accurate valuations.
A firm conducting an initial public offering (IPO) needs to have its stock valued before the IPO, in order to determine a price range within which the stock will be offered to the public. There are several methods available for stock valuation. The most widely used valuation approaches are the dividend discount model (DDM), the discounted free cash flow (DFCF) method, and valuation approaches that rely on multiples of firms in similar industries and firms involved in similar transactions.
While there is a very extensive literature on IPOs, and there are many papers on the choice and accuracy of valuation methods, few papers focus on the valuation of IPOs. We are aware of only one study that investigates the choice of valuation methods for IPOs, and two studies that focus on the accuracy of IPO valuation.1Roosenboom (2007) investigates how French underwriters value the stocks of firms they take public, and finds that the valuation methods used depend on IPO firm characteristics, aggregate stock market returns and stock market volatility. Kim and Ritter (1999) value a sample of IPOs in the US using price-earnings (P/E) and price-to-book comparables, and find that these methods lead to very imprecise valuations when historical accounting numbers are used. However, when forecasted earnings are used, the accuracy of the valuation improves substantially. Berkman, Bradbury and Ferguson (2000), who value 45 newly listed firms in New Zealand, conclude that the best discounted cash flow and P/E valuations have similar accuracy.
An important feature of existing studies on the accuracy of IPO valuation methods is that they use ex post value estimations by the researcher(s). In Belgium, pre-IPO value estimates by the lead underwriting investment bank are often published in the IPO-prospectus, which is made available at the start of the public offering. Until recently, the Banking, Finance and Insurance Commission, the supervising authority for Belgian financial markets, requested that this information should be included in the prospectus for domestic public offerings. In most other countries, this information is not publicly available. Furthermore, even in Belgium, this is not the case anymore due to the ‘Prospectus Directive’ (Directive 2003/71/EC), which had to be transposed into national law by all EU member states before July 1, 2005. In all member states, the IPO prospectus will need to comply with international standards, and a section on the company's valuation/justification of the pricing will not be included anymore.
The availability of this information in Belgian prospectuses allows us to examine how IPOs are valued, and how this valuation affects the pricing of IPOs. It can be expected that the accuracy of ex ante valuation by investment banks will differ from the valuation accuracy measured by academics, for several reasons. Value estimates by investment banks may be less accurate because academics are more objective than investment banks, who may be tempted to report valuations that justify a high price, for instance by choosing comparables with high multiples, or a low price, in order to reduce marketing efforts. On the other hand, value estimates by investment banks may be more accurate than value estimates by academics because investment banks have more information for valuation available. Moreover, as the stock market is pricing perceptions of the future and not the future itself, the value estimates by lead underwriters and the offer price, which to some extent will be based on these value estimates, may influence these perceptions and therefore the stock price. However, in an efficient market mispricing by underwriters should not affect market valuation.
In this paper, we investigate the valuation and pricing by the underwriters of 49 IPOs on Euronext Brussels (formerly the Brussels Stock Exchange) in the 1993–2001 period. We address two research questions. First, we want to know how IPOs are valued. What valuation models do underwriters use, and how do they set the offer price, given the value estimates? Second, we investigate which of the valuation models, as used by underwriters, provides the best estimation of the stock market price. Do these valuation methods produce unbiased results, and what is the accuracy of the valuations? We find that for each IPO several valuation methods are used, of which DFCF is the most popular method: the DFCF model is used to value all IPOs in the sample. We also find that the offer price is set closer to DDM estimates if DDM is applied. This is remarkable, as a comparison of pre-IPO valuations to the average stock price in the first month of listing and to the stock price on post-IPO days +1, +10, +20 and +30 suggests that DDM tends to underestimate value, while DFCF produces unbiased results. Interviews with investment bankers indicate that underwriters consciously underprice IPOs, by applying a deliberate discount to DFCF value estimates. DFCF is considered to be the most reliable method. DDM estimates are on average closer to the preliminary offer price than other value estimates, because DDM tends to underestimate value. Furthermore, we find that P/E and price/cash flow (P/CF) multiples using earnings and cash flows in the IPO-year lead to less accurate valuations than multiples using forecasted earnings and cash flows in the year after the IPO, which is consistent with results of Kim and Ritter (1999). Finally, our results indicate that the final offer price is closer to the stock market price than pre-IPO value estimates. This is consistent with the expectation that the final offer price incorporates valuable information about investor demand, obtained by the underwriter during the public offering.
This paper makes two contributions to the literature. First, it explores the frequency of use of alternative valuation models within the IPO pricing context, providing evidence that changes the consensus in the valuation model choice research area. Our finding that DFCF is the most popular valuation model contrasts with previous work, which shows that financial analysts primarily focus on multiples and tend to ignore discounted cash flow models (e.g., Block, 1999; Barker, 1999a and 1999b; Bradshaw, 2002; Demirakos et al., 2004; and Asquith et al., 2005). We put forward various possible explanations for the extensive use of DFCF in our sample. First, there may be a lack of comparable firms, which makes methods based on multiples difficult to implement. Second, there may be differences in the perceived importance that investment banks attach to alternative valuation techniques. A third reason may involve time-specific effects: the majority of our sample went public at the end of the 1990s, a period of very high stock market levels during which the use of multiples might have led to overvaluation. Finally, differences in the quality and objectives of IPO prospectuses and equity research papers may account for our finding.
The second contribution is that the paper uses ‘real world’ estimations to investigate the accuracy of valuation models within the IPO pricing context, in contrast to Kim and Ritter (1999) and Berkman, Bradbury and Ferguson (2000) who use ex-post value estimates by academics. Some studies examine the accuracy of earnings forecasts in IPO prospectuses, but do not focus on valuation accuracy (e.g., Firth and Smith, 1992; Jaggi, 1997; Jelic, Saaudouni and Briston, 1998; Cheng and Firth, 2000; Gonoupolis, 2003; and Jog and McConomy, 2003). Other papers (e.g., DeAngelo, 1990, for management buyouts) investigate the ‘real world’ use of valuation models in different settings but do not discuss valuation accuracy.
This paper is closely related to Roosenboom (2007), who investigates how French underwriters value the stocks of firms they bring public. Our paper complements and extends the work of Roosenboom in several respects. First, while the study of Roosenboom specifically focuses on how IPOs are valued, we also offer empirical evidence on the accuracy and bias of the implemented valuation methods. Second, we provide much more detailed information about multiple valuation, whereas Roosenboom combines all multiple valuation methods into one single group. Third, we use interviews to interpret the quantitative results and explain the relation between the estimates derived by alternative valuation models and the preliminary offer price. Fourth, while Roosenboom investigates the French stock market, our study offers empirical evidence on the valuation of IPOs for another, smaller stock market. Our findings reveal some interesting differences as compared to Roosenboom. Contrary to Roosenboom, we find that DFCF and not multiples is the most important valuation model. Our results also suggest that investment bankers base the preliminary offer price primarily on DFCF regardless of the characteristics of the analyzed firm. The preliminary offer price is derived by applying a discount to the DFCF valuation estimate.2
The remainder of the paper is organized as follows. In the next section we briefly discuss the choice and accuracy of valuation models. Section 3 describes the IPO process in Belgium. The sample and the methodology we use are discussed in Section 4. In Section 5 the valuation outcomes are compared to the IPO offer price and the average stock market price in the first month of listing. Section 6 discusses the relation between IPO valuation and underpricing. Finally, Section 7 presents some conclusions.
