UFDC Home  Search all Groups  UF Institutional Repository  UF Institutional Repository  UF Theses & Dissertations  Vendor Digitized Files   Help 
Material Information
Subjects
Notes
Record Information

Full Text 
THE ROLE OF ACCOUNTING NUMBERS IN THE "LONGTERM" MISPRICING OF INITIAL PUBLIC SECURITY OFFERINGS By JOHN D. NEILL, III A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNJI'.'ERSITY OF FLORIDA 1990 Copyright 1990 by John D. Neill, III ACKNOWLEDGEMENTS There are a number of people whom I wish to thank for their support and encouragement during the course of my graduate studies. First and foremost, I want to express thanks to my dissertation committee chairman Bipin Ajinkya for his advice in general and for his insightful comments on this dissertation in particular. I will always be appreciative of the tremendous amount of time he spent in directing this dissertation research. I also want to thank the remainder of my dissertation committee, David Brown, Miles Livingston, and Dan Smith, for their encouragement and helpful suggestions and comments throughout this project. I also wish to acknowledge Jay Ritter and Chris James for their contributions to this dissertation. The data that they supplied me with proved to be invaluable in this research. Finally, I want to especially thank my wife Karene for her unfailing love and support over the past five years of graduate studies. Without her continual encouragement, this dissertation might never have become a reality. iii TABLE OF CONTENTS page ACKNOWLEDGMENTS .............................................. iii ABSTRACT ......................................................... vi CHAPTERS 1 INTRODUCTION............................................ 1 2 PRIOR RESEARCH ON INITIAL PUBLIC OFFERINGS............. 9 Introduction ............................................ 9 ShortTerm Mispricing of IPOs........................... 9 LongTerm Mispricing of IPOs........................... 16 3 HYPOTHESIS DEVELOPMENT................................. 24 Introduction........................................... 24 Assumptions ............................................ 24 Decomposition of Systematic and Unsystematic Risk into Accounting and Business Risk Components......... 27 Hypotheses ............................................. 29 4 EMPIRICAL METHODOLOGY.................................. 33 Introduction........................................... 33 Three Measures of Accounting Risk....................... 33 Measurement of RiskAdjusted Abnormal Security Returns ............................ 42 Univariate Tests....................................... 47 Multivariate Tests..................................... 54 Summary of the Empirical Methodology.................... 65 5 DATA, SAMPLE SELECTION CRITERIA, AND RESULTS........... 67 Data Sources and Sample Selection Criteria............. 67 Descriptive Statistics................................. 69 Results of Univariate Tests............................ 77 Results of Multivariate Tests.......................... 87 6 CONCLUSIONS AND SUGGESTIONS FOR FUTURE RESEARCH........ 109 Conclusions............................................ 109 Suggestions for Future Research........................ ll APPENDICES A SHORTTERM UNDERPRICING AND ACCOUNTING RISK............ 113 B EMPIRICAL RESULTS BASED ON A FIVE YEAR TIME HORIZON.... 119 REFERENCES ................................................... 143 BIOGRAPHICAL SKETCH.......................................... .147 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy THE ROLE OF ACCOUNTING NUMBERS IN THE "LONGTERM" MISPRICING OF INITIAL PUBLIC SECURITY OFFERINGS By John D. Neill, III August, 1990 Chairman: Bipin B. Ajinkya Major Department: Fisher School of Accounting This dissertation examines the impact of the quality of accounting numbers supplied to potential investors on the "longterm" mispricing of initial public offerings (IPOs) of securities. Such mispricing is assessed in relation to the market's valuation after more information is revealed. Initial offerings are of interest from an accounting research perspective due to the lack of market information sources prior to the IPO. Three hypotheses relating accounting quality and longterm mispricing are developed and empirically tested. First, the lack of quality accounting numbers at the IPO date is predicted to have no effect on the mean abnormal returns earned by investors. However, the second hypothesis states that the variance of these abnormal returns is greater in instances where quality accounting information is lacking. Finally, the third hypothesis predicts that the influence of the quality of accounting numbers on abnormal return variances diminishes over time. Three empirical surrogates for accounting quality are employed. The first is the reputation of the auditor utilized in the IPO process. It is measured according to whether a Big Eight or a nonBig Eight auditing firm is employed. The second is the potential managerial manipulation of reported earnings as reflected in managers' accounting procedure choices, measured by whether the inventory and depreciation methods are income increasing or income decreasing. The third proxy is whether the SEC registration statement filed is a Form S1 or Form S18 registration. This third surrogate measures the quantity of accounting disclosures since Form S1 registrations are more extensive. The empirical results indicate that the three surrogates for accounting quality have no appreciable impact upon the mean abnormal returns of investors, as hypothesized. However, these same quality surrogates, with the exception of auditor reputation, influence the variance of abnormal returns, also as predicted. Thus, the results indicate that uncertainty relating to accounting quality translates into an increased variance of IPO abnormal returns, rather than a systematic overvaluation or undervaluation. As for the third hypothesis, the results do not reveal a diminution of the quality surrogates' ability to explain abnormal return variances as time progresses. CHAPTER 1 INTRODUCTION Initial public offerings (IPOs) of securities have been the subject of academic research for over two decades.' An aspect of the IPO market which makes it a particularly rich research area is the relative scarcity of publicly available information about these previously privately held firms.2 Specifically, potential IPO investors are at an informational disadvantage compared to those who invest in more established publicly traded firms, since there is no past series of security prices and only very limited financial statement information is available upon which to make determinations of IPO firm value.3 In addition, IPO firms are not subject to the continuing scrutiny of the financial analyst community. The motivation of this dissertation is to understand the degree to which accounting numbers influence IPO valuation. The preceding 1 See Smith [1986] and Ibbotson et al. [1988] for reviews of this literature. 2 Privately held firms which have publicly traded debt outstanding are notable exceptions. Publicly available financial statement information is produced by these firms. 3 The Securities and Exchange Commission (SEC) requires the inclusion of two (in a Form S18 registration) or three (in a Form S1 registration) years of audited financial statements in the registration statement and accompanying prospectus. However, Weiss [1988] documents that 529 out of her sample of 1510 IPOs reported less than two years of earnings in the prospectus. These noncomplying firms are presumably startup firms which have operating histories of less than two years. discussion indicates that accounting numbers potentially play a greater role in the valuation of IPOs than in the valuation of more established publicly traded firms, due to the reduced number of competing information sources about IPOs.4 Thus, initial public offerings represent an opportunity to explore the role of accounting numbers in security valuation in instances of considerable uncertainty surrounding the value of a firm. Specifically, this dissertation examines the impact of the quality of accounting numbers available to investors on the "longterm" mispricing of IPOs. In this thesis, accounting quality refers to both the reliability and the quantity of the accounting numbers supplied to investors. In other words, neither a sufficient quantity of unreliable accounting numbers nor an insufficient quantity of reliable accounting numbers are considered quality accounting disclosures in this dissertation. The discussion to follow describes the differences between "longterm" and "shortterm" IPO mispricing, and explains why the longer term perspective is adopted in this study. Bower states that "there are two types of mispricing inherent in initial offerings" (Bower [1989, p. 647]). The first type of mispricing 4 Prior research has consistently demonstrated that accounting