- Tltaeoq and Evidence -

Transcription

1 Tijdschrift voor Econornie en Management Vol. XXXVIII, 4, 1993 Initial Public Offerings in Belgium - Tltaeoq and Evidence - by C.VAN HULLE:':, M.CASSELMAN:': and M.O.1MA.M:" I. INTRODUCTION New listings are a matter of long run survival for any stock market : it neutralizes the ongoing process of delisting through take overs, bankruptcies, going private decisions... One kind of new listings are initial public offerings (IPOs): privately held firms obtain a quotation on a stock exchange and sell shares to the public at large. Completing an IPO successfully is not easy as this operation is surrounded by conflicts of interests and uncertainty. The two most visible decisions concern the choice of introduction method and the choice of the price at which the shares are sold to the public. In this paper we consider these two decision problems from several perspectives. First we give an overview of theories and international empirical evidence. We confront this with the second perspective, i.e. the Belgian IPO experience over the period This confrontation allows us to characterize the Belgian IPO market and formulate an answer to the following two questions: - given the characteristics of the Belgian IPOs, has the choice of introduction method been appropriate? - have Beigian IPOs been correctly priced? The literature on IPOs is an interesting one : it is permeated with strong and persistent anomalies at odds with the predictions OS the capital asset pricing model (C.A.P.M.). In fact it is the area in which Departelnent Toegepaste Econornie, K.U.Leuven. The authors are thankful to Wim Cools and Veerle Van Langendonck for con~putational assistance. They also thank Mr.Maertens from the Brussels' Stock Exchange for providing data.

2 the notion of market efficiency has come most strongly and convincingly under attack. The earliest and most strongly documented anomaly concerns the abnormally large returns investors earn between subscription and the first trading day in the aftermarket. Not only are initial returns abnormally large but they also seem to be governed by time varying patterns. And finally, but not in the least, there is some evidence that in the long run, returns on IPO-shares are inexplicably low. Over the last fifteen years many theories explaining the large initial IPO returns have been developed. According to many of these models the average size of the initial returns is not at odds with efficiency: these returns arise because of a rational reaction of the market participants to the special circumstances characterizing IPOs iii the time span between siibscription a~id the beginning of altermarket trading. Up to now rational explanations of time patterns are less well developed ; these rationalizations typically are a direct side product of the initial return models. Also no convincing rationale has been found for long run IPO underperformance. However the evidence supporting the existence of this latter anomaly is weaker and less convincing. Nevertheless it has reopened the question about the rationality of pricing in the IPO market and revived interest in non rational explanations for the initial return anomaly. Consequently a study of initial returns should not be separated from an analysis of the long run performance. In this paper we consider all the above IPO anomalies and test to what extent they occur in the Belgian market. The outcome of these tests allows us to evaluate the rationality and the price setting of Belgian IPOs. The paper is organized as follows : Section 11 gives a brief overview of the most important introduction methods ancl of the methods used in Belgium ; Section 111 is concerned with the phenomenon of the large initial returns ; Section IV considers the pattern of successive "hot" and "cold" issue markets; Section V discusses long run IPO performance and finally Section VI offers some conclusions. II. INTRODUCTION METHODS Various institutional arrangements are used for marketing IPOs. The following three arc frequently employed around the world : an offer for sale at a fixed price, an offer for sale with tender. and a place-

3 ment. In a placement the investment banker invites select customers (usually institutional investoi-s) to participate in the issue so that the offering is not available to the general public. An offer for sale at fixed price is an offering available to the gcneral public whereby the issuing firm preannounces a price and the number of shares to be sold. Individual investors specify the number of shares to which they wish to subscribe. If the offer is oversubscribed rationing occurs. In the opposite case, part of the offered shares remain unsold. This method is obligatory for public issues in the U.S., and heavily uscd in a.0. the U.K., Germany, the Netherlands, Italy, Finland, Sweden, Singapore and to a lesser extend in France. In Belgium it is only used in a minority of cases : of the 31 IPOs in the period, only 9 issues came to market under a fixed price arrangement (see Appendix 1). The offer price is set in cooperation with one or more investment bankers in the framework of an underwriting contract. The underwriting contract specifies the services the investment banker provides to the issuing firm and typically includes either the "firm commitment" or the "best efforts" clause. Under the former the investment banker commits itself to purchase all unsold shares at a predetermined price (below the preannounced offer price); under the latter the investment banker only takes care of the practical realization of the offer (e.g. collect and process the bids of the investors,...). In an offer for sale by tender the public is invited to bid for the shares at any price over a stated minimum. This method is commonly used in a.0. France and Belgium (in 20 out of 31 IPO cases). In principle, in a tender, the issuing price is determined at the level where demand equals supply. However in Belgium tenders are often "dirty": the price is determined at a level where some oversubscription still remains. The services provided by the investment banker are similar to those mentioned for the fixed price arrangement. Two versions of the tender method are commonly employed in Belgium : the "classical" tender and the "direct" tender. Under the classical method the offer price is determined and shares assigned to investors a few days before trading on the exchange starts. The initial or first day return for subscribers is then equal to : (first day trading price - offer price)/ offcr price. Under the direct method however. the offer price to be paid by the investors is equal to the price on the first trading day.