2. VALUATION MODEL CHOICE AND THE ACCURACY OF VALUATION MODELS
(i) Valuation Model Choice
A number of authors have argued that multi-period valuation models based on discounted cash flows or residual income are superior to single-period multiple valuation approaches, which are likely to result in less accurate valuations (e.g., Copeland et al., 2000; Palepu et al., 2000; and Penman, 2007). However, the empirical evidence on valuation models used by professional investors and financial analysts stands in contrast to the theoretical superiority of multi-period valuation models. The strongest and most consistent empirical finding is the primary importance of the P/E ratio (e.g., Govindarajan, 1980; Previts et al., 1994; and Yap, 1997). Another consistent finding is that DFCF models, technical analysis and beta analysis are rarely used in investment decisions (Arnold and Moizer, 1984; Pike et al., 1993; and Block, 1999). For equity analysts, the most prevalent base for target prices are P/E ratios and forecasted long-term earnings growth rates, or a combination of these two constructs (the PEG ratio). They rarely use present value techniques in equity valuation (Block, 1999; and Bradshaw, 2002). The inherent uncertainty of projecting future cash flows and determining an appropriate discount rate may make multi-period cash flow valuation appear too difficult (Barker, 1999a). Asquith et al. (2005) find that almost all analysts use earnings multiples in their reports. Only 12.8% of the analyst reports in their sample include any variant of DFCF, while 25.1% of the reports include asset multiples. Another finding of the empirical literature is that the use of option pricing models or the residual income model is extremely limited (Demirakos, Strong and Walker, 2004).
The use of valuation models in analyst reports varies according to industry. Barker (1999b) finds that the P/E ratio is the dominant valuation model in the services, industrials and consumer goods sector. The dividend yield is more important for financials and utilities. Comparative valuation models are more popular in relatively stable sectors where conventional accounting does a better job of capturing the value of the firm (Demirakos, Strong and Walker, 2004).
A number of studies focus on the valuation of high-tech and internet stocks, examining whether traditional valuation frameworks are relevant for such firms (e.g., Hand, 1999 and 2000; Schwartz and Moon, 2000; Trueman et al., 2000; Demers and Lev, 2001; and Bartov et al., 2002). The results of these studies are mixed: while some studies find that the traditional valuation models play a key role, others find that valuations tend to differ from the traditional methods. A recent survey with telecommunication company equity analysts by Glaum and Friedrich (2006) shows that both DFCF and multiples are always used, but that DFCF is the dominant technique. Analysts use multiples but only to validate their DFCF results, not as an independent valuation tool. Most analysts indicated changes in the relative importance of valuation models with DFCF now being more important than at the end of the 1990s.
(ii) The Accuracy of Valuation Models
Various studies have examined the accuracy of valuation models. Some of these studies focus on multiples valuation, and provide mixed results on which multiples have the highest valuation accuracy (e.g., Beatty, Riffe and Thompson, 1999; Kim and Ritter, 1999; and Liu, Nissim and Thomas, 2002). A number of papers investigate how the choice of comparable firms affects accuracy of multiples. Profitability, growth and risk are important variables in peer group selection, and the use of harmonic means generates the best results (e.g., Boatsman and Baskin, 1981; Alford, 1992; Cheng and McNamara, 2000; and Bhojraj and Lee, 2002). A consistent result is that multiples based on forecasted earnings lead to higher valuation accuracy than multiples using trailing earnings (see Kim and Ritter, 1999; Liu, Nissim and Thomas, 2002; and Lie and Lie, 2002). This is not unexpected, as other studies suggest that earnings forecasts capture information on value that is not reflected by historical earnings (e.g., Tse and Yaansah, 1999; Liu and Thomas, 2000; and Barniv and Myring, 2006). Yee (2004) shows that in a setting of unobserved information, forward earnings are better valuation attributes than trailing earnings. Furthermore, the value-relevance of earnings is enhanced by a focus on permanent, as opposed to transitory components of earnings (see Barker, 1999b, for a review).
Some studies compare the accuracy of multiples valuation and discounted cash flow valuation. Kaplan and Ruback (1995) examine the discounted cash flow and comparable firm approaches in the context of highly leveraged transactions, and conclude that both approaches are useful and reliable. According to Kaplan and Ruback, discounted cash flow valuation methods perform at least as well as valuation approaches using companies in similar industries and companies involved in similar transactions. Gilson, Hotchkiss and Ruback (2000) find that, for firms that reorganize in bankruptcy, the discounted cash flow and comparable firm approaches have about the same degree of accuracy and lead to estimates that are generally unbiased but not very precise. Berkman, Bradbury and Ferguson (2000), value 45 newly listed firms in New Zealand, and also conclude that the best discounted cash flow and multiples valuations have similar accuracy. Asquith et al. (2005) find that the accuracy of earnings multiples and DFCF used by equity analysts in predicting price targets is fairly similar. Analysts are least successful in predicting target prices when they use EVA or a ‘unique’ alternative valuation method.
Penman and Sougiannis (1998) and Francis, Olsson and Oswald (2000) compare accrual earnings valuation to discounted cash flow valuation. Interestingly, both studies find that accrual earnings techniques produce lower valuation errors than discounted cash flows and dividends. However, Lundholm and O'Keefe (2001) claim that the superiority of accrual earnings valuation found by these papers is misguided, and is only due to the researchers' actual implementation containing inconsistencies, as accrual earnings valuation and discounted cash flow valuation are theoretically equivalent (Ohlson, 1995). Bernard (1995) and Walker (1997) argue that accrual earnings valuation is superior under information-constrained conditions.
3. THE IPO PROCESS IN BELGIUM
As our analysis focuses on the valuation of IPOs, it is useful to briefly describe the IPO process in Belgium. This process is very similar to IPO procedures in other countries, such as the US and the UK. Once the board of directors of a company has decided to go public, the company will hire an investment bank to underwrite the offering. Usually, a group of co-underwriters is formed to help sell the issue to the public. A profound due diligence is carried out, checking business, legal, financial and tax issues, and all legal documentation is put together. Limited pre-marketing activity may take place to obtain some feedback from institutional investors about their potential interest for the new issue. A prospectus is drafted which contains, amongst other things, financial information about the company and the terms of the offer. It is the only document the company can use to communicate with potential investors during the IPO. In Belgium, it will very often contain estimates of the company value by the lead underwriter. Before the shares can be offered to the public, the prospectus has to be approved by the Belgian Banking, Finance and Insurance Commission.
During the offering period, which lasts one or two weeks, investors can place bids for shares, usually within a range of potential offer prices, and sometimes at a unique offer price. The offer price range depends on the outcomes of the underwriter's valuations, and may also reflect information obtained from pre-marketing. Most IPOs in Belgium make use of the bookbuilding method, in which the underwriter builds a book of likely orders and uses this information to set the final offer price. The underwriter organizes road shows during which the new issue is marketed to investors. A few days after the public offering period, the final offer price is set within the offer price range (if applicable), shares are allocated to investors, and the share starts trading on the stock market. The final offer price is therefore set after the underwriter has obtained information about investor demand, which is not available at the time when the underwriter sets the preliminary offer price range on the basis of the value estimates. The final offer price also takes into account current market conditions, and is the outcome of a negotiation process between the issuer, and the underwriter's corporate finance and sales team.
4. SAMPLE AND METHODOLOGY
Our sample includes 49 IPOs on Euronext Brussels from 1993 to 2001, for which valuation information is available. Between January 1984 and June 2005, 103 companies have been introduced on the First Market or on the New Market of Euronext Brussels. The New Market (also called Euro.NM Belgium) was set up in 1997 for young, high growth firms. The 30 IPOs before 1993 could not be included in the sample because the prospectus did not contain information on value estimates. For the limited number of IPOs after 2001, the prospectus also contains insufficient valuation information, due to the EU ‘Prospectus Directive’. Some IPOs in the period 1993–2001 which were primarily aimed at international investors, also had to be left out of the sample due to a lack of detailed valuation information (e.g., Interbrew and Mobistar). After excluding IPOs of mutual funds, financial institutions and holding companies, we eventually retain 49 IPOs, of which 15 were listed on the New Market.