numbers and security prices are statistically associated. Please refer to Beaver [1981, Chapter 5] and Watts and Zimmerman [1986, Chapter 3] for comprehensive reviews of this literature. However, changes in accounting numbers (i.e., "unexpected" earnings) typically only explain five to ten percent of the variation in security returns. This low level of explanatory power is presumably due to the abundance of competing sources of information (e.g., financial analysts, industry publications, the Wall Street Journal, etc.) about publicly traded firms. is the widely documented shortterm underpricingg" phenomenon. Underpricing refers to the empirical finding that, on average, the offering price set by the underwriter6 (hereafter referred to as the IPO price) is below the consensus price set by market participants shortly after the securities begin trading publicly. On average, this underpricing exceeds 15% (see Smith [1986]).7 Shortterm underpricing is typically calculated as the return that an investor could have earned by purchasing the IPO from the underwriter and selling at the market clearing price at the end of the first day of public trading.8 Thus, the shortterm mispricing research assesses the accuracy of underwriters' pricing decisions by comparing an underwriter's IPO price to the price which prevails at the close of the first day of public trading (hereafter referred to as the first aftermarket price). 5 Shortterm underpricing has been documented by Reilly and Hatfield [1969], McDonald and Fisher [1972], Logue [1973], Ibbotson [1975], Ibbotson and Jaffe [1975], Ritter [1984b, 1987], Beatty and Ritter [1986], Miller and Reilly [1987], Tinic [1988], Weiss [1988], and others. Reviews of these empirical findings are provided by Smith [1986] and Ibbotson et al. [1988]. 6 More precisely, the initial offering price is set by the underwriter after consultation with the entrepreneur of the IPO firm. Thus, the issuing firm may play a significant role in the setting of the IPO price. Alternatively, the issuing firm may completely delegate the pricing decision to the underwriter (as in Baron's [1982] model). 7 Proposed explanations of the shortterm underpricing phenomenon will be detailed in Chapter 2. 8 However, underpricing has also been calculated as the return over the first week (e.g., Tinic [1988]) or month (e.g., Ibbotson [1975]) of public trading. Such departures from the usual underpricing calculation are typically due to the unavailability of the market price at the end of the first trading day. Bower describes the second type of IPO mispricing as follows: The second type of mispricing is a longer term effect that relates to the lack of publicly available information about the firm. Firm value is not likely to be revealed immediately, but rather investors' expectations about firm value improve over a long period of time as more information is revealed about the firm. (Bower [1989, p. 650]) This second type of IPO mispricing "does not deal with initial underpricing, but rather is concerned with the change in the market price from the immediate aftermarket to the long term" (Bower [1989, p. 656]). Hence this longterm perspective of mispricing relates to the accuracy of market participants' initial consensus assessment of IPO firm value (i.e., the first aftermarket price). The longterm mispricing of the market's consensus valuation on the first day of public trading9 is assessed in relation to the market's valuation of the firm after more information is revealed, via examining abnormal aftermarket security returns. There are three principal reasons why the longterm perspective of IPO mispricing is adopted in this study. The first is that scant empirical evidence exists on the longterm mispricing of IPOs. In contrast to the widely documented shortterm underpricing by underwriters, the existing empirical findings do not provide conclusive evidence as to whether the market overvalues, undervalues, or accurately values new security issues, on average.10 Further, the impact of 9 This first day of public trading is presumably a time of great uncertainty about the true value of an IPO due to lack of quality information about the firm. 10 The empirical findings on longterm mispricing of IPOs will be reviewed in detail in Chapter 2. accounting numbers on the longterm accuracy of IPO valuations has not been examined. Second, the aftermarket performance of initial offerings may represent a more appropriate measure of the results of the typical investor's experience with new issues than shortterm assessments of mispricing. Specifically, it is widely contended that investors may only receive a small percentage (if any) of their desired number of shares of an initial offering due to quantity rationing of typically oversubscribed issues.11 If quantity rationing is indeed prevalent in the IPO market, then the typical investor's experience with IPOs may not be represented by the shortterm underpricing of the IPO price since many investors are unable to purchase shares at that price. The third and most important reason for the longterm mispricing perspective is that accounting numbers potentially play a greater role in the valuation decisions of market participants than in underwriters' pricing decisions. Specifically, in modern finance theory equilibrium security prices such as an IPO's first aftermarket price equal the present value of the firm's expected future cash flows. Therefore, if accounting numbers are useful in estimating such cash flows, then accounting numbers may be beneficial to market participants in their assessment of the first aftermarket price. By contrast, prior IPO research has proposed that the underpricing of the IPO price by underwriters is due to a variety of nonaccounting factors. In 11 See Rock [1986] and Beatty and Ritter [1986] for analytical explanations of shortterm underpricing based on the quantity rationing of IPOs. particular, explanations of shortterm underpricing based on potential legal liabilities, information asymmetries, and the desire to receive a higher price for subsequent seasoned security offerings have been proposed.12 Therefore, from an accounting perspective, the examination of the longterm mispricing of IPOs may be more appropriate. Three hypotheses relating accounting quality and longterm IPO mispricing are developed and empirically tested in this thesis. The first hypothesis is that the lack of quality accounting numbers at the IPO date has no effect on the mean abnormal aftermarket returns of IPO investors. However, the second hypothesis states that the dispersion (variance) of these abnormal returns is greater in instances where quality accounting information is lacking. In other words, these first two hypotheses suggest that investors are able to value IPOs in an unbiased fashion (i.e., not systematically overvalue or undervalue) even in the presence of considerable uncertainty due to insufficient and/or unreliable accounting numbers. However, the lack of quality accounting numbers influences the precision of investors' valuations of IPOs. Thus, it is postulated that the absolute magnitude (rather than the direction) of longterm IPO mispricing is related to the quality of accounting disclosures provided to potential investors. The third hypothesis provides the time path of Hypothesis 2. Specifically, it predicts that the influence of the quality of accounting numbers available at the IPO date on the variance of investors' abnormal returns diminishes over time. Hypothesis 3 implies that the accounting 12 See footnote 7. information supplied at the IPO date becomes less important in the valuation of the firm over time as more timely information becomes known about these newly public firms. Three empirical surrogates for the quality of accounting information available to IPO investors are employed in this study. The first is the reputation (or quality) of the auditor utilized in the IPO process. Auditor reputation is measured dichotomously according to whether a Big Eight or a nonBig Eight auditing firm is utilized in the going public process. A measure of the potential managerial manipulation of reported accounting earnings as reflected in managers' accounting procedure choices forms the second surrogate for quality. This accounting procedure choice proxy is measured dichotomously according to whether the methods utilized in accounting for inventory and depreciation are income decreasing (i.e., "conservative") or income increasing (i.e., "liberal"). The third proxy is whether the SEC registration statement filed by the IPO issuer is a Form S1 or Form S18 registration. This third surrogate measures the quantity component of accounting quality since the accounting disclosure requirements of Form S1 registrations are much more extensive than those of Form S18 filings. The empirical results provide support for the first two hypotheses. Specifically, the results indicate that