4 Hence the initial return under this method is equal to : (second day trading price - first day trading price)/first day trading price.' Finally one also distinguishes between a primary offering, a secondary offering and a mixture of the two. With a primary offering the issuing firm receives the proceeds from the IPO as the company itself sells the publicly placed shares. With a secondary offering the initial owners pocket the proceeds as they sell off part of their ownership in the firm. In the U.S. most IPOs are primary offerings or a mixture (Prasad (1992)). In Belgium, most IPOs have either been secondaiy or a mixture (see Appendix) UNDERPRICING : WHY DOES IT OCCUR? Usually empirical studies on first day returns, do not report the first day returns themselves but the abnormal first day returns, i.e. the surplus of the first day IPO return over the return of a contemporaneous investment in the stock market index. These studies consistently find that, on average, these first day surplus returns are unusually large and positive. Table 1A shows that in past periods the nverage abnormal return (before transactions costs) of subscribers selling their allotted shares at the first opportunity on the exchange, could range between % in Malaysia to 4.2 % in France. These abnormal returns have been earned over a very short time interval (usually less than two weeks) so that if expressed on a yearly basis the surplus returns have even been much higher. Table 1B provides information for Belgian IPOs on the initial abnormal returns as well as a measure of the return performance over the first six months of trading. We come back to this table in more detail later on. For the present we only note that, by international standards, with an average of , the 31 Belgian IPOs of the period offcred a relatively low initial abnormal return on a~erage~.~! Nevertheless, even after transactions costs, subscribers still have received returns much higher than those normally earned in a one to two week time span on the exchange. Hence the following questions : are these high initial return caused by issuers deliberately determining the offer prices below the stocks' normal market valuation (i.e. issuers underprice)? And if so, why would issuers do this? Or, rather is optimistic investor behaviour responsible for pushing prices up too high in the aftermarket? As more and more reasons were discovered

5 why issuers would rationally wish to underprice, it was generally accepted, at least in the academic world, that high initial returns resulted from deliberate underpricing. This view was supported by the finding that in most markets, immediately after the beginning of normal trading, no systematic abnormal returns were realized anymore (i.e. returns as of trading day 2, 3 etc. exhibit all the return properties of an efficient market5). Hence the view that on the first normal trading day, the market simply adjusted the share price from the abnormally low offer price level to the correct value. More recently however one has discovered that if - instead of analyzing daily, weekly or monthly returns - a longer term view is taken (i.e. two or three year returns after introduction), the IPO shares may on average underperform the market. This finding obviously has reopened the discussion on he validity of rhe ra~ional explanations of the initial return phenomenon : the long run underperformance may be seen as evidence that optimistic investors drive up the price too high at the first normal trading day ; subsequently over a two to three year time period the bubble slowly unravels. Even if this latter explanation of the long run behaviour is correct, still the question remains why underpricing occurs: why wouldn't issuers - being aware that in the aftermarket optimistic investors consistently drive up the price - try to benefit from this optimism and increase the offer price to a level where, on average, no abnormal initial returns are left? Likewise thc persistence of patterns in initial returns cannot be discardcd by simply referring to investors' irrationality. Generally the alternative explanations for the high initial returns arc not mutually exclusive. For presentational purposes we have classified them into three broad groupings, viz. rational models based on information asymmetries, rational models based on litigation considerations and models implying some irrationality. When applying these insights to the IPO market in Belgium, it is important to bear in mind that all subsequent explanatioils prcsume fixed price offers! Although the "dirty" tender is the most common sales method in Belgium, these explailations remain of interest for the Belgian market. To see this we only have to bear in mmd that in a fixed price system the company lteeps the offer price more in its own hands whereas this is not or far less the casc in a tender. The models show that a fixed price system allows firms to use the offer price as a tool to achieve certain goals. However the models also show that the use of this tool is costly. Hence by confronting the

6 benefits and the costs of the fixed price system (as indicated by the models) with the company's choice of issue method, we can make inferences about what the main concerns of the issuer likely have been. A. Models based otz asytnrnetr-ic info~mation Most models explaining initial IPO returns fall into this class. Therefore we have subdivided these asymmetric information explanations into subclasses 1, 2 and Winner's curse as an explanation for high initial returns Winner's curse implies tl~at winning a contract, shares... in a bidding contest is bad news because it is highly probable that the winner has won for the wrong reasons. To see how this comes about in the IPO context, let us consider the model developed by Rock (1986). For a winner's curse model to work it is essential that at the time of subscription, (a) there is uncertainty about the way the market will price the issue in the aftermarket and (b) information about this pricing is unevenly distributed over investors. To illustrate the argument assume for a moment that no underpricing would occur so that on average the offer price would be equal to the price in the aftermarket. As in Rock, let us presume that at the time of subscription some better informed investors Itnow whether or not the aftermarket price will be above the (fixed) offer price. Hence the better informed place large quantity bids whenever the offer price is below the aftermarket price. When the offer price is higher, these latter investors place no bids. By contrast, the uninformed investors cannot distinguish between good and bad offers and hence always place the same bids. It follows that although good offers arc heavily oversubscribed and rationed (because of the large informed orders), the informed still receive a large share; of the bad issues however they receive nothing. Consequently, as the share of the issue received by the uninformed is the mirror image of the sharc received by the informed, the uninformed investors systematically end up with all of the bad offers and with only a small fraction of the good issues (i.e. the uninformed face a winner's curse). Hence, if there would be no underpricing, uninformed investors would loose money on average. Consequently to induce the uninformed to participate in the subscrip-