Table 1 shows that most companies in our sample went public in the hot IPO market at the end of the 1990s. The IPO firms are active in a wide range of industries, including both ‘high tech’ industries and industries that do not rely on sophisticated technologies (see Table 2).
Distribution Across Years
Forty Nine IPOs on Euronext Brussels Between 1993 and 2001 by Year of Offering
Distribution Across Industries
Forty Nine IPOs on Euronext Brussels Between 1993 and 2001 by Industry
|Clothing and footwear||2|
|Medical equipment and supplies||2|
|Restaurants and pubs||2|
|Textiles and leather goods||2|
|Chemicals – advanced materials||1|
|Publishing and printing||1|
Table 3 presents some descriptive statistics. The median firm introduced at Euronext Brussels offered 858,678 shares at a preliminary offer price of 28.8 EURO. The preliminary offer price is the midpoint of the minimum and maximum offer prices, or, if the shares are not offered within a price range, the unique offer price (see Kim and Ritter, 1999).3 The initial returns, which are calculated as [average price in the first month of listing/offer price]–1, were generally substantial on Euronext Brussels: the median initial return is +9.5%, and the mean initial return is +19.3%.4
|Number of shares offered to the public||1,132,361||1,066,284||116,809||858,678||6,250,000|
|Preliminary offer price (EURO)||40.4||96.4|| 1.7||28.8||694.1|
|Average price in the first month of listing (EURO)||47.0||98.1|| 2.1||32.6||695.8|
|Initial return (%)||19.3||37.3||−22.3|| 9.5||190.5|
| ||Number of Firms Offering|| || |
|Only existing shares||9|| || |
|Existing and new shares||25|| || |
|Only new shares||15|| || |
For most IPOs, at least a part of the shares are sold by existing shareholders: nine firms offer only existing shares; 25 firms offer both existing and new shares; 15 firms offer only new shares.5 These findings are consistent with evidence for other European countries that a major motivation for European firms to go public is to allow the controlling shareholder to divest from the firm (see Rydqvist and Högholm, 1995; Pagano, Panetta and Zingales, 1998; and Ritter, 2003).
All lead underwriters are Belgian banks, with the exception of ABN Amro Rothschild, which is co-lead underwriter of two IPOs, and Indosuez, which is co-lead underwriter of one IPO. The (co-)lead underwriters are Generale Bank/Fortis Bank (14 IPOs), Smeets Securities/Delta Lloyd Securities (12 IPOs), Bank Brussel Lambert (9 IPOs), KBC Securities (7 IPOs), Petercam (7 IPOs), Paribas/Artesia Bank/Dexia (5 IPOs), Bank De Groof (4 IPOs), Lessius (1 IPO), Delen & Co (1 IPO), Van Moer Santerre (1 IPO) and Nedee (1 IPO).
In a first step, we examine how IPOs are valued by (a) considering the frequency of use of different valuation methods and (b) assessing the relative importance of valuation methods in determining the preliminary offer price. For each valuation method we compute the proximity of value estimates to the preliminary offer price by the natural log of the ratio of the estimated value to the preliminary offer price. While we will refer to this measure as a ‘pricing error’, it should not be interpreted as an error in the strict sense, but only as a quantification of the proximity of value estimate and preliminary offer price. We evaluate the degree of central tendency of the preliminary offer price to value estimates by the percentage of differences within 15% and the mean absolute error. Mean and median pricing errors are calculated to indicate the extent to which the preliminary offer price is set higher (negative pricing error) or lower (positive pricing error) than the value estimates. We will focus on median errors, which are less affected by outliers than mean errors (although the difference in our sample is limited).
In a second step, we will investigate which of the valuation models provides the best estimation of the stock market price. The proximity of estimated value to the market value is computed as the natural log of the estimated value over market value. As in other studies of valuation methods, we use mean and median valuation errors to indicate the extent to which value estimates are biased: do the value estimates tend to be on average higher or lower than the market value? The accuracy of valuation is measured by the percentage of valuation errors within 15% and the mean absolute error, both indicating the dispersion of the valuation errors.6
Our measure of valuation accuracy, which is the standard measure in the literature, is based on a comparison of value estimates with stock market prices. This measure assumes that (1) value estimates reported in the prospectus are the actual value estimates of the underwriter, and (2) stock market prices reflect fair value. Regarding the first assumption, it cannot be ruled out that the reported value estimates are lower than the actual estimations of the underwriter. It is widely acknowledged that IPOs are underpriced. If underwriters consciously underprice IPOs, they may have an incentive to report value estimates that are lower than their ‘true’ value estimates. If this is the case for the IPOs in our sample, we expect to find that the reported value estimates for different methods tend to underestimate stock market prices. However, it can be argued that even if the reported value estimates are lower than the actual value estimates of the underwriter, a comparison of valuation errors will yield consistent results, as long as reported value estimates are underpriced to the same degree. Moreover, we interviewed representatives of seven lead underwriters, who covered 45 of the 49 IPOs in our sample.7 They all claim that the valuations reported in the IPO prospectus reflect the true outcomes of their valuation models.
The assumption that stock market prices reflect fair value, implies that the stock market is efficient and that any underpricing is corrected for very fast when trading starts. Most studies on underpricing indeed find abnormal initial returns on the first day of trading, but not afterwards, which suggests that underpricing is corrected at the first trading day (e.g., Ritter, 1998). Investors may also overreact when trading starts, thereby driving stock prices above fair values. Some studies on the long-run underperformance of IPOs find support for this theory (e.g., Aggarwal and Rivoli, 1990; Ritter, 1991; Loughran and Ritter, 1995; Ritter and Welch, 2002; Purnanandam and Swaminathan, 2004; and Álvarez and González, 2005). Stocks can also become overvalued because underwriters tend to support prices in the after-market (e.g., Aggarwal and Rivoli, 1990; and Schultz and Zaman, 1994). Stock market prices might therefore not reflect fair value at all times. However, it can be argued that if the underwriter bases his valuations upon the same expectations as the stock market, a comparison of valuation errors will yield consistent results, regardless of overvaluation in the stock market. In order to minimize the risk that our empirical results are driven by short-term underpricing or overpricing, we measure stock market prices (‘fair value’) over a range of different time periods (post-IPO day 1, +10, +20, +30, average first month).
Another potential problem when examining the valuation accuracy is that the information set changes when trading of the IPO begins, and investors are not restricted to the investment bank's information set (Berger, 2002). This would mean that even though the value estimates of underwriters may be right at the time they were made (ex ante), they may be wrong afterwards (ex post), potentially distorting our results. However, given the rather short period of time between the publication of the IPO prospectus and the first day of trading, we expect this effect to be limited. Moreover, the period between the publication date of the prospectus and the first trading date falls within the ‘blackout’ or ‘silent period’, during which there are strong restrictions with respect to the release of new information about the company.