certain accounting choices which provide indications of accounting quality (i.e., the quality of auditor utilized in the IPO process, the type of SEC registration form filed, and the degree of conservatism inherent in inventory and depreciation choices) have no appreciable impact upon the 8 mean abnormal aftermarket returns of IPO investors. However, these same accounting choices (with the exception of auditor quality) influence the variance of investors' aftermarket abnormal returns. Thus, the results indicate that uncertainty relating to accounting quality translates into an increased variance of IPO abnormal returns, rather than a systematic overvaluation or undervaluation of IPOs by investors. However, the empirical results do not support the third hypothesis. Specifically, the results do not reveal a strong tendency for the accounting quality proxy variables to decrease in their ability to explain abnormal return variances as time progresses subsequent to the initial offering. CHAPTER 2 PRIOR RESEARCH ON INITIAL PUBLIC OFFERINGS Introduction Bower's [1989] specification of the two types of mispricing inherent in IPO valuation is utilized in this chapter as an aid in classifying and summarizing the relevant prior IPO research.' Thus, this chapter will review in detail (1) the shortterm underpricingg" literature and (2) the extant longterm mispricing literature. ShortTerm Mispricing of IPOs Summary Results As indicated in Chapter 1, the empirical evidence on the short term average underpricing of initial offers is extensive. In his review of this area of research, Smith [1986] reports that the average level of underpricing exceeds 15%. In other words, on average, an investor could purchase an initial offering at the price set by the underwriter and then sell the security at the prevailing market price at the end of the 1 This classification scheme ignores all IPO research not pertaining to the "mispricing" of the security issue. Thus, the major line of research relating to the entrepreneur's ability to convey the firm's "type" to investors via various signaling mechanisms will not be reviewed here. The interested reader is referred to Leland and Pyle [1977], Downes and Heinkel [1982], Ritter [1984a], Titman and Trueman [1986], Hughes [1986], Feltham et al. [1988], Krinsky and Rotenberg [1989], and Gale and Stiglitz [1989]. first day of public trading and earn a one day return (before transactions costs) in excess of 15%.2 Thus, conclusive empirical evidence exists concerning the shortterm underpricing of initial offers by underwriters. Table 21 summarizes the results of a number of studies in this area. Proposed Explanations of ShortTerm Underpricing Various analytical models have been proposed as explanations of the shortterm IPO underpricing phenomenon. Rock's [1986] model is based on an asymmetry of information between "informed" and "uninformed" investors. The informed investors have purchased perfect information concerning the true value of an IPO. A major result of Rock's analysis is that underpricing is required to entice the uninformed investors to enter the IPO market. In the model, the informationally advantaged investors only submit purchase orders for IPOs which are underpriced (i.e., those in which the IPO price is less than the known true price). By contrast, uninformed investors are unable to differentiate underpriced and overpriced IPOs and must either not participate in the IPO market or submit purchase orders for both underpriced and overpriced issues. Thus, assuming uninformed demand is sufficient to fully subscribe a new issue, uninformed investors are allocated 100% of the shares in an overpriced issue and only a fraction of the shares in an (oversubscribed) underpriced issue. This rationing of underpriced 2 It is important to note that on average IPOs are underpriced and that any one particular initial offer could be overpriced. In fact, evidence in Tinic [1988] and Miller and Reilly [1987] indicates that approximately 3040% of IPOs are not underpriced. 10 0 W) 4 a) 0'4 W a r14 4i 4n 0 .,4 41i Cfl "l4 a r) m c w u CN40 z 0. c4 4l4 0 0 04 1 E4 4, Q) 0 cu 4 44i 4 (414(41 0 0 P. i1 co If 1~O o 00 00 "I CM C! il O aN 4 4 4 00 %O I I CN,  !~ in so s t4 a' a'> 4 *a) 0 4) ca ca >)> a) u a) 'a Ca Hi a) Ca 4) 44J 4> 44 44 U] ul U) U) V) 4 4 4 14 4 4 1 (1i (440 0 as os 0 0 4 "do0 4i (44 4 Q) '4 co Ma~ 414 cl ul U a)a 4 4 4 00 00 ON 4 ca o 0 4i 0 414 "4 Q U issues between informed and uninformed investors forms the basis of the underpricing phenomenon in Rock's model. Beatty and Ritter [1986] summarize the model's major result as follows: Consequently, if an uninformed investor is allocated shares in an initial public offering, there is a greater than usual chance that the issue will start trading at a discount in the aftermarket. In other words, for an uninformed investor, the expected return conditional upon being allocated shares is less than the expected return conditional upon submitting a purchase order. But an uninformed investor will participate in the market only if the expected return conditional upon being allocated shares is nonnegative. This can only happen if, on average, issuers underprice their shares. The owners of a firm going public, who typically have a large proportion of their wealth invested in the firm, would be willing to pay this price if they are sufficiently riskaverse. (Beatty and Ritter [1986, p. 228]) Beatty and Ritter's [1986] extension of Rock's model indicates that the greater the ex ante uncertainty (i.e., uncertainty about the true value of the issue), the greater the informed investors' informational advantage, and thus the greater the level of expected underpricing required to ensure uninformed investors' participation in the offering. Thus, Beatty and Ritter conclude that ex ante uncertainty and expected underpricing are positively related. This positive association between ex ante uncertainty and underpricing has been empirically verified by Beatty and Ritter [1986], Ritter [1984b], and Miller and Reilly [1987]. Baron [1982] also attempts to explain the average shortterm underpricing of IPOs via asymmetric information arguments. In his model the underwriter possesses an informational advantage over the IPO issuer regarding investor demand for the upcoming security issue. In an analytical principalagent framework, Baron demonstrates that the optimal contract between the issuer (i.e., the principal) and the underwriter (i.e., the agent) involves the delegation of the offer price decision to the better informed underwriter. This is in addition to the underwriter's role as distributor of the issue. A major result of the model is that the issuer will be willing to accept a lower offer price decision from the underwriter (and hence greater underpricing) as the issuer's uncertainty surrounding investor demand for the upcoming issue increases. Thus, Baron's model also implies a positive association between underpricing and uncertainty. Tinic [1988] proposes an "implicit insurance" explanation of underpricing. He summarizes the implicit insurance hypothesis as follows: Stated briefly, I demonstrate that gross underpricing serves as an efficient form of protection against legal liabilities and the associated damages to the reputations of both the investment bankers and the issuers. In other words, it is a form of implicit insurance against potential liabilities that may arise from the "duediligence" and disclosure requirements of the federal securities regulations. Although avenues for legal recompense against misinformation and fraud existed before the Securities Act of 1933, the Act, in effect, replaced the principle of caveat emptor with the dictum: "Let the issuer and the investment banker beware." (Tinic [1988, p. 790]) Even though Tinic's shortterm underpricing explanation is not based on informational asymmetries, his model is consistent with Beatty and Ritter's [1986] conclusion that underpricing is positively associated with ex ante uncertainty. A recent line of research proposes that shortterm underpricing can be utilized as a signaling mechanism by entrepreneurs of high quality IPO firms. In Welch's [1989] model, high quality firms underprice their initial offering in order to obtain a higher price in subsequent seasoned security offerings. A separating equilibrium results due to the prohibitive costs which must be borne by lesser quality firms attempting to imitate the underpricing signal of higher quality firms. Thus, Welch's model is consistent with the hypothesis that the purpose of underpricing is to "leave a good taste in investors' mouths" (Ibbotson [1975, p. 264]). Grinblatt and Hwang [1989] formulate a two signal model in which underpricing and the percentage of ownership retained by the entrepreneur serve as signals of the IPO firm's unobservable "intrinsic" value and cash flow variance. Since underpricing signals the "intrinsic" value of the IPO