7 tion process, firms must underprice on average so as to compensate for the bias in the allocation. Hence if a winner's curse force is at work in the IPO market one should observe, not only underpricing, but also a positive relationship between initial returns and the degree of subscription (i.e. high subscription is good news). In an extension Rock, Beatty and Ritter (1986) also show that, on average, underpricing increases with increasing uncertainty about the price in the afterma~ket.~ The predictions of the winner's curse models have been tested in many studies. ICoh and Walter (1989) find for data from Singapore a positive relationship between the subscription rate and initial underpricing; similar results are found by Levis (1990) for the U.K., by Suchard and Woo (1993) for Australia and by Beatty and Ritter (1986) for the =.S.. Empirical data aiso consistently confirm the hypothesized relationship between initial returns and uncertainty about the aftermarkct price, e.g. for the U.S.: Logue (1973), Ritter ((1984a), (1987)), Ibbotson, Sindelar and Ritter ((1988), (1991)), Miller and Reilly (1987), Beatty and Ritter (19861, Muscarella and Vetsuypens (1989b); for the U.K.: Levis (1993); for Germany: Ljungqvist (1993), for Australia : Suchard and Woo (1993). It is not hard to see that a chance of rationing (as is clearly the case with a dirty tender) together with asymmetric information among investors, are sufficient conditions for a winner's curse effect. However, because of issue procedures, it has likely been less important in the Belgian IPO market as compared to e.g. the U.S. market. First, because in a (dirty) tender the offer price depends on the bids made by all investors, the offer price at least partially reveals the information of the best informed. Meilce it dissipatcs the advantage of superior information. Sccond, in case of oversubscription, allotmenr is not necessarily proportional to the submitted bid quantities and, most importantly, the allotment rule is not always announced beforehand. Hence. although an informed investor may have made a large bid, the cx post chosen allotment 1-ulc need not allocate him/ her many shares. In sum, Belgian IPO procedures reduce he benefits from superior information about the aftermarltet price. In fact this fits the goals of the issuers : Belgian IPO firms generally do not seek a clientele of investors who subscribe with the intention to pocket the initial return and sell out immediately in the aftermarket. Consequently, in Belgian data we would expect only a weak relationship between initial return and oversubscription ; similarly we would also

8 expect a weak relationship between initial return and riskiness of the issue. These expectations are confirmed by the available evidence. In an earlier study on Belgian IPOs (over the period and including 19 issues), Van Hulle and Vanthienen ((1989a), (1989b)) find a positive statistically significant correlation between oversubscription and initial abnormal return ; however the correlation between initial abnormal return and risk (as measured by the standard deviation of daily aftermarket stock returns) is not statistically significant. Rogiers and Manigart (1992) report in their study (over the period and including 28 IPOs) similar findings except that in their sample the relationship between risk and initial abnormal return is significant. Finally there is evidence that Belgian issuers have been successful in attracting their preferred clientele. Casselm2n (1993) reports on a survey involving 27 investment advisors and 108 individual investors, all of whom operate in the Belgian market. 82 % of the survey participants indicate that they subscribe with the intention of keeping the IPO-shares for at least a few months. Only 2 % indicates their sole interest is in pocketing the initial return. 2. Signalling as an explanation for high initial returns (i.e. the image problem) Basically the signalling models try to capture the idea that IPOfirms may improve their image among investors by allowing subscribers to pocket a high initial return. The models in the literature focus mainly on the improvement of the circumstances under which, at a later time additional shares can be issued, as the explanation why firms wish to improve their image. The cornerstone of these signalling models is the asymmetric information between insidersissuers and investors. In these models issuers use the degree of underpricing to signal their true quality to the market. They predict, somewhat surprisingly, that initial return and firm quality are positively related (i.e. the better the prospects of the firm the lower its offer price). This result follows from the fact that only companies with good prospects can over time recover the cost of setting the offer price below their true market value. In a typical signalling model (e.g. Welch (1989), Allen and Faulhaber (1989)) firins make a partial offering of stock through an IPO in phase 1. Issuing firms may have either good or bad future cash flow prospects. During this first time interval only firms know what type they are (good or bad).

9 Hence, unless the good firms send a signal to the market in which they distinguish themselves from the bad ones, all firms look alike to the investors. In phase 2 information about the true quality of the IPO firms is revealed to the market with positive probability. Finally in phase 3 more stock is sold by the IPO firms through "seasoned" offerings. %t IPO time good firms may use this structure of events and, through signalling, distinguish themselves from the bad firms. However a signal can only be effective if it is unprofitable for bad firms to mimic it. In the present setting the offer price provides the good firms with the signalling tool they need. To see this, let us denote by V, (V,) the true value of the good (bad) firms (i.e. V, >V,); g denotes the proportion of good firms in the total IPO population. Suppose for simplicity that there is no risk aversion, discounting, transactions costs, taxes etc. Then, if there is no signalling effort by the good firms at IPO-time, all firms sell their shares at a uniform price P = [(gxv, + (l-g)xv,], i.e. at this price investors, on average, make no abnormal gains nor losses. Also note that, as P is simply a weighted average of V, and V,, necessarily V, <P <V,. Let us suppose now that good firms try to distinguish theinselves by fixing their offer price P, below the value of the bad firms, i.e. P, <V,. Then a situation may materialize in which bad firms find it unprofitable to mimic this offer price while simultaneously good firms are better off selling their IPO shares at P, instead of at P. To see this first note that there is some benefit to bad firms in mimicking the good firms and sell their IPO issue at P,: by copying the low offer price policy the bad firms buy themselves a chance to be able to overprice their seasoned offerings during phase 3. However if the seasoned offerings are small andlor the chances are coilsiderable that during phase 2 the true nature of the bad firms is revealed, the latter are better off not mimicking the good firms. The good firms may recover the loss from selling their IPO shares at too low a price if the chances of their true nature being revealed in phase 2 are remote and if the seasoned offerings are large. Hence at IPO time a "separating" equilibrium may emerge whereby good firms offer their shares at a lower price than the bad firms. Note that in this equilibrium bad firms may only expect to sell their shares at V, instead of at P = [gxv, + (l-g)xv,] >V, since presently investors can distinguish them from good firms. Hence as bad firms offer their shares for (maximally) their fair value while good firms offer their shares below their fair value, one should observe, on average, underpricing