Our quantitative measures, as described above, do however suffer from some drawbacks. First, they do not provide information about the practical implementation of the valuation models. Second, they do not give us a clear answer on which valuation method underwriters really rely to price IPOs. Qualitative research methods allow identifying hitherto unexpected and undisclosed relationships. Combining quantitative and qualitative research methods compensates for one another's weaknesses, leading to more relevant and reliable findings than would be achieved by either method in isolation (Barker, 1999b). Thus, in order to gain insight into the way underwriters value and price IPOs, seven investment bankers, who were the lead underwriter of 45 of the 49 IPOs in our sample, were interviewed. The semi-structured interviews lasted 60–90 minutes; they were tape-recorded and transcribed in text documents. The major advantage of semi-structured interviews is that they allow interviewees to express opinions on wide-ranging predetermined issues and also to respond to supplementary questions seeking clarity, consistency and full explanation. These benefits have to be weighted against potential problems of subjectivity and bias in the data and the interpretation of the data (e.g., Barker, 1999a; and Silverman, 2006). Conducting interviews enables us to obtain differentiated answers, especially in the case of complex issues. Moreover, the interviewer can remove possible misunderstandings by clarifying certain aspects or seeking further explanation which results in higher validity of data (Neuman, 1999). A recent study by Imam, Barker and Clubb (2008) also combines qualitative and quantitative research techniques. They investigate analysts' use of valuation models by offering a content based analysis of equity research reports and using a semi-structured interview based approach to interpret the quantitative results.
(i) Valuation Model Choice
How do the lead underwriters value the firm?Table 4 contains the valuation methods used to value the 49 IPOs, and the number of cases in which they are applied.8 All lead underwriters mention only the use of generally accepted valuation methods and seem to avoid eccentric multiple valuation methods as the ones described by Fernandez (2001).9 DFCF is the most popular method: it is used to value all IPOs. DDM is used for 24 IPOs, and a multiples approach is used to value 40 IPOs. All underwriters use at least two different valuation methods. Although multiple valuation is often used in practice, it is flawed from a theoretical perspective. Its economic rationale is questionable as the analysis is not anchored in fundamentals like future cash flows, growth opportunities and risk that tell us about value independently of market prices. Next, the method of multiples assumes that the market is efficient in setting prices for the comparables. Moreover, multiple valuation is often of a static nature, whereas discounted cash flow methods can better handle the dynamic nature of a business environment.
Valuation Methods Used by Lead Underwriters
|Discounted free cash flow||49|
|Dividend discount model||24|
| Peer group||34|
| Stock market||14|
| Growth shares|| 2|
|▪ Price/cash flow||17|
| Peer group||15|
| Stock market|| 8|
| Growth shares|| 3|
|▪ EnterpriseValue/EBITDA (peer group)|| 8|
|▪ EnterpriseValue/sales|| 3|
| Peer group|| 3|
| Stock market|| 1|
|▪ Price/book (peer group)|| 1|
|▪ Dividend yield (peer group)|| 2|
|▪ P/E-to-growth (peer group)|| 1|
Conceptual problems aside the method of comparables also has problems in implementation. Identifying peers with similar operating and financial characteristics is difficult. The method leaves too much room for ‘playing with mirrors’ and too much freedom for the analyst to obtain a desired valuation. In addition, different multiples give different valuations—it is not clear which is most reliable (Penman, 2007). Moreover, failure to make appropriate adjustments to peer companies' financial statements and a simple reliance on mean or median peer group multiples without comparative analysis may further cause relative valuation by multiples to produce inaccurate outcomes (Pratt, Reilly and Schweihs, 2000). However, in contrast to DFCF, multiple valuation does not require the tough work of adequately estimating cash flows and appropriate discount rates (cf. Lie and Lie, 2002). P/E and P/CF are the most popular multiple approaches: P/E is applied for 37 IPOs and P/CF is applied for 17 IPOs. Other multiple approaches used are EnterpriseValue/EBITDA (8x), EnterpriseValue/Sales (3x), Price/Book (1x), Dividend yield (2x) and P/E-to-Growth (1x).
A problem with using multiples to value Belgian IPOs is that the number of firms listed on Euronext Brussels is limited. At the end of 1999 only 144 firms were listed, many of which are financial institutions and holding companies. It is therefore often difficult to find a sufficient number of comparable firms. In several cases, the underwriter compares the IPO to the Euronext Brussels stock market, as well as to a peer group of firms, in order to estimate value. For a few IPOs, value is also estimated using the average P/E or P/CF of growth shares on Euronext Brussels.
Multiples can be based on historical earnings or cash flows, but also on forecasted earnings and cash flows. Most multiples used in our sample are based on current year's forecasted earnings and cash flow (year 0) or next year's forecasted earnings or cash flow (year +1). In a limited number of cases the underwriter also uses historical earnings and cash flow in the year before the IPO (year −1) and/or the forecasted earnings and cash flow for the second post-IPO year (year +2). This leads to a wide range of multiples used by investment banks, along three dimensions:  type of multiple (P/E, P/CF, Price/book ..),  the firms to which the IPO is compared (peer group, stock market, growth shares), and  the timing of the multiple (years −1, 0, +1, +2). In the remainder of the paper, we will investigate the estimations of the most frequently used multiples: P/E and P/CF, for a peer group and for the stock market, in years 0 and +1.
The usefulness of valuation methods may vary according to firm characteristics (Roosenboom, 2007). In order to identify distinctions between the firms that are valued with each valuation model, we report median values for key firm characteristics drawn from the prospectuses and test the significance of differences using the non-parametric Mann-Whitney test. The firm characteristics we consider are firm age, firm size, profitability, sales growth and dividend payout. We measure firm age as the difference between the IPO year and the founding year. According to Kim and Ritter (1999), it is more difficult to forecast the future dividends of younger firms without established track records. Firm size is proxied by total assets in the accounting year preceding the IPO (expressed in million EURO). Larger firms are likely to have more stable dividends which are easier to forecast. Profitability is measured by EBIT/Sales, which is the ratio of current year's forecasted earnings before interest and taxes to current year's forecasted sales, and by EBIT/Total assets, which is the ratio of current year's forecasted earnings before interest and taxes to current year's total assets. Underwriters are more likely to value highly profitable firms with DDM, as these firms can pay out substantial dividends. Furthermore, firms that are only marginally profitable or loss reporting in the current year are more likely to be valued with forward looking multiples than with multiples based on current profits. Sales growth is equal to forecasted sales growth during the current year. We expect that high growth firms are more likely to be valued with multiples since multiples are better suited to value growth opportunities than DDM (Roosenboom, 2007). Moreover, high growth firms are more likely to retain earnings instead of paying them out as dividends. They are therefore less likely to be valued with DDM. Dividend payout is dividend over net income expected for the year after the IPO. (High) dividend paying firms are more likely to be valued with DDM.
Table 5 compares firms valued with DDM to those for which DDM is not used and presents similar information for P/E and P/CF. For firms valued with P/E, we further differentiate between firms valued with current P/E only, forward P/E only and both current and forward P/E.10 As all the firms in our sample use DFCF, we cannot differentiate between firms valued with DFCF and other firms. Table 5 reveals that firms valued with DDM are relatively older, larger and more profitable. They also have a higher dividend payout and lower forecasted sales growth. This is in line with the findings of Roosenboom (2007). While there are no significant differences between firms valued with P/E and other firms, we do find that young, loss making firms without dividends are more likely to be valued with forward P/E only than with current (and forward) P/E. Remarkably, companies valued with the P/CF multiple tend to be relatively old and large, have low growth prospects and pay dividends. The median number of methods used to value P/CF firms is seven, while the overall median number of methods used is only three. This may suggest that the P/CF multiple is not considered to be a central valuation method, but is only used as a supplementary method by underwriters of large, mature firms.