firm to investors, this model is also consistent with the hypothesis that the underpricing of the initial offering can be utilized by high quality firms in order to receive a higher price in subsequent security offerings. The prior paragraphs describe numerous proposed explanations for the shortterm underpricing phenomenon. However, it appears that Ibbotson et al. [1988] are correct in their assessment of the current state of underpricing research. They state: "A number of hypotheses have been offered to explain the underpricing, but to date there is still no persuasive, widely accepted, and testsupported explanation of IPO underpricing" (Ibbotson et al. [1988, p. 37]). Accounting Research Related to ShortTerm Underpricing Accounting research related to shortterm IPO underpricing examines the impact of auditor reputation on initial underpricing. Beatty [1989a] and Balvers et al. [1988] hypothesize that auditor reputation and initial underpricing are inversely related. Beatty utilizes the major result of Beatty and Ritter [1986] (i.e., expected underpricing is positively associated with ex ante uncertainty) in the formulation of this hypothesis. Specifically, since the attestation of an IPO's financial statements by a high reputation auditor serves to reduce investors' ex ante uncertainty, then auditor reputation is hypothesized to be inversely related to underpricing. Balvers et al. extend the models of Rock [1986] and Beatty and Ritter [1986] by explicitly considering the relationship between an IPO's auditor and underwriter. In the model, high auditor reputation reduces investors' uncertainty regarding an IPO's earnings and serves as a positive signal of the reputation of the underwriter. Thus, auditor reputation is inversely related to ex ante uncertainty, and therefore is hypothesized to be inversely related to underpricing. Empirical support of this hypothesized relationship between auditor reputation and shortterm underpricing is documented by Beatty [1989a, 1989b] and Balvers et al. [1988].3 Thus, the results of prior research seem to suggest that an entrepreneur of a firm contemplating an initial offering may be able to 3 However, Balvers et al. [1988] conclude that the impact of auditor reputation on underpricing is reduced when a high reputation underwriter is also involved in the registration process. Thus, auditor and investment banker reputation appear to be (at least partial) substitutes. influence the shortterm underpricing of the issue by his auditor choice.4 LongTerm Mispricing of IPOs Summary Results As indicated in Chapter 1, longterm mispricing refers to the accuracy of the market's initial valuation of an IPO. The accuracy of this potentially "uninformed" first aftermarket price is assessed in relation to security prices which prevail after additional information about the firm is revealed, via examination of abnormal aftermarket returns. In contrast to the extensive literature on shortterm IPO mispricing, little prior research exists on the longterm mispricing of initial security offerings. Also, the evidence to date is not conclusive as to whether, on average, market participants overvalue, undervalue, or correctly value initial offerings of securities. Table 22 summarizes the results of prior examinations of longterm IPO mispricing. With the exception of Ritter [1989], crosssectional explanations of longterm mispricing are unavailable. Ritter documents considerable variation in the level of longterm mispricing across industries and '4 The impact of other accounting choices on initial underpricing has not been examined. For example, prior research has not explored the effect of the accounting procedures chosen by IPO firms or the amount and content of accounting information disclosed on shortterm underpricing. Even though this dissertation examines the effect of accounting numbers on longterm IPO mispricing, some exploratory results of the impact of certain accounting choices on shortterm underpricing are provided in Appendix A. ("n CV CT 0% cn M" mo r m\ Om 0N ON 4 4 4 4 0N 0' CT^ 0' 4J1 4.J 41 co co cc 14 LW 44 (444.. ". 4 4 4 14 ho tko bo ) ,4 .4 "4 in .4 .4 T4 4J 0 .1I u( ,4 441 C*4 ca cU CU cz Cd CU ca ca >4 >4 >4 > >4 >4 >4, t) a) ) a) a) a) a) a4 (a 10 a) 3 3 U, 14 I 4J a) to 4 0 co c 4 04I 0A 4 at) > 0 1 .401 I., kn I 0N ON 40 It 04 CY) ON N 41 a 4 ) 0 4 & co 4 ^1 CL En h o P4 0 o4 z ON M 0 4, a) 171 co 4 (a 0 4 C/) 0 41 4 (a1 (j o 0 A4 P6 year of IPO issuance. This dissertation contributes to the literature by providing explanations of longterm IPO mispricing based on cross sectional differences in accounting quality. Research on the LongTerm Aftermarket Performance of IPOs Reilly and Hatfield hypothesize that underwriters' shortterm underpricing of IPOs continues into the longterm "as the market continues to recognize and adjust for the underpricing" (Reilly and Hatfield [1969, p. 74]). Thus, they argue that the market for initial security offerings is inefficient in that the market initially undervalues IPO firms, on average. They calculate shortterm mispricing as the percentage change from the IPO price to the prevailing price one week (or month) following the IPO. Longterm performance is defined as the return from one week (or month) following the IPO to one year subsequent to the commencement of public trading. Reilly and Hatfield do not empirically assess the average abnormal aftermarket returns of their entire sample. Rather, they demonstrate that the subset of firms which outperformed the market in the shortterm also had abovemarket longterm returns. Thus, Reilly and Hatfield's results do not provide an indication of the average longterm mispricing of IPOs; however, the results may indicate that shortterm and longterm mispricing are correlated. McDonald and Fisher hypothesize that in an efficient market "subsequent price behavior of the stock should be independent of the initial rate of return at offering" (McDonald and Fisher [1972, p. 97]). They calculate abnormal returns by subtracting the return on the market portfolio from an IPO's return. Thus, this adjustment for overall market performance does not vary across securities (i.e., the Capital Asset Pricing Model (CAPM) beta parameter is assumed to equal one for all firms). The authors' results support their contention that long term mispricing is independent of the level of shortterm mispricing. However, average abnormal returns of 18.1% are documented from one week after the offering to one year after the offering. This negative abnormal return implies that investors were overly optimistic in their initial valuations of IPO firms, and therefore lost 18.1% of their investment in the first year of public trading due to such incorrect valuations. Thus, McDonald and Fisher's results indicate that the market overvalued, on average, IPOs issued in the first quarter of 1969. Ibbotson [1975] examines the abnormal aftermarket performance of a sample of firms which went public between 1960 and 1969. Abnormal returns are calculated over various holding periods up to 60 months subsequent to the IPO by means of the twoparameter Sharpe [1964] and Lintner [1965] CAPM. The empirical results demonstrate very few statistically significant abnormal returns in the aftermarket. Thus, Ibbotson concludes that IPO aftermarket returns do not deviate from market efficiency. In other words, Ibbotson's results are consistent with the hypothesis that the market utilizes all available information in an unbiased fashion and hence, on average, accurately values new issues. The most extensive examination of the aftermarket performance of IPOs is provided by Ritter [1989]. He explores the longterm mispricing of 1,526 firms that made initial offerings between 1975 and 1984. Monthly abnormal returns are calculated by subtracting the return on a market index from an IPO's raw return for each of the first 36 months following the IPO. Thus, marketadjusted rather than riskadjusted abnormal returns are computed. Ritter documents average abnormal returns of 24.33% over the first three years of public trading. Thus, the results are consistent with the hypothesis that the market overvalued IPOs in the 19751984 period. However, considerable variation in the level of this overvaluation across year of issuance and industries is documented. In particular, Ritter states that the overvaluation is "concentrated among relatively young growth companies, especially those going public in the 1980s" (Ritter [1989, p. 3]). Comparison of the LongTerm Mispricing of New Issues Before and After the 1933 Securities Act Stigler [1964] and Jarrell [1981] compare the longterm mispricing of new security issues5 offered before and after the 1933 Securities Act (the Act). Both researchers test the hypothesis that the period prior to the disclosure provisions of the Act was characterized by exaggerated claims of issuers and underwriters, resulting in overvaluation of new issues by investors. Specifically, both Stigler and Jarrell examine whether the disclosure requirements of the 1933 Securities Act resulted in increased investor returns for post1933 new issues as compared to the returns earned on potentially overvalued pre1933 issues. 