10 and high initial returns. It is also clear that good firms can easily afford to underprice if only a limited number of shares is floated during the IPO. Hence percentage ownership retained and degree of underpricing should be positively reiated. Grinblatt and Hwang (1989) include risk aversion in their signalling model and derive several interesting results on the interaction between underpricing, percentage ownership retained, firm value and riskiness of the firm's activities. Another extension is Chemmanur (1993) who considers the case where no separating equilibrium materializes. Underpricing may still occur and is driven by the good firms who wish to give investors an incentive to engage in information gathering so that the chances of firm quality revelation increase. The main implications of the signalling models can be summarized as follows : - the higher rhe inirial remrn the higher the iikeiihood that subsequently additional shares are sold to the market (either by insiders selling off more of their stock or by an increase in the firm's share capital); - the higher the initial return the higher the probability that a subsequent share offering is large; - firms with higher initial returns are more likely to issue additional equity quickly after the IPO. For the U.S., Welch (1989) finds evidence in line with the predictions of the signalling models. Similar results are found by Keloharju (1993) with Finnish data. However not all studies find support for the signalling models. For example in a sample of U.S. data, different from that of Welch (1989), Garfinkel (1993) finds no relationship between initial returns and the probability of a subsequent additional offer. Furthermore Jagadeesh, Weinstein and Welch (1991) correctly point out that the empirically observed relationships need not have anything to do with signalling. What could be at work instead is a "market feedback" mechanism: the aftermarket performance of the issue provides the issuer with feedback information about the market's valuation of the firm; a higher than expected aftermarket price, for example, conveys the information that the issuer estimated the value of the firm Iower than what the market considers appropriate ; this information may induce the issuer to raise more capital through an additional offering after the IPO. When we apply the insights from the signalling models to the Belgian IPO market, we are immediately faced with the following problem : why do most Belgian issuers' forego the possibility to use the

11 offer price as a signalling tool by opting for the (dirty) tender method? As the tender method reveals thc opinion of the market rather than the information of the issuers, this choice of issue mechanism implies that most issuers have been more concerned with the uncertainty of the market reaction than with the consequences of asymmetric information between issuers and investors. This observation, in combination with the logic of the signalling models imply that, for Belgian IPOs : - either the information asymmetry between issuers and investors has been relatively unimportant ; - andlor the issuers had no plans to tap the market again soon. Consequently we would expect Belgian IPOs to be well established older firms (i.e. with only limited asymmetric information bctween issuers and investors) engaging in few seasoned offerings afterwards. Furthermore as such firms would have no signalling needs (and hence no use to underprice for signalling purposes), we would expect relatively small initial returns9. The Belgian evidence is consistent with the preceding remarks. At IPO-time, the median age of the 31 firms in the sample is 19 years (see Appendix 1 for details). Clearly the typical Belgian IPO is no recent upstartlo. In addition, by the end of 1992 none of these companies tapped the market again with a seasoned offering. Also the fact that about half of the issues were pure secondary offerings (see Appendix 1) indicates that, at IPO time, many issuers did not have major investment plans for which they urgently needed additional capital ; hence no need for any signalling about the prospects of such plans. Also the relatively small initial returns of the Belgian IPOs fit in with the argument. So, oddly enough, by abstaining from tapping the market again, the behaviour of the Belgian IPO firms is consistent with the logic of the signalling models. However as one digs deeper into the data, fitting all of the details into the models is more difficult. In particular, after 1987 relatively more issuers used the fixed price system on the Belgian market (in fact, after 1987, in 5 out of 12 cases as compared to only 4 out of 19 cases in the period 83-87). Is this purely a matter of accident? If we speculate on the possibility that this may not be the case, this post 1987 behaviour represents an odd phenomenon for two reasons. First, similarly to the earlier IPOs, by the end of 1992, post 1957 IPOs have not returned to the market with a seasoned issue. Why then this increased preference for the fixed price method? Secondly, both the signal-