Valuation Methods Used by Lead Underwriters: Median Values for Firm Characteristics
|Firm age (Years)||21.50***||7.00||10.00||12.00||16.00||11.50||11.50**||6.00||26.50***||8.00|
|Firm size (Mio EURO)||55.63***||10.25||21.73||17.60||147.82||42.88||42.88||2.84||72.45***||16.51|
|EBIT/Total assets (%)||9.52***||1.29||4.89||6.29||7.77||10.33||10.33***||−7.06||8.37||5.00|
|Sales growth (%)||8.89***||56.67||28.20||36.00||10.15||15.31||15.31||66.38||9.82**||40.87|
|Dividend payout (%)||30.00***||0.00||0.00||10.00||35.00||30.00||30.00**||0.00||32.50***||0.00|
|Number of methods used||5||2||4||2||5||6||6||2||7||2|
|Number of observations||24||25||37||12||4||20||20||12||17||32|
(ii) Valuation Models and IPO Pricing
In this section, we investigate which valuation model is most closely linked to the preliminary offer price and on which valuation method(s) lead underwriters rely most to set the preliminary offer price.11 As discussed in Section 4(ii), we use both quantitative and qualitative analysis.
(a) Quantitative Analysis
Table 6 presents results of our quantitative analysis for DFCF, DDM and the most commonly used multiple approaches: P/E and P/CF based on a peer group or stock market, calculated for year 0 and year +1.12 The median pricing error for DFCF is +14.1%: lead underwriters set the preliminary offer price significantly lower than the value estimates based on DFCF (p-value of the Wilcoxon signed rank test is less than 0.001). The median error for DDM is +3.4% and only significant at the 10% level (p-value = 0.094). The median errors for estimates based on multiples vary widely: they range from −9.8% to +21.7%. These results would suggest that the lead underwriters rely primarily on DDM to determine the preliminary offer price.
Pre-IPO Value Estimates and Preliminary Offer Price: A Comparison of Different Valuation Methods
|Median||14.1%|| 3.4%||−1.0% ||11.2%||−7.2% || 7.6%||21.7%||20.1%||−9.8% || 7.7%|
|Mean||14.4%|| 3.2%||−25.8% || 7.4%||−16.8% || 9.5%||24.9%||24.1%||−3.8% || 5.8%|
|Standard deviation||13.2%|| 8.7%||121.5% ||14.1%||30.8%||23.5%||34.1%||31.1%||21.6%||14.4%|
|Interquartile range||15.0%||12.0%||34.0%||16.8%||26.8%|| 6.6%||32.4%||22.4%||18.5%||13.4%|
|Percentage within 15%||55.1%||91.7%||42.9%||63.0%||50.0%||66.7%||33.3%||25.0%||62.5%||50.0%|
|Mean absolute error||15.8%||7.9%||47.0%||13.9%||23.2%||16.6%||30.4%||29.4%||17.6%||13.5%|
|Number of observations||49||24||21||27||12||12||12||12||8||6|
We measure the degree of central tendency of value estimates towards the preliminary offer price by the percentage of differences within 15% and the mean absolute error. For only 27 out of 49 IPOs (55.1% of the sample), the estimates based on DFCF are within 15% of the preliminary offer price. On the other hand, the DDM estimate is within 15% of the preliminary offer price for 22 out of 24 IPOs (91.7%) for which a DDM value is estimated. For the multiples estimates, the percentages within 15% are also much lower than for DDM. A comparison of the mean absolute error for the different valuation methods leads to the same conclusions as the comparison based on the percentage within 15%. A t-test of differences in the mean absolute error reveals that the mean absolute error for DDM estimates is significantly smaller than the mean absolute error of DFCF estimates (p-value = 0.002), and smaller than the mean absolute error for P/E peer group estimates in year +1 (p-value = 0.003), the most frequently used multiple estimation method. The mean absolute error for DFCF and P/E peer group (year +1) estimates are not significantly different (p-value = 0.40). Again these results seem to suggest that the preliminary offer price is driven by DDM if applied.
Another interesting finding in Table 6 is that the multiples valuations for year +1 are consistently closer to the preliminary offer price than the multiples valuations for the IPO year 0: underwriters seem to rely more on forecasted future multiples than on current multiples. The p-value of a t-test of differences in the mean absolute error between P/E peer group (year 0) and P/E peer group (year +1) is 0.002.
Some valuation methods will be more appropriate to use than others. Individual valuation models may have strengths and weaknesses that are complementary. Alternatively, the usefulness of a given valuation model may vary according to certain exogenous factors (Barker, 1999b). The underwriter has to choose which methods are appropriate and which are not. The results in Table 6 may be influenced by this choice. For example, the difference between DFCF and DDM might be caused by the 15 IPOs for which DFCF was used but DDM was not. For the 24 IPOs for which both methods were used, the DFCF estimates might then be much closer to the preliminary offer price than the results in Table 6 suggest. We therefore also investigate the relation between the IPO preliminary offer price and different value estimates by a pairwise comparison of valuation methods, for those IPOs that are valued with both methods. The results are presented in Table 7. DFCF, DDM and P/E based on a peer group for year +1 are compared pairwise. We concentrate on the P/E based on a peer group for year +1 because our performance measures indicate that this multiple is the one closest to the preliminary offer price. Moreover, P/E based on a peer group is the most commonly used multiple. We also compare P/E based on a peer group in year 0 and year +1. All results in Table 7 confirm those in Table 6.
Pre-IPO Value Estimates and Preliminary Offer Price: A Pairwise Comparison of Different Valuation Methods
|Median||13.8%|| 3.4%||15.6%||11.2%|| 7.7%||11.2%|| −5.8%|| 8.4%|
|Mean||12.9%|| 3.2%||15.0%|| 7.4%|| 4.8%|| 7.4%||−38.4%|| 5.7%|
|Standard deviation|| 9.3%|| 8.7%||11.6%||14.1%|| 9.2%||14.1%|| 131.7%||13.3%|
|Interquartile range||12.0%||12.0%||18.5%||16.8%|| 9.8%||13.6%|| 41.3%||13.5%|
|Percentage within 15%||62.5%||91.7%||48.1%||63.0%||84.6%||61.5%|| 41.2%||70.6%|
|Mean absolute error||13.7%|| 7.9%||15.9%||13.9%|| 9.1%||13.9%|| 51.6%||12.6%|
|Number of observations||24||24||27||27||13||13||17||17|
So far, we have relied on univariate analysis to investigate the relation between pre-IPO value estimates and preliminary offer price. For the IPOs in our sample, underwriters always use at least two valuation approaches. It therefore seems likely that the preliminary offer price is based on more than one value estimate. Table 8 reports the results of OLS-regressions of the natural logarithm of the preliminary offer price on the natural logarithm of value estimates by the lead underwriter. All variables are expressed as a natural logarithm because of large size differences. The coefficients of the explanatory variables should provide an estimate of the weight underwriters attach to a particular valuation method. However, the results in Table 8 have to be interpreted with caution: the number of observations for each combination of valuation methods is limited, and the correlations between the pre-IPO value estimates of the valuation methods are very high.
Relation of the Preliminary Offer Price to Pre-IPO Value Estimates
|Ln (DFCF estimate)||0.09||0.38||0.62||–|
|Ln (DDM estimate)||0.68||0.66||–||–|
|(3.15)***||(3.82)***|| || |
|Ln (P/E Peer Group Year +1 estimate)||0.31||–||0.38||0.82|
|Ln (P/E Peer Group Year 0 estimate)||–||–||–||0.03|
| || || ||(0.59)|
|Number of observations:||13||24||27||17|
The first column of Table 8 reports results for 13 IPOs which were valued with DFCF, DDM and P/E Peer Group Year +1 (the most commonly used multiple). The DDM coefficient is 0.68 and significant at the 1% level; the P/E Peer Group Year +1 is 0.31 and significant at the 5% level; the DFCF-coefficient is not significant. This again suggests that the preliminary offer price of IPOs which are valued with DFCF, DDM and P/E Peer Group Year +1, is primarily determined by DDM, and to a lesser extent by P/E Peer Group Year +1.