5 These papers examine the aftermarket abnormal returns on all new issues of securities. In other words, both IPOs and new security offerings of existing publicly held corporations (i.e., seasoned issues) are examined. Stigler [1964] reports negative average marketadjusted returns for both pre1933 and post1933 new security issues over each of the first five years after issuance. Thus, Stigler concludes that the disclosure requirements of the Act did not result in increased returns for investors in the post1933 era. Stigler controls for market influences by comparing aftermarket prices to the value of a market index. The preSEC sample is taken from the years 19231928, while 19491955 is utilized as the postSEC period. Jarrell [1981] also rejects the hypothesis that average aftermarket performance improved significantly after passage of the Act. He calculates abnormal returns by means of the twofactor CAPM. His preSEC sample covers the years 19261933, while the postSEC period runs from 19341939. Jarrell's results indicate positive mean risk adjusted returns for both groups over a five year holding period. Contrary to the hypothesis tested however, the cumulative returns for the preSEC sample exceeded those of the postSEC sample. Simon [1989] also examines the aftermarket returns of preSEC versus postSEC new security issues. However, she hypothesizes that the disclosure provisions of the 1933 Securities Act should have no effect on the average abnormal returns earned by investors. In other words, she postulates that rational investors were able to accurately value new issues, on average, prior to the Act's increased disclosure requirements since "the existence of substantial uncertainty about the true value of a security need not imply that the issue will be, on average, overvalued or undervalued" (Simon [1989, p. 295]). However, Simon theorizes that the lack of quality information about the value of a firm undertaking a 22 new security issue which characterized the preSEC period increased the level of riskiness associated with the purchase of the security. Thus, she hypothesizes that the variance of aftermarket abnormal returns decreased after the passage of the 1933 Securities Act. Simon utilizes a multifactor asset pricing model in the determination of riskadjusted abnormal returns. In addition to the return on the market, the model includes the industry return, the unexpected variance of the market return, and a business cycle indicator as explanators of equilibrium security returns. She classifies new issues from 19261933 as preSEC issues, while 19341940 offerings are deemed postSEC issues. Contrary to expectations, Simon documents significantly negative average abnormal returns for nonNYSE initial public offerings in the preSEC period. Thus, it appears that investors overvalued, on average, IPOs of nonNYSE firms prior to the passage of the 1933 Securities Act. However, results consistent with the hypothesis that average abnormal returns would not differ significantly from zero are documented for NYSE issues and seasoned nonNYSE issues in both the pre and postSEC periods. Further, in contrast to Stigler [1964] and Jarrell [1981], Simon demonstrates a statistically significant increase in average aftermarket abnormal returns for non NYSE initial public offerings from the preSEC to postSEC periods. No such increase was present for either NYSE issues or nonNYSE issues of currently publicly traded corporations. Finally, the hypothesis that the variance of aftermarket returns significantly decreased following the disclosure requirements of the Act is empirically supported for each portfolio of new security offerings. 23 Simon's study represents an intervention analysis of a timeseries nature (i.e., differences between the pre and postSEC periods are examined). In contrast, the current study deals with information variables from a crosssectional perspective. Summary of Longterm Mispricing Research The results of prior research conflict as to whether, on average, the market overvalues, undervalues, or accurately values new issues of securities. In assessing the longterm performance of new issues, prior researchers have employed diverse methods of calculating abnormal returns and widely divergent sample years. It is possible that the contradictory findings are time period specific and/or dependent upon the asset pricing model utilized in the determination of abnormal returns. This dissertation extends prior research by (1) reexamining the longterm mispricing of initial public offerings by utilizing a refined model of equilibrium security pricing6 on a sample of recent IPOs7 and (2) testing hypotheses relating to the effect of accounting numbers on such mispricing. 6 A modified version of Simon's [1989] multifactor asset pricing model will be implemented in this study. The model will be described in detail in Chapter 4. 7 In contrast to Stigler [1964], Jarrell [1981], and Simon [1989], only initial public security offerings will be examined. Thus, new seasoned issues will be excluded. Also, this study will only examine the longterm mispricing of IPOs initiated since 1980. With the exception of Ritter's [1989] 19751984 sample period, previous studies have not utilized post1970 IPOs. CHAPTER 3 HYPOTHESIS DEVELOPMENT Introduction The three hypotheses developed and empirically tested in this dissertation relate to the impact of the quality of accounting information available to IPO investors on longterm IPO mispricing. With the aid of three assumptions, this chapter develops these hypotheses in a Capital Asset Pricing Model (CAPM) framework.1 Assumptions First, it is assumed that the total uncertainty surrounding an initial offering arises from two sources. Specifically, assume that an IPO's firmspecific risk may be decomposed2 as follows: Firmspecific risk the uncertainty related to the firm's production, investment, and financing activities + the uncertainty related to the quality of the accounting numbers supplied to potential investors. (la) In mathematical terms, Equation (la) becomes FSR1 BUSRISKi + ACCTRISKi, (lb) SThe model utilized in this chapter to derive the hypotheses is an adaptation of the model found in Jarrell [1981]. 2 This partitioning of firmspecific risk is in addition to the CAPM's systematic (i.e., nondiversifiable) versus unsystematic (i.e., diversifiable) decomposition. where FSRj is firm specific risk, BUSRISKi equals firms i's business risk, and ACCTRISK1 denotes accounting risk. The first component of the preceding definition describes the cash flow uncertainty associated with the firm's production, investment, and financing activities, and will hereafter be referred to as business risk. In other words, business risk represents the underlying uncertainty related to the overall business activities (both operating and financial) of the firm. The second component in the above characterization of an IPO's firmspecific risk (hereafter referred to as accounting risk) reflects the influence of accounting quality on IPO firm uncertainty. According to Equation (Ib), the total level of uncertainty surrounding an IPO equals the firm's business risk (i.e., the lower bound) only when there is no uncertainty surrounding the quality of the accounting numbers supplied to potential investors. However, when uncertainty surrounding accounting reliability exists, the total uncertainty surrounding the security issue exceeds the IPO's business risk, and increases as accounting risk increases.3 3 Equation (Ib) characterizes the total risk of an initial offering as a linear and additive function of business risk and accounting risk. Thus, the equation assumes the independence of the two concepts. This assumption appears to be reasonable, since there is no a priori reason to postulate the absence of any of the four possible combinations of business and accounting risk. Specifically, one could reasonably expect to find IPO firms in each of the following categories: (1) high business risk and a high level of accounting reliability and accuracy, (2) high business risk and a low level of accounting reliability, (3) low business risk and a high level of accounting reliability, and (4) low business risk and a low level of accounting reliability. Thus, it seems reasonable to assume that the level of uncertainty surrounding an IPO firm's production, investment, and financing activities (i.e., its inherent cash flow uncertainty) is independent of the precision of its accounting valuations. Second, it is assumed that the twofactor SharpeLintner CAPM appropriately describes equilibrium security valuation.4 The model characterizes the expected return on a security as a linear