12 ling argument and the winner's curse argument indicate that the higher the proportion of fixed price issues, the higher the average initial returns should be. But Table 1B indicates just the opposite. Nevertheless it is possible that neither of these two observations are at odds with the theoretical models because applying the theoretical insights requires awareness of the "ceteris paribus" assumptions. The signalling models, for example presume that, given the knowledge of the IPO's type (good or bad), there is no uncertainty about the market's reception and hence no uncertainty whatsoever about the aftermarket price. Suppose now that, as is the case in reality, we also allow for the fact that the uncertainty about pricing of the shares in the aftermarket is larger for some firms than for others. It is clear that when this uncertainty is important, determining the offer price is a major problem, a problem iargely resoived by the (dirty) tender method. Conversely, suppose that if there is little uncertainty about the market's reception of the IPO, firms prefer the fixed price method even if they do not wish to signal with a view to later seasoned issues (e.g. because of the relative simplicity of the fixed price procedure and because of the fact that with limited uncertainty about the aftermarket price there is liltely to be little winner's curse effect anyway). If after the stock market crash of 1987 a larger portion of the IPO firms were easy to evaluate "sure thingu-companies (in fact of the 5 fixed price issues 4 were holding companies), it is possible to find simultaneously more fixed price offers and smaller average initial returns. In line with this argument we find that the average of the initial abnormal return of all the fixed price issues between 1984 and end of 1991, but exclusive of outlier Glaverbel, amounts to only 0.33 % as compared to a global average of , also exclusive of Claverbell'-l'. However as mentioned before, in view of the small sample these latter remarks are rather speculative. 3. Investment banking activity and underpricing An amalgam of models investigate the interaction of investment banking activity and the average levcl of initial abnormal returns. One of the earliest models is the one developed by Baron (1982). It contends that underpricing is due to the superior information of the investment banker that advises the issuing company. The model is based on the assumption that the aftermarket stock price of an

13 issue is affected by the size of the market demand and the investment banker's selling effort. This latter effort is assumed to be unobservable by the issuing firm. Before the IPO the investment banker also has superior information about the market demand for the issue. To induce the investment banker to reveal his superior information about market demand, the firm has to accept underpricing because underpricing reduces the required selling effort of the banker. The empirical results in Muscarella and Vetsuypens (1989) however refute this model. Another set of models considers the interaction between underwriter's reputation and underpricing. As a rule, these models predict a negative relationship between underwriter's reputation and initial returns. Contrary to the Baron model, investment banker activity is not the cause of underpricing ; rather the banker's actions are aimed at reducing it. The basic idea in each of these models is that the investment banker's reputation can be used either to diminish the earlier discussed winner's curse problem or to reduce the need for good firms to credibly signal their quality through costly underpricmg. As an example of how underwriter reputation may help reduce winner's curse effects, we consider the model of Carter and Manaster (1990). These authors assume that accurately predicting the aftermarket price is more difficult for some issues than for others. Issuers and investment bankers know what issues are a "sure thing" and for which ones predicting aftermarket price is difficult. As in the winner's curse models, when investors submit their quantity bids, some investors are better informed than others. It is assumed that some investment bankers (i.e. the more prestigious ones) have deve- Iqed a reputation of marketing only "sure thing" - IPOs. In these IPOs the investment bank sets the offer price close to the price it is convinced will prevail in the aftermarket. All investors know the reputation of the investment banker. Hence by simply observing this offer price, all investors become equally and perfectly informed about the aftermarket price. Hence there can be no winner's curse and therefore there is no need for underpricing. Of course in the case of risky IPOs this mechanism cannot be applied and the winner's curse effect remains. The resulting implication is that investment banker prestige and initial returns are negatively related. Alternatively, in Booth and Smith (1986) and Titman and Trueman (1986), investment bankers build a reputation for correctly valuing firms. Hence if investors know that the investment bankers can distinguish bet-

14 ween good and bad firms (see signalling models above) and set the offer price so that it reflects this information, there is no need anymore for good firms to underprice in order to distinguish themselves from bad ones : the "certification" of the underwriter suffices. If high quality investment bankers are better at assessing the true nature of IPO firms, good firms have more interest in employing high quality investment bankers. So good firms employ prestigious investment bankers, whereas bad firms are more likely to employ less reputed investment bankers. Taken together, the main testable implications of this class of reputation models are as follows : - IPOs handled by less established investment bankers tend to be more underpriced ; - the higher the degree cf "certificzticn" by the investment banker the less underpricing ; so firm commitment offers should on average offer less initial return than best efforts issues ; l3 - the higher the quality of the underwriter the smaller the riskiness of the IPO. For the U.S. Booth and Smith (1986), Carter and Manaster (1990), Balvers, McDonald and Miller (1988) find a significant negative relationship between underwriter reputation and initial return. Ritter (1987), Chalh and Peavy (1985) and Kumer and Tsetsekos (1993) also find that best efforts IPOs have significantly higher initial returns as compared to firm commitment offers. In Belgium, Rogiers and Manigart (1992) test the relationship between investment banking prestige and average initial returns for Belgian IPOs. They find no support for the reputation model14. This result is in line with our earlier inferences. For these models presume that the reputation of the investment banker either reduces the winner's curse effect or else reduces the cost of signalling. However, our preceding analysis indicates that, in Belgium, introduction methods have limited winner's curse problems and in addition issuers may not have had strong signalling needs. Finally a third type of model on the interaction between investment banking activity and underpricing focuses on the process of information gathering concerning the likely market reception of the issue. Although these models are very much tailored to the U.S. institutional setting, the rationale for certain regularities in the way U.S. investment bankers market issues is very insightful. An important phenomenon in the U.S. investment banking industry is that of