Column (2) of Table 8 reports regression results for 24 IPOs which were valued with DFCF and DDM. The preliminary offer price of these IPOs seems to be primarily determined by DDM, and to a lesser extent by DFCF. For 27 IPOs which were valued with DFCF and P/E Peer Group Year +1 (see column (3)), the underwriter seems to rely primarily on DFCF, but also on P/E Peer Group Year +1. The results in column (4), based on 17 IPOs, suggest that when P/E Peer Group year 0 and year +1 are used, underwriters rely only on P/E Peer Group in year +1.13 Taken together, the regression results confirm those of the univariate analysis: the preliminary offer price seems to be primarily determined by DDM, and underwriters rely more on forecasted multiples than on current multiples.
(b) Qualitative Analysis
In order to gain further insight into the way underwriters value and price IPOs, a qualitative analysis was carried out, by means of interviews with investment bankers, who were the lead underwriter of 45 of the 49 IPOs in our sample. They provided a consistent story on how Belgian IPOs are valued and priced. All seven investment bankers consider DFCF the most important valuation method. DDM is regarded to be too ‘conservative’ and is less relied on. One banker described DDM-valuation as merely a control check on DFCF-valuation. Multiples, on the other hand, are driven by market sentiment and may provide valuations that are ‘too high’ in a hot market (see for instance Fernandez, 2001). Glaum and Friedrich (2006) report that during the high market valuation period at the end of the 1990s analysts had to use comparative valuation or even come up with new metrics like subscriber multiples for start-ups or high-tech stocks, to produce company values that were in line with concurrent stock prices.14 An offer price that is too high may severely damage the reputation of the underwriter.
According to the investment bankers, the preliminary offer price and the offer price range are primarily determined by DFCF estimates, with DDM providing lower value estimates and multiples providing higher value estimates. The preliminary offer price is not actually based on a weighted sum of value estimates, but is set more loosely, taking into account the available value estimates (especially DFCF), and then applying a price discount. The discount is required in order to ensure that the stock offers a good opportunity to investors and has upside potential, and that the issue will be oversubscribed. This guarantees a good deal to all parties involved in the transaction and ensures sufficient liquidity for the stock when trading begins. The final offer price is the result of a negotiation between the issuer and the underwriter, and takes into account relevant market circumstances.
This implies that our finding in the quantitative analysis that the preliminary offer price is closest to DDM is not a consequence of underwriters primarily relying on DDM, but rather coincidentally as it tends to produce low valuations. The preliminary offer price is closest to DDM because underwriters rely on valuation methods that produce higher value estimates, and apply a discount to these valuations.
The substantial difference between valuation according to DFCF and DDM is remarkable, as both valuation methods should yield the same value if consistent assumptions are made. Nevertheless, it is well recognized that dividend discounting techniques are subject to some practical issues. In Penman and Sougiannis (1998), DDM produces considerably lower valuation results than DFCF. Francis, Olsson and Oswald (2000) find that all their models underestimate value, but DDM to the largest extent. They state that models which should give the same results in theory, differ in practice due to inconsistencies in forecasted attributes, growth rates and discount rates (e.g., clean surplus relationship violation, non-constant growth rates, discount rates inconsistent with no arbitrage principle, …). Lundholm and O'Keefe (2001) relate differences in the models to inconsistent forecast, inconsistent discount rate and missing cash flow errors. Damodaran (1994) explains the undervaluation of DDM because most practical valuation models do not allow (in an appropriate way) for the build-up in cash when firms pay out less than they are able to, which is often the case, and the use of incoherent assumptions. He claims that in reality, DFCF values often exceed DDM values.
Our interviews with the investment banks involved in the IPOs further shed some light on the inconsistencies between the application of DFCF and DDM, and provide an explanation why the latter often results in lower valuations. The investment banks are aware that DDM valuations in the prospectuses tend to be lower than the DFCF valuations. One explanation they offer is that the DDM valuations do not take into account the value of non-operating assets (such as excess cash). Moreover, many Belgian companies pay out only a fraction of their free cash flow as a dividend.15 As a result, most of these companies are underlevered and overcapitalized. By consequence, if DDM is based on actual dividends paid out, it will on average produce lower valuations than DFCF, which assumes that all free cash flows are returned to the investors when they are available.
Another explanation put forward by the investment banks is that companies typically go public when they need funds for large investments. This results at first in low dividend payouts, but leads to a high dividend growth later on. Often, the terminal (or continuing) value estimates used in the DDM model are too pessimistic, as the payout is not set at a level reflecting that the current high growth will slow down, and that less financial resources will be needed for new investments. In other words, this is an example of an inconsistent adjustment of the payout ratio and the growth rate.
Most investment banks prefer the DFCF method as it considers the overall picture. The DDM, on the other hand, produces an all-in-one value for the shareholder. Although the investment banks are aware of most of the above inconsistencies, they do not really attempt to reconcile both methods, since the majority of investors attach little importance to the DDM-outcome; most only care about DFCF and multiples. Block (1999) argues that a firm's dividend policy (and thus the DDM) is relatively unimportant in a financial analyst's analytical process, due to the irrelevance of dividends theory of Modigliani and Miller (1961), the fact that dividends do not count for much in an analyst's mind when annual returns are 20–30% as in the late nineties, and the desire of corporations to buy back shares rather than increase cash dividends.
(iii) The Performance of Valuation Models
In this section, we investigate the performance of the valuation models. Section 5(iii)(a) investigates the bias and accuracy of valuation models. In Section 5(iii)(b), we investigate whether the final offer price is closer to the stock market price than the pre-IPO value estimates.
(a) The Bias and Accuracy of Valuation Models
Results on the bias and accuracy of the valuation methods are presented in Table 9. Results based on the stock price on post-IPO days +1, +10, +20 and +30 are very similar and are therefore not reported in the paper. DFCF seems to be an unbiased value estimator: the median valuation error is only 4.6% and not significant (Wilcoxon signed rank test p= 0.681). DDM on the other hand produces biased estimates of value: the median valuation error is −10.6% and is significant (p= 0.006). If the stock market prices IPOs correctly, then DDM tends to underestimate value. For most multiple valuation methods, we also find a negative median valuation error, but this error is significant at the 10% level or higher for only three methods: P/E Peer Group (year 0) (p-value = 0.082), P/E Stock Market (year 0) (p-value = 0.009) and P/CF Stock Market (year +1) (p-value = 0.031). For P/E Peer Group (year +1) and P/CF Peer Group (year +1) the median valuation error is not significantly different from zero, which suggests that these multiples provide unbiased value estimates. Of course, for most multiples the sample is very small, which affects the quality of statistical testing.
Pre-IPO Value Estimates and Stock Price: A Comparison of Different Valuation Methods
|Median|| 4.6%||−10.6% ||−9.3% ||−4.3% ||−23.7% ||−5.0% || 4.6%||−1.3% ||−27.3% ||−20.4% |
|Mean|| 0.0%||−16.0% ||−45.0% ||−11.9% ||−33.4% ||−11.4% ||14.7%|| 8.0%||−24.9% ||−20.5% |
|Standard deviation||30.5%||28.6%||123.5% ||30.7%||34.8%||34.8%||36.1%||32.9%||30.8%|| 9.6%|
|Percentage within 15%||46.9%||50.0%||38.1%||48.2%||33.3%||41.7%||50.0%||66.7%|| 0.0%||33.3%|
|Mean absolute error||21.6%||22.3%||58.1%||23.9%||36.7%||27.1%||25.4%||21.2%||35.9%||20.5%|
|Number of observations||49||24||21||27||12||12||12||12||8||6|
Again, we use the percentage within 15% and the mean absolute error to measure the accuracy of the valuation methods. The results suggest that DFCF, DDM and the most commonly used multiples (P/E Peer Group year 0 and year +1) have similar accuracy. For each of these methods, about half of the valuations is within 15% of the average stock price in the first month of listing. The mean absolute errors are also very similar: they are not significantly different from each other, except the mean absolute error of P/E Peer Group year 0, which is significantly larger than DFCF mean absolute error (p-value = 0.043).