function of the expected market return. The model (in expectation form) may be written E(Rit) RFt + Pi(E(RMt) RFt), (2) where E the expectation operator, Rit the security return on IPO firm i over time period t, RFt the riskfree rate for time t, Rmt the time t return on the market portfolio, and i firm i's systematic risk. The CAPM is often utilized in its realized form for empirical purposes. In realized form, the model is Rit RFt + Pi(RMt RFt) + eit, (3) where eit is an error term and all other terms are as previously defined. Systematic risk refers to that portion of firmspecific risk which cannot be diversified away. In other words, Oi reflects the riskiness of security i in relation to marketwide risk, and is calculated as i Cov(Rit, RMt) / Var(Rmt). (4) According to the CAPM, the remaining portion of firmspecific risk is labelled unsystematic risk. Unsystematic risk is calculated as the variance of the error terms in the realized formulation of the model 4 The assumption that the CAPM accurately reflects equilibrium security valuation is implemented in this section for expositional purposes only. In fact, in the empirical analysis to follow, a multifactor model is implemented in an attempt to overcome certain deficiencies of the CAPM. (i.e., Var(eit)). In a CAPM framework, unsystematic risk does not affect the expected return on a security since its effects may be diversified away by holding a welldiversified portfolio. Finally, this study makes use of the assumption that investors are rational. In other words, this rational expectations assumption implies that market participants form unbiased estimations of (1) an IPO's systematic risk and (2) the future payouts arising from IPO ownership.5 Decomposition of Systematic and Unsystematic Risk into Accounting and Business Risk Components The CAPM partitions firmspecific risk into systematic and unsystematic components. This section further decomposes the CAPM's firmspecific risk components into the business risk and accounting risk partition of Equation (Ib). Specifically, i may be defined as Pi = BiB + piA, (5) where PiB systematic risk attributable to firm i's production, investment, and financing activities, and piA = systematic risk attributable to uncertainty surrounding the quality of the accounting numbers supplied to investors. The systematic risk attributable to accounting risk is modelled as follows: ^ _i^(TQ), (6) SThese expected future payouts can be thought of as either dividend distributions or the firm's cash flows. where T = time subsequent to the IPO and Q the quality of accounting information disclosed to investors at the time of the IPO. It is assumed that systematic accounting risk is a decreasing function of both time and accounting quality at the IPO date. Specifically, apiA/aT < 0 and (7a) afiA/aQ < 0. (7b) In other words, subsequent to the IPO, systematic accounting risk decreases over time as more information is revealed about the firm.6 Also, the disclosure of a sufficient quantity of reliable accounting information at the IPO date reduces systematic accounting risk. Similarly, unsystematic risk may be divided into business and accounting risk components as follows: Var(ei) Var(ejB) + Var(eiA), (8) where Var(ej) = firm i's unsystematic risk which is calculated as the variance of the error terms in the realized formulation of the CAPM, and Var(e B) and Var(eiA) equal the business risk and accounting risk components of unsystematic risk, respectively. As was the case for systematic accounting risk, unsystematic accounting risk is modelled as a decreasing function of time subsequent to the IPO as well as accounting quality at the IPO date. Specifically, Var(e A) = Var(eiA)(T,Q), (9) 6 This additional information revealed in the periods following the IPO may consist of both accounting (e.g., quarterly and/or yearly financial statements or analysts' forecasts of earnings) and nonaccounting (e.g., security prices or Wall Street Journal articles) information. This additional information will reduce accounting risk only if it is deemed reliable. 29 where aVar(eiA)/aT < 0 and (10a) aVar(eiA)/aQ < 0. (10b) Hypotheses Overview This section demonstrates the effects of changes in accounting risk on the distribution of investors' aftermarket abnormal returns (i.e., longterm mispricing). First, the effect of changes in accounting risk on mean abnormal returns is explored. Next, accounting risk's effects on the dispersion (variance) of abnormal riskadjusted returns is examined. Finally, the impact over time of accounting risk on the variance of abnormal returns is portrayed. Accounting Risk and Mean Abnormal Returns An increase in accounting risk (ACCTRISK) leads to an increase in both the systematic and unsystematic components of accounting risk. According to Equations (5) and (8), such increases in iA and Var(eiA) result in increases in both systematic and unsystematic risk. It is evident from Equation (2) that such an increase in unsystematic risk has no effect upon expected mean returns in a CAPM framework. The effects of increases in unsystematic risk may be eliminated in a well diversified portfolio and thus do not affect mean expected security returns. In contrast, it is apparent from Equation (2) that an increase in systematic risk increases investors' expected raw (i.e., unadjusted) returns. However, assuming that investors are rational, it is readily seen that an increase in systematic accounting risk should not cause riskadjusted abnormal returns to deviate from zero since the increased raw returns demanded by investors merely compensate them for bearing greater risks. The rational expectations assumption implies that investors are able to form unbiased estimates of an IPO's future payouts even in the presence of considerable uncertainty surrounding the value of the firm (e.g., in instances of high accounting risk). In other words, on average, rational investors do not overestimate or underestimate a firm's cash flows due to unreliable and/or incomplete accounting disclosures. Increased accounting risk merely induces investors to demand a greater return per dollar of expected future payout (cash flow) to compensate them for bearing increased risks. Thus, on a riskadjusted basis, IPOs are on average correctly valued owing to the unbiased nature of the estimated valuerelevant cash flows. In other words, rational investors do not misprice IPOs, on average, even in the presence of a high degree of systematic accounting risk. Thus, neither unsystematic nor systematic accounting risk should cause mean riskadjusted aftermarket returns to deviate from zero. This assertion gives rise to the first hypothesis, which may be formally stated as follows: Hypothesis 1: On average, the longterm mispricing of IPO firms is zero regardless of the level of accounting risk faced by investors. Accounting Risk and the Variance of Abnormal Returns The rational expectations assumption implies that even though investors' estimates of future payouts are unbiased even in the presence of high accounting risk, the precision (variance) of these estimates decreases (increases) with accounting risk. This increase in the variability of cash flow estimates should translate into greater variability in returns. In fact, just such an increase in variability can be demonstrated in a CAPM framework. The variance of a security's returns according to the CAPM (in realized form) is Var(Ri) Oi2Var(RM) + Var(ei), (11) where Var(Ri) the variance of security i's returns, Var(RP) the variance of the return on the market portfolio, and the remaining terms are as previously defined. As previously noted, Equations (5) and (8) demonstrate that an increase in accounting risk results in an increase in both systematic (i.e., Pi) and unsystematic (i.e., Var(ei)) risk. Therefore, it follows directly from Equation (11) that the variance of a security's returns increases with accounting risk. The preceding assertion forms the basis for the second hypothesis, which may be stated as Hypothesis 2: The variance of IPO firms' abnormal returns is positively associated with accounting risk. Thus, it is consistent with rational expectations that the absolute magnitude (rather than the direction) of IPO mispricing is related to the level of accounting risk surrounding an issue, even though IPOs are not systematically mispriced. Accounting Risk and the Variance Over Time of Abnormal Returns Equations (7a) and (10a) model systematic and unsystematic accounting risk, respectively, as decreasing functions of time subsequent to the IPO. Thus, this decrease in fliA and Var(ejA) over time results in a reduction in both systematic (i.e., 1i) and unsystematic (i.e., Var(ei)) risk. It then follows directly from Equation (11) that such decreases in systematic and unsystematic risk result in an IPO firm's abnormal return variance decreasing over time. The preceding implication of the model gives rise to the third