15 "presales": investment bankers contact regular customers before fixing the offer price and sound them out about their interest for the IPO ; at the time of the allotment of an oversubscribed issue, regular customers receive priority in the allocation and so earn even higher abacrma! retsrns. Benveniste and Spindt (1989), Spatt and Srivastava (1991) show that underpricing and "presales" go hand in hand. Thc reason why a system of presales implies underpricing is simple. In the presaies period the investment banker collects information about the likely aftermarket price with select customers. However for this information to be useful it is necessary that these investors correctly reveal how much they are prepared to pay for shares of the issue. Furthermore customers must also have an incentive to participate in this information gathering process. Hence the following procedure : only customers who communicate a price in the tapper range of all communicated prices get priority allocation ; the investment banker estimates the level of the aftermarket price on the basis of communicated prices and sets the offer price so that subscription leaves some positive abnormal returns to investors. The underpricing guarantees that select customers have an incentive to participate in the process. Furthermore, since the priority allocation is reserved to those comn~unicating the highest prices, this process also induces investors not to communicate a price (too much) below the price they are willing to pay. No direct empirical tests of these models have been completed so far. However some of these models' implications are in accordance with observed regularities. For example, Benveniste and Spindt show that best efforts IPOs should be more underpriced than firm commitment offers. This is in line with empirical findings reported earlier. When we compare the presales system with the dirty tender method in Belgium, we see a surprising similarity: the latter method is also a system in which (a) investors have an incentive to cooperate in the information gathering and (b) investors have an incentive to truthfully reveal the price they are prepared to pay. Just as in the U.S., investors are first solicited to submit price and quantity bids; participants in this bid process receive a priority allocation (i.e. investors who submit no bids, normally do not participate in the initial allotment of the issue). Also similarly to the presales method, only the bids in the higher price range ultimately receive a priority allocation. After observing the investors' bids, the investment banker

16 and issuer determine the offer price, so that, on average, rationing and some underpricing remains. Hence, ultimately the investment bank receives the information it seeks and the participants most truthfully revealing the price they are willing to pay are most likely to receive a share of an underpriced issue. In fact, the dirty tender method can be regarded as a simplified version of the U.S. marketing procedure. For in the U.S., the floating of the issue occurs in two stages : in the first (presales) stage some investors are approached and part of the issue is "sold" through priority allocation ; then in a second stage the remainder of the issue is floated through the (legally enforced) fixed price procedure. In comparison with this process the Belgian dirty tender method is a one stage "presales" procedure in which all investors can participate and through which the issue is fully sold. From this "presales" view of Belgian IPO procedures we would suspect a positive relation between initial abnormal returns and the percentage ownership initial owners retain : the less shares are floated, the smaller the cost of underpricing and hence the cost of remunerating investors for participating in the "presales". Furthermore, in practice, a sufficiently high initial return earns the issuer and the investment banker favourable comments on the success of the operation, so that the cost of underpricing may very well be compensated by the benefits of this free publicity in the financial press. This is our interpretation of the empirically observed positive relationship between retention rate and initial abnormal return as reported in Van Hulle and Vanthienen (1989) and Rogiers and Manigart (1992), albeit that in the former study the relationship is not statistically significant. The presales logic applies equally well to one stage (i.e. non U.S. style) fixed price issues, at least in case of "sure thing1'-ipos whereby the issuer is without signalling needs. However instead of predicting a sizeable initial return, here the presales argument would predict only a small initial return: with a fixed price "sure thingu- IPO the submission of bids process is used solely as a distributive mechanism by the issuer. Consequently, the issuer does not need much underpricing as an incentive for participants to reveal information ; a compensation for participation costs (e.g. transactions costs) suffices. If we combine this result with the predictions of the signalling models the following hypothesis ensues : in countries where issuers can freely choose between the fixed price and the tender method, the variation in initial abnormal returns of the fixed price

17 !POs should be much larger than the variation in the initial abnormal returns of the tenders. In fact, among the fixed price IPOs we expect two types of firms : firms with signalling needs and which therefore heavily underprice, and "sure thing" IPOs that hardly leave subscribers any money on the table. This discussion formalizes of our eariier speculative considerations on the use of the fixed price method in Belgium. B. Litigation corzsiderzlfions A number of models (e.g. Tinic (1988), Huges and Thakor (1992)) hypothesize that underpricing of IPOs is made deliberately to avoid lawsuits. In the U.S. these lawsuits may arise fro= the requirement of "due-diligence" to be shown by the investment bankers and the requirement that no information, important for valuation purposes, may be withheld. In the earlier model of Tinic the cost of underpricing is weighted against the lower expected costs from a lawsuit. Huges and Thakor add investment banker reputation to the setting and study the interaction between reputation, offer prices and the probability of a lawsuit. As the chances of a lawsuit decrease with investment banker reputation, Huges and Thakor predict (in accordance with the reputation models above) that the higher the underwriter's reputation, the smaller the underpricing and initial return will be. As lawsuits only occur if the aftermarket price is sufficiently bad in comparison with the offer price, the model also predicts (and this in accordance with the winner's curse models) that issuers underprice more as uncertainty about the aftermarket price increases. Earlier on we referred to empirical evidence confirming these relationships. However, direct tests of the litigation hypothesis show conflicting results. Tinic (1988) reports that in the US., after a regulatory change increased the likelihood of lawsuits, initial IPO returns went up on average. Drake and Vetsuypens (1993) investigate a sample of U.S. IPO firms that were subsequently sued. They find that the average court settlement in their sample represents about 15 % of the offering value. Drake and Vetsuypens point out that prior studies have documented that, in the U.S., average underpricing also roughly amounts to 15 % or more. Hence taking into account the fact that the chances of being sued is far less than 1, this