It is interesting to compare our results on value accuracy with the valuation accuracies obtained by Kim and Ritter (1999), who investigate the value accuracy of multiples using a sample of 190 US IPOs from 1992 to 1993. They use recent IPOs as comparables, which are chosen with a mechanical algorithm, and comparables chosen by a firm specializing in IPO research. Comparing IPO multiples to the median comparables multiple and a predicted multiple using regressions, they find much lower valuation accuracy than we do. It may seem surprising that objective valuations by academics are less accurate than valuations by lead underwriters. As we have noted in the introduction, value estimates by lead underwriters may be more accurate because they often have better access to information that is useful for valuation. Moreover, the lead underwriter may be affected by the market mood at the time of valuation. On the other hand, the post-IPO stock price may be affected by the valuation of the lead underwriter, if the market is willing to be guided by the valuations made at the pre-IPO stage. In that case, the valuation influences the market, rather than the other way around. The Belgian IPOs in our sample are also often mature firms that are comparatively easy to value. Finally, it can be expected that lead underwriters choose to report only valuation results that are appropriate for the type of firm that needs to be valued, while academics report all estimates of the valuation method(s) they investigate.
When we compare the valuation accuracy of the different multiples approaches in Table 9, it is striking that the valuations based on the forecasted earnings and cash flows in year +1 are consistently less biased and more accurate than the valuations based on the forecasted current year's earnings and cash flows: this result holds for both the P/E and the P/CF peer group multiples, and for the P/E and the P/CF stock market multiples, for all measures of valuation accuracy.16
Another interesting result is the quite good performance of the P/CF Peer Group (year +1) multiple: compared to the other valuation methods it has the lowest median valuation error and the highest percentage within 15%. A possible explanation for this result might be that P/CF is mainly used by underwriters to value firms which are relatively easy to value: older and larger firms with low growth prospects and a high dividend payout (cf. Table 5).
Again, we make a pairwise comparison of valuation methods: DFCF, DDM and P/E based on a peer group for year +1 are mutually compared, as well as P/E based on a peer group in year 0 and year +1. The results, which are presented in Table 10, confirm those of Table 9.17
Pre-IPO Value Estimates and Stock Price: A Pairwise Comparison of Different Valuation Methods
|Median|| 0.3%||−10.6% || 4.6%||−4.3% ||−16.1% ||−4.3% ||−18.5% ||−4.3% |
|Mean||−6.3% ||−16.0% ||−4.3% ||−11.9% ||−23.6% ||−21.0% ||−59.4% ||−15.3% |
|Percentage within 15%||50.0%||50.0%||51.9%||48.2%||46.1%||46.1%||35.3%||52.9%|
|Mean absolute error||21.5%||22.3%||20.4%||23.9%||26.3%||32.2%||66.8%||24.1%|
|Number of observations||24||24||27||27||13||13||17||17|
Given the dissimilar characteristics of firms that seek listing on the First and New Market, there might be a divergence in the frequency of use and performance of alternative valuation models across both groups of firms. Panel A of Table 11 reports the valuation methods used to value IPOs on the First and New Market. There are indeed some differences between both markets. The median IPO on the First Market is valued with four methods while the median IPO on the New Market is valued with DFCF and only one other method, which is typically P/E Peer Group (year +1). Thus, underwriters use primarily valuation models based on forecasted amounts of cash flows or earnings for valuing New Market IPOs. Yet, it is more difficult to make forecasts for these young high growth technology firms. This finding reveals a degree of sophistication in the valuation model choice of underwriters. A possible explanation is that New Market IPOs do not have sustainable current earnings or cash flows to be used as measures of future value. Table 11 also confirms our finding in Table 5 that young high growth firms, which are generally introduced on the New Market, are not valued with DDM or P/CF.
First Market versus New Market
|Discounted Free Cash Flow||34||15 |
|Dividend Discount Model||23||1|
|P/E Peer Group (year 0)||19||2|
|P/E Peer Group (year +1)||20||7|
|P/E Stock Market (year 0)||11||1|
|P/E Stock Market (year +1)||11||1|
|P/CF Peer Group (year 0)||12||0|
|P/CF Peer Group (year +1)||12||0|
|P/CF Stock Market (year 0)|| 8||0|
|P/CF Stock Market (year +1)|| 6||0|
|Median number of methods used|| 4||2|
|Median|| 0.98%||15.72%||−5.42% ||13.88%|
|Mean||−6.28% ||15.56%||−16.39% || 1.02%|
|Percentage within 15%||50.00%||40.00%||50.00%||42.86%|
|Mean absolute error||19.92%||25.39%||25.36%||19.68%|
Panel B reports valuation errors for the two methods which are frequently used for valuation on the New Market: DFCF and P/E Peer Group (year +1). For New Market IPOs, the median DFCF valuation error is quite high at 15.72%, while for First Market IPOs it is close to zero. The difference between the New and First Market is significant at the 5% level. This suggests that DFCF tends to overestimate the stock market value of IPOs on the New Market, which usually involve young and small high growth firms. On the other hand, we do not find a significant difference in the accuracy of DFCF measured by the mean absolute valuation error between the New and First Market. As for the P/E Peer Group (year +1) multiple, Panel B of Table 11 reports higher mean and median valuation errors for the New Market than for the First Market, but the differences are not statistically significant. The mean absolute error is also not significantly different between First and New Market IPOs.
(b) Pre-IPO Value Estimates and the Final Offer Price
During the public offering period, the underwriter obtains information about investor demand and updates the price. In addition, the bank may incorporate any changes in general market conditions, industry or firm-specific outlooks. The final offer price should therefore be a more accurate predictor of the stock price than the estimates of individual valuation methods. To test whether this is indeed the case, we compare the relation between the final offer price and the stock price on the one hand, to the relation between the estimated value and the average stock price in the first month of listing on the other hand. For each valuation approach, the offer price should be closer to the stock price than the value estimate if the lead underwriter also uses other valuable information to determine the price at which the shares will be offered. Results for the most commonly used valuation methods are presented in Table 12. The number of observations for each valuation method depends on the number of IPOs which were initially offered within a price range (not at a unique offer price), and which were valued with this estimation method.
Pre-IPO Value Estimates, Stock Price, and Final Offer Price: A Comparison of Different Valuation Methods
|Ln (estimated value/average stock price in the first month of listing)|
|Percentage within 15%||45.4%||47.6%||38.9%||50.0%|
|Mean absolute error||22.1%||23.4%||34.4%||25.4%|
|Ln (offer price/average stock price in the first month of listing)|
|Percentage within 15%||57.6%||66.7%||61.1%||55.0%|
|Mean absolute error||21.3%||20.4%||20.9%||22.1%|
|Number of observations||33||21||18||20|
For the 33 firms that are valued using DFCF, 15 firms (45.4%) have a DFCF estimated value within 15% of the stock price, while 19 firms (57.6%) have an offer price within 15% of the stock price. For the 21 firms that are valued using DDM, 10 firms (47.6%) have a DDM estimated value within 15% of the stock price, while 14 firms (66.7%) have an offer price within 15% of the stock price. A comparison of mean absolute errors also suggests that the offer price is closer to the stock price than the DFCF and/or DDM value estimates. However, the differences in the mean absolute errors are never statistically significant. Finally, the results for the most commonly used multiple estimates indicate that the offer price is closer to the stock price than the multiple estimates, but again, the differences in the mean absolute errors are not statistically significant.18
6. VALUATION AND UNDERPRICING
Probably the most often researched anomaly in the IPO market concerns underpricing and high average initial returns. These are found over a wide range of time periods and countries. (see e.g., Ibbotson and Ritter, 1995; and Ritter, 2003, for an overview of the evidence). Ritter and Welch (2002) find an average initial return on the first trading day of 18.8% in their sample of US firms going public from 1980 to 2001. In our sample, the initial return averaged 19.3%, which is a strong indication of underpricing.