hypothesis, which may be stated as follows: Hypothesis 3: The variance of IPO firms' abnormal returns decreases over time. This third hypothesis is consistent with a diminishing effect of accounting risk on the precision of IPO valuation (i.e., longterm mispricing) over time as the information disclosed at the time of the IPO takes on a lesser role in the valuation of the firm as new information is revealed.7 7 See footnote 6. CHAPTER 4 EMPIRICAL METHODOLOGY Introduction The empirical methodology implemented in the testing of the three hypotheses is described in this chapter. First, the three variables utilized as empirical proxies for the accounting risk theoretical construct are discussed. The measurement of abnormal riskadjusted aftermarket returns (i.e., longterm mispricing) is then illustrated. Next, both univariate and multivariate tests of the three hypotheses are detailed. Finally, a summary of the empirical methodology is provided. Three Measures of Accounting Risk Overview of the Three Accounting Risk Surrogates The accounting risk theoretical construct is a multifaceted concept which is difficult to empirically capture with any one proxy variable. Thus, this dissertation utilizes three surrogates for accounting risk, each of which is intended to capture a different aspect of accounting riskiness. The three unique components of accounting risk captured by this study's proxy variables are (1) the level of credibility associated with the reported accounting numbers, (2) the level of bias inherent in those numbers, and (3) the quantity of accounting disclosures supplied to potential IPO investors. Auditor reputation, the degree of conservatism inherent in a firm's inventory and depreciation accounting procedure choices, and the type of SEC registration statement filed, respectively, are employed in this study as empirical surrogates for these three aspects of accounting risk. The remainder of this section provides detailed descriptions of the three accounting risk proxy variables. Auditor Reputation (Quality) Proxy The reputation (or quality) of the auditor utilized in the initial public offering process serves as the first proxy for accounting risk. The primary function of the auditor in the IPO process is to express an opinion on the IPO firm's financial statements. Therefore, it is reasonable to assume that a high quality audit lends credibility to the financial statements, thereby reducing accounting risk. Conversely, a low quality audit adds little credibility to the issuer's financial statements, and thus investors' level of uncertainty surrounding the reliability and accuracy of accounting numbers will be correspondingly greater. Thus, audit quality is a surrogate for the level of credibility associated with reported accounting numbers, which is an aspect of accounting riskiness. An IPO firm's accounting risk (based on auditor quality) is measured in terms of whether the accounting numbers supplied to potential investors at the time of the initial offer were audited by a 35 Big Eight or nonBig Eight auditing firm.' DeAngelo [1981] models audit quality as a positive function of auditing firm size. She argues that this relationship results from the greater level of independence which is employed by large audit firms in an effort to maintain their reputation capital. In an IPO context, this presumed higher quality of large audit firms might also be due to larger firms' greater expertise in industryrelated accounting matters, greater experience in auditing developmental stage companies, stricter internal control requirements, etc. If audit quality is indeed positively associated with audit firm size, accounting risk should be inversely related to auditor size. In other words, the preceding arguments imply that Big Eight audited IPO firms should possess less accounting risk than IPOs audited by nonBig Eight firms. However, a number of arguments against this presumed positive association between audit quality and audit firm size can be raised. First, DeAngelo's [1981] analytical model which predicts this relationship has not been empirically verified. A second argument against DeAngelo's hypothesized relationship is that individual audit 1 NonBig Eight firms consist of both "national" (i.e., the mid sized firms with offices throughout the country) and local/regional auditing firms. Thus, it is possible to classify auditing firms into three categories, rather than the traditional Big Eight, nonBig Eight dichotomy. For example, Simunic and Stein [1986] classify auditing firms into three categories (i.e., Big Eight, "national," and local/regional) and then discard the middle group in order to increase the power of the test of auditor reputation on IPO valuation. Due to the small number of nonBig Eight audited IPO firms in the sample analyzed in this dissertation (see Chapter 5), a further breakdown of this group into "national" and local/regional categories does not appear to be a viable alternative in this study. Therefore, the traditional Big Eight, nonBig Eight dichotomy is employed. engagement partners in large decentralized auditing firms may have the same incentives in regards to compromising their independence as partners in small local/regional firms. In each case, a particular audit client may represent a significant proportion of the revenue generated by a partner. Therefore, audit engagement partners in large decentralized firms may not consider the ramifications of certain actions on the firm as a whole. Thus, the predictions of DeAngelo's model may not hold in large decentralized audit firms. Finally, it is possible to dispute the argument that large audit firms have greater expertise in industryrelated accounting matters. Specifically, regional auditing firms may specialize in the predominant industry in that region. For example, a regional auditing firm in Texas may have greater expertise in the oil and gas industry than a large national firm headquartered in New York. The preceding arguments against the presumed positive association between audit quality and audit firm size imply that the Big Eight, nonBig Eight dichotomy employed in this study may not accurately proxy for accounting risk. Accounting Procedure Choice Proxy The second accounting risk surrogate utilized in this study is measured according to whether the accounting numbers provided to potential IPO investors were calculated by utilizing income decreasing (i.e., "conservative") or income increasing (i.e., "liberal") accounting methods. This proxy is utilized in recognition of the potential of a firm's managers (or entrepreneurs in the IPO context) to manipulate reported earnings via certain accounting procedure choices.2 An entrepreneur of a firm about to undertake an initial public security offering might be motivated to manipulate accounting numbers in order to send an exaggerated (i.e., positively biased) signal to potential investors concerning the earnings history (and potential) of his firm. By definition, accounting risk (i.e., the uncertainty surrounding the reliability and accuracy of accounting numbers) increases as the level of managerial manipulation of accounting numbers increases. Thus, an estimate of the degree of such manipulation is employed as a proxy for accounting risk. This surrogate is intended to capture the component of accounting risk which relates to the level of bias inherent in reported accounting numbers. Imhoff and Thomas [1989] provide empirical support for the utilization of this conservative versus liberal accounting procedure choice dichotomy as a proxy for accounting quality. They demonstrate a statistically significant positive association between the ratings of financial analysts regarding the accounting quality of firms in the particular industry in which each analyst specializes and the degree of conservatism of the accounting methods employed by these firms. Thus, Imhoff and Thomas [1989] provide evidence that financial analysts view 2 Watts and Zimmerman [1986] review the literature (both theoretical and empirical) which suggests that managerial manipulation of earnings is related to the existence of political costs, debt covenants, and bonus plans denominated in terms of accounting numbers. firms which utilize conservative accounting methods as possessing higher accounting quality than firms which utilize more liberal methods.3 In this study, the liberal versus conservative accounting procedure choice proxy is measured based on whether an IPO firm's chosen inventory and depreciation methods4 tend to increase or decrease reported accounting earnings. Specifically, conservative accounting choices are defined as the lastin, firstout (LIFO) inventory valuation 3 There are other possible explanations of this perceived relationship besides the managerial manipulation of earnings explanation utilized in this study. For example, it could be argued that accounting risk is less when the "conservative" lastin, firstout (LIFO) inventory valuation method is used in comparison to the more "liberal" firstin, firstout (FIFO) method. Specifically, the LIFO method tends to smooth reported income numbers in periods of rapidly changing prices, both increasing and decreasing. These less volatile income numbers may be perceived as being of higher quality than the more variable FIFObased numbers. Also, LIFO users often report in their financial statements what income would have been under the FIFO method. However, the opposite does not hold since FIFO users typically do not report LIFO based net income. Therefore, it is possible that LIFObased numbers may be considered co be of higher quality since more information is typically available under LIFO than under FIFO. 4 Other accounting procedure choices have been examined in the accounting literature. For example, the methods employed to account for the investment tax credit (ITC) and oil and gas exploration costs have been widely examined (see Watts and Zimmerman [1986] for a review and Imhoff and Thomas [1989] for a recent example). These accounting choices were also originally planned to be included in this study's definition of conservative versus liberal accounting procedure choices. However, only 2 out of the 372 sample firms used a procedure other than the flow through ITC method and oil and gas exploration firms comprise only 5% of the sample firms. Hence, these two accounting procedure choices appear to be of little use in the current sample in classifying firms into conservative versus liberal accounting procedure choice groups. Thus, the level of conservatism of a firm's accounting procedure choices is measured based only upon inventory and depreciation method choices. The inability to examine other than inventory and depreciation choices is considered to be only a minor limitation since depreciation and inventory represent the most important (in dollar terms) accounting procedure choices for most firms. procedure and the utilization of any of a number of accelerated depreciation methods (e.g., doubledeclining balance or sumofthe years'digits). Liberal (i.e., income increasing) choices are defined as any inventory valuation method other than LIFO (e.g., firstin, firstout (FIFO), weighted average, or specific identification)5 and the straightline depreciation method. This accounting procedure choice proxy variable is measured dichotomously. The current paragraph provides an overview of the classification scheme utilized in the grouping of sample IPO firms into high and low accounting risk portfolios based on accounting procedure choices, while the following paragraph details the specifics of the classification scheme. An IPO firm is considered to possess high accounting risk if each of its available accounting method choices is income increasing (i.e., "liberal"). Conversely, an IPO is considered to be a low accounting risk firm if at least one of the two relevant accounting procedure choices is income decreasing. This accounting procedure choice variable was originally measured over three levels. Specifically, firms were classified into accounting risk portfolios based on whether their inventory and depreciation choices were (1) consistently income increasing, (2) consistently income decreasing, or (3) a combination of income increasing and income decreasing methods. However, in the sample analyzed in this study (see Chapter 5), only 10 out of the 372 sample firms used income decreasing 5 Consistent with current accounting practices, in the sample of IPOs examined in this study (see Chapter 5), the vast majority of the nonLIFO inventory choices are FIFO choices. (i.e., conservative) methods for both inventory and depreciation. Therefore, this consistently "conservative" group was combined with the group whose accounting procedure choices consisted of a combination of income increasing and income decreasing methods to form the low accounting risk portfolio in the dichotomous classification scheme. Further, a number of the sample firms did not have both inventories and longterm depreciable assets. In cases where a firm possessed no inventory (longterm depreciable assets), the measurement of this accounting procedure choice variable was based strictly upon the firm's depreciation (inventory valuation) choice. In summary, the high accounting risk firms either (1) utilized income increasing methods for both inventories and longterm depreciable assets or (2) employed an income increasing method for the only applicable accounting choice. Conversely, the low accounting risk firms either (1) utilized income decreasing methods for both accounting choices, (2) used a combination of income increasing and income decreasing methods, or (3) employed an income decreasing method for the only applicable accounting choice.6 6 Various methods of classifying firms into accounting risk groups based on the firms' accounting procedure choices were employed in the empirical analysis, with no qualitative differences in results. SEC Registration Form Proxy The final accounting risk surrogate employed in this dissertation is whether the registration statement7 filed with the SEC was on Form S18 or Form S1. Form S18 offerings were instituted by the SEC in April 1979. Prior to that time all initial offerings (with the exception of unregistered Regulation A offerings which had a limit on the amount of proceeds raised of $1.5 million) were required to be registered on Form S1. The limit on the amount of capital raised with an S18 offering is $7.5 million.8 Important differences exist in the disclosure requirements in these two types of registration statements. Form S1 registrations are more comprehensive both in terms of the number of years of data required and in the extent of data required per year. Specifically, Form S18 allows the issuer to present two years of audited financial statements, rather than the three required in an S1 offering. Further, "Form S18 relaxes or eliminates the disclosure requirements for several areas, including property, segment data, foreign operations, order backlogs, and research and development" (Manegold and Arnold [1986 p. 29]). The elimination of these potentially valuable accounting disclosures in an 7 Before securities may be sold to investors in a public offering, the SEC must approve an issuing firm's registration statement. The registration statement contains information about the proposed offering, as well as information about the issuing firm itself (e.g., audited financial statements). Similar information about the offering and the issuer is provided to potential investors in a prospectus. 8 For a detailed account of the differences between the two types of registrations, please see Manegold [1986] or Manegold and Arnold [1986]. S18 offering should lead to an increase in accounting risk, and hence the type of SEC registration statement filed comprises the third proxy for accounting risk. This third surrogate is intended to capture the quantity (i.e., extent of disclosure) aspect of the accounting risk theoretical construct. Measurement of RiskAdjusted Abnormal Security Returns In order to test Hypotheses 1 through 3, aftermarket abnormal security returns (i.e., longterm mispricing) must be calculated. A modified version of Simon's [1989] multifactor security pricing model is employed in the measurement of such excess returns.9 The model utilized is Rit RFt i(RMt RFt) + 6i(INDUSTRYit RMt RFt) 3 + EjiDjt + Eit, (12) where Rit the return on firm i in time period t, RFt = the return on riskless assets over time t, RM = the time t return on the market portfolio, INDUSTRYit = the equallyweighted average return of firms in the same industry (based on 2digit SIC codes) as firm i during time period t, Djt are time period specific dummy variables which are designed to capture abnormal returns, 6j, 6j, and 8ji are regression parameters, and eit is an error term. SSimon's model includes factors relating to business cycle activity and the unanticipated portion of the variance in market returns in addition to the variables employed in this study in the specification of the equilibrium return generating process. The Rit terms in Equation (12) represent IPO firm i's raw (i.e., unadjusted) returns over month t following the initial offering. These monthly returns are constructed from the CRSP NASDAQ daily returns file utilizing a procedure similar to that employed by Ritter [1989]. The return for month 1 is calculated as the multiplicative return over the 21 trading day period beginning with the day following the IPO.10 Thus, returns are calculated based on "event time" (with the event in question being the IPO), rather than calendar time. The monthly returns for months subsequent to month 1 are then calculated using the returns on successive 21 trading day periods. In equation form, Rit is calculated 21+21(t1) Rit n (1 + Rid) 1, (13) d=l+21(tl) where Rid firm i