18 underpricing is too large to be explained as a mere insurance against possible future lawsuits. C. Other explanations Additional explanations presume some irrationality on the part of market participants. (For expositional purposes we distinguish between three classes of models) 1. Underwriter price support or pegging hypothesis According to Ruud (1991) an important factor in explaining the documented high initial retmns boils down to price support behaviour of the underwriter in the aftermarket. Ruud claims that issuers do not wish the price to fall below some threshold level in a first period after the IPO. Consequently the cross sectional distribution of initial returns should be skewed to the right: the investment bankers' actions allow the right tail of the distribution (i.e. good outcomes) to be observed but not the "true" left tail (i.e. bad aftermarket price drops are eliminated by price supporting activities). Another consequence of those price support activities is that, on average, the first day aftermarket price rises above the true market value of the stock. Hence as the price support is gradually withdrawn, realized returns during the first few weeks of trading should tend to be negative. Ruud finds for U.S. data that the cross sectional distribution of the initial returns indeed satisfies this predicted pattern. However the author does not report on the aftermarket performance of his IPO sample. Such information is given in Ljungqvist (1993) for Germany. Ljungqvist also finds a strongly skewed (to the right) distribution of initial returns ; however over the subsequent 20 first trading days no evidence of systematic negative abnormal returns is found. So, at least for Germany, the pegging hypothesis cannot explain the initial abnormal returns. Van Hulle and Vanthienen (1989) report average abnormal returns of the first 50 trading days of Belgian IPOs. They find some weak price pressure in the second and third day return; on other days no price pressure is observed. The price pressure of the first two days is not present anymore in the larger sample of Rogiers and Manigart (1992) nor in the sample of the present study1? Furthermore Table 1B shows that the returns over the first six months of

19 aftermarket trading above a comparable investment in the market (but excluding the IPO return of the first three days), reveal no sign of IPO underperformance during the early trading period. So, even if in Belgium price pegging has taken place, it has not resulted in a pattcrn of massive initial overpricing subsequently followed by a gradual price decrease as the support is withdrawnl6. 2. Impresario hypothesis Shiller (1990) contends that underpricing may be cxplained by the behaviour of underwritcrs. These underwriters (i.e. the impresarios) underprice IPOs because they wish to create an impression of excess demand and "waiting lines" for the investment products they market. As subscribers do not understand this game and as individual investors are influenced by what other investors seem to think, the bank uses this policy to create the impression that its investment products are considered attractive by many individuals. This reputation attracts additional investors thereby stimulating demand. Shil-!er argues that any given underwriter probably cannot expect to promote a fad, and certainly not repeatedly, but still may find it profitable to follow a standard policy of underpricing. This theory has not been tested directly as yet. However Shiller's U.S. survey data support the hypothesis that individual investors are influenced by the opinions of others : only 26 % of the respondents indicate they do fundamental analysis of the relationship between the offer price and the firm's intrinsic value. 3. Miller's uncertainty and divergence of opinion model In practice, uncertainty and risk imply divergence of opinion among investors about future expected returns. Miller (1977) argues that underwriters price new issues on the basis of their own best estimates and on the prices of comparable seasoned securities. The mean of their appraisals would resemble the mean appraisal of the average investor. However the investors likely to buy shares from a particular IPO are the ones who are most optimistic about it. These overoptimistic investors create the high initial abnormal returns as they bid up the prices in the aftermarket above the true values. Consequently, on average, as time goes by, these overoptimistic investors gradually are proved wrong. Hence, in the long run, share re-

20 turns of IPOs should be abnormally low. Because of this latter prediction, Miller's view has recently gained more attention as empirical evidence indicates that the long run performance of IPOs may indeed be bad (see Section V below). IV. "HOT" AND "COLD" ISSUE MARKETS Another persistent phenomenon is a wavelike pattern in both the number of issues and the size of the initial returns : first comes a period with high initial returns (i.e. the market is "hot"); it is followed by a period with unusually many IPOs and low initial returns (i.e. the "cold" market). The heavy volume period tends to lag the "hot" market with six to twelve months. In the U.S. Ibbotson and Jaffe (1975) and Ritter (1984a) identified hot issue markets at the end and the beginning of the sixties and in the early eighties. Suchard and Woo (1993) find a hot market in Australia between the end of 1986 and the stock market crash a year later. Uhler (1993) reports the existence of a hot issue market in Germany during and It is not well understood why such cycles occur and current explanations are only partial in nature. All current theories presume that the composition of firm types entering the IPO market, varies over time. Under this assumption, most of the underpricing theories discussed above would predict corresponding changes in the level of average initial returns. For example, the winner's curse model implies that, if for some (unexplained) reason the proportion of IPO firms for which information asymmetry among investors is small, increases, the winner's curse effect and hence required initial returns decrease. Also the signalling models produce cycles in initial returns if we assume that the number of good investment projects available to firms, varies over time. As an illustration, let us consider a simplified version of the scenario of Allen and Faulhaber (1989). Suppose that initially no surplus gains can be made in certain sectors of the economy. Hence few IPO firms expect additional financing needs in the future through a seasoned offering. Consequently they have no incentive to engage in costly signalling. Suppose then that a shock occurs (e.g. an economic upturn) so that at least some firms expect increased economic rents for some time in the future. Hence to meet the financing needs of additional investments, more IPO firms expect to engage subsequently in a seasoned offering. Therefore more