In the context of our paper, underpricing could be consistent with three different explanations. First, underwriters may rely on DDM because they believe that DDM produces the most accurate value estimates, not realizing that DDM tends to underestimate value. Alternatively, underwriters may consciously underprice the IPO, by relying on DDM. Third, underwriters may deliberately underprice the IPO by applying a discount to other value estimates (such as DFCF), or to a weighted average of valuations. Given the results in Section 5(ii) (preliminary offer price is on average closest to DDM estimates, but underwriters rely on DFCF estimates applying a discount to it) and the results in Section 5(iii) (DDM tends to underestimate value while DFCF is an unbiased value estimator), the third explanation should be retained, in that underwriters consciously underprice the IPO by applying a discount to the DFCF estimates.
The hypothesis that underwriters consciously underprice IPOs is further supported by the fact that for 21 out of 24 IPOs for which both DDM and DFCF were used, the DDM valuation, which is on average closest to the preliminary offer price, was lower than the DFCF valuation. For only one IPO, DDM and DFCF lead to the same value estimate.
Another indication that underwriters consciously underprice the IPO is that for 14 of the 49 IPOs, the preliminary offer price was set lower than all value estimates published in the prospectus. An example is the IPO of Real Software, a Belgian software company: DFCF, minimum multiple and maximum multiple estimates were respectively 21.8%, 48.1% and 50.0% higher than the preliminary offer price. The average stock price of Real Software in the first month of listing was 64.6% higher than the preliminary offer price.
While literature on IPOs is abundant and several studies investigate the accuracy of valuation approaches, very few studies have investigated the valuation of IPOs by underwriting investment banks. We investigate the valuation by the lead underwriters of 49 IPOs on Euronext Brussels in the 1993–2001 period. We find that the lead underwriter always uses several valuation approaches, of which DFCF is the most popular. This is in sharp contrast to the standard literature on valuation model choice, which typically finds that multiples are much more often used than present value techniques. Our analyses show that DDM tends to underestimate value, while DFCF produces unbiased value estimates. However, DDM, DFCF and the most commonly used multiples have similar accuracy. Interviews with investment bankers suggest that underwriters consciously underprice IPOs, by applying a deliberate discount to DFCF value estimates. DFCF is considered to be the most reliable method. DDM estimates are on average closer to the preliminary offer price than other value estimates, because DDM tends to underestimate value.
When multiples valuation is used, investment banks rely mostly on forecasted future earnings and cash flows. We find that multiples valuation based on post-IPO forecasted earnings and cash flows indeed leads to more accurate valuations than multiples valuation based on earnings and cash flows in the IPO-year. Our results also indicate that the IPO final offer price is closer to the post-IPO stock price than pre-IPO value estimates, which is consistent with the lead underwriter using not only value estimates but also other valuable information to set the final offer price.
Our findings offer some important insights into the valuation of IPOs and the accuracy of valuation methods. The results suggest that discounted cash flow models are not superior (or inferior) to multiple valuation models when applied in the real world. However, we do find DDM has a tendency to underestimate value. Yet, DFCF and even DDM models are commonly used by investment bankers, which seem to have more trust in DFCF than in multiples or DDM. This paper also confirms that underwriters consciously underprice IPOs by applying a discount to their estimated value.
We identify various directions for future research. It would be interesting to examine what valuation methods investment banks use and how accurate underwriters are in predicting the stock market prices in countries such as the US, the UK and Japan. Another attractive research avenue would be to investigate how IPO firm or deal specific variables affect valuation accuracy. For instance, one could study whether larger venture capital backed firms taken public by a reputable underwriter are valued with greater accuracy. Such research would provide further insights in the motives of IPO underwriters and in the characteristics of IPOs. Finally, studies about both the use and accuracy of valuation methods should be conducted in related contexts like private equity investments and M&A.
A number of papers investigate determinants of IPO valuation, but do not consider valuation accuracy (e.g. Krinsky and Rotenberg, 1989; Clarkson, Dontoh, Richardson and Sefcik, 1992; McGuinness, 1993; Klein, 1996; Roosenboom and Van der Goot, 2003).
Another difference between our study and Roosenboom (2007) is that our analysis is based on information included in the IPO-prospectus, which is reviewed by and requires approval from a regulatory agency (in Belgium: the Banking, Finance and Insurance Commission) before it can be published, while the analysis of Roosenboom is based on pre-IPO underwriter analyst reports.
Thirty six out of 49 IPOs in our sample made use of bookbuilding, the remaining IPOs involved fixed price mechanisms.
Stock prices were taken from Datastream. Stock prices for three IPOs were not reported in Datastream; they were manually collected from De Tijd, the main Belgian financial newspaper.
Not surprisingly, all firms introduced on the New Market, which aims at young high growth firms, offer new shares.
See e.g., Kaplan and Ruback (1995), Kim and Ritter (1998) and Gilson et al. (2000). Kaplan and Ruback also use the mean squared error as a measure of valuation accuracy. The mean squared error assumes that the ‘cost’ of valuation errors for the user of the valuation method, such as costs arising from mispricing, increase quadraticly, while the mean absolute error assumes that the cost increases are linear. We measured the mean squared error for all the analyses presented in this paper. The results (not reported) fully confirm those of the mean absolute error.
The objective of these interviews is explained at the end of this section.
Some IPO prospectuses mention the use of a valuation method for which no estimation result is given. These are not included in Table 4.
Fernandez (2001) discusses the valuation of internet provider Terra-Lycos in 2000 by a number of banks, which use weighted averages of a curious mix of multiples, based on e.g., GNP per capita, number of inhabitants, capitalization per subscriber, enterprise value per page view, and capitalization per page view.
One IPO was valued with P/E based on earnings in the previous year only.
The results presented in this section are qualitatively similar if we use the final offer price instead of the preliminary offer price.
These multiples are consistently closer to the preliminary offer price than the multiples for which no results are presented.
We did not estimate a separate regression for DDM and P/E Peer Group year +1, because the 13 IPOs which were valued with DDM and P/E Peer Group year +1, were also valued with DFCF: see column (1).
In contrast to our expectations and as already reported, the multiple valuations in our sample were generally lower than the DFCF valuations. Results for the subsample of IPOs that went public during the hot market in 1997–1999 are qualitatively similar.
In Belgium there is also a legal restriction that dividend payments cannot be higher than the net profit after taxes.
This result is further confirmed if we take into account valuations based on the earnings and cash flows in years –1 and +2. We did not include these valuations in Table 9 because they are based on a very limited number of observations (1 to 5 observations).
We also compared P/CF Peer Group (year +1) pairwise with DFCF, DDM and P/E Peer Group (year +1). The results (unreported but available from the authors upon request) also confirm those of Table 9.
P-values are 0.89 for DFCF, 0.69 for DDM, 0.18 for P/E Peer Group (year 0), and 0.70 for P/CF Peer Group (year +1).