21 firms may find it worthwhile to signal at IPO time. To explain not only the patterns in initial returns but also the corresponding wavelike behaviour in the number of IPOs, additional assumptions on competitive conditions in industries are needed. Allen and Faulhaber (1989) construct a scenario that produces the hot-cold market pattern in both initial returns and IPO volume. A deeper discussion involves industrial competitive analysis and is outside the scope of this article. As in the U.S., Australia and Germany, Belgian average initial returns also have decreased after 1987 (see Table 1B). However our discussion in Section 111 indicates that this may mainly have been due to a more frequent use of the fixed price method after 87. V- THE LONG-RUN UNDERPERFORMANCE OF IPOS Recently Agganval and Rivoli (1990) and Ritter (1991) reopened the debate about the nature of the initial returns, i.e. whether or not the high initial returns are a consequence of rational and deliberate underpricing or a consequence of too much optimism by investors and hence initial overpricing in the aftermarket. Ritter (1991) reports that in the three years after going public, the IPO firms that entered the U.S. stock market in the period, significantly underperformed a set of comparable firms matched by size and industry with about 27 % (i.e. on average 27 % / 3 = 9 % per year). However the underperformance is concentrated among relatively young growth companies, especially those going public in the highvolume years of the early 80s. Ritter finds no long-run underperformance for more established IPO firms and for the light-volume years of the mid and late 1970s. Aggarwal and Rivoli (1990) compare the long run return on IPO shares with the market return and also report underperformance. Table 2A gives an overview of the findings of recent studies that compare the long run return on IPO shares with the market return. These figures concern solely the long run aftermarket performance in the sense that initial returns are excluded. Table 2A indicates that underperformance is sometimes, but not always present. In fact, only in relatively few cases is the underperformance statistically significant. When interpreting the results of Table 2A it should be borne in mind that the numbers only represent sample averages and hence obscure several observed patterns in long run IPO performance. One of these patterns is that

22 long run IPO performance varies widely through time. At least for the U.S., companies that went public in heavy volun~e years fare badly, while there is no poor aftermarket performance for IPO firms that went public in light volun~e years (Ritter (19911, Loughran and Ritter (1993), Aggarwal and Rivoli (1990)). However, in the U.K. this relationship bctween long run performance and IPO volu~iie does not seem to be present (Levis (1993)). Furthermore, consistent with Miller's model, some studies report a tendency 101- firms with high initial abnormal returns to have the worst long run performance (Ritter (1991), Loughran and Ritter (1993), Levis (1993)). And in view of the fact that Ritter reports the worst long run perforn~ance for young growth companies, it is not surprising that this author finds a strong positive relationship bctween firm age at IPO time and long run return (Ritter (1991)). Finally the interpretation of the results in Table 2A is hampered by measurement problems. Leleux (1993) correctly points out that the long run performance studies usually assume that the average sample beta coefficient of the IPOs is equal to 1. Under this assumption the capital asset pricing model (C.A.P.M.) implies that, on average, the long run return of IPOs should equal the market return over the corresponding period. Consequently to check whether or not IPO returns correspond with the risk return trade-off of the C.A.P.M., these latter studies subtract from the long run IPO returns the corresponding market returns and calculate the average of these differences. However, in his IPO sample, Leleux finds that IPO firms, on average, show low systematic risk (i.e. the average beta coefficient is below 1). Hence, the market return may not be the appropriate yardstick against which to evaluate long run IPO performance. Consequently the studies of Table 2A may wrongly indicate underperformance17. To evaluate the long run performance of Belgian IPOs we have proceeded in two steps: first we evaluate as in Table 2A; second, we check simultaneously for the possible existence of Leleux's "beta" bias and for the possible existence of patterns in long run IPO performance. The results of the performance evaluation of the first step are contained in Table 2B. It shows detailed information about the performance of Belgian IPOs during their first two years of trading. Reported long run returns exclude the returns of the first three days to ascertain that any abnormal price movement linked with the initial abnormal returns is excluded. The data in Table 2B also exclude the IPOs of 1991 because by the end of 1992 (the last data included

23 in our study), the IPOs of 1991 had been quoted for too short a period (see also Note 2). Next to a comparison with the return on the market index we add, as another test, a comparison with the return on a portfolio of matching firms, selected to match simultaileously as much as possible both for size (in terms of market capitalization) and sector. However in some cases finding a comparable firm has been impossible ; so we may expect the matching methodology to produce noisy performance figures that should be interpreted with care. According to Table 2B, the Belgian IPOs coming to the exchange before 1987 underperformed, on average, both the market and the portfolio of matching firms while those of the post 1957 period overperformed on average. But neither the average underperformance nor overperfcrmance is statistically significant, except possibly the average post 1987 long run IPO performance relative to the matching firm (i.e. significant at the 10% level)'! However, this latter ave- rage is only based on a very small subsample (i.e. 9 sample points). Also, in the present case, the statistical significance output as reported in Table 2B, should be considered with precaution. The applied tests (i.e. t-statistics) only produce meaningful results if the sample data are normally distributed. In the present sample there are too many outliers implying that the sample distribution shows "fat tails" in comparison to a normal distribution. Hence we need a transformation of the data to obtain a sample distribution that more closely approximates normality. It turns out that if we replace the two year abnormal returns by the equivalent one year abnormal returns (i.e. the equivalent one year abnormal return satisfies the equation: [l + equivalent one year abnormal returni2 = [l + two year abnormal return]), we get a sample of data approximating the normal distribution". As the sample of two year abnormal returns relative to a matching firm contains abnormal returns below 100 %, we only calculate transformed returns for the series of abnormal returns relative to the market (i.e. it is not possible to take the square root of a negative number). In terms of the one year equivalent abnormal returns, the underperformance of the pre-l987 IPOs is significant at the 10 % level". However, even after this transformation of variables, we still cannot trust the preceding statistical significance tests because beta bias may cause overestimation as well as underestimation of the performance. To test for beta bias and the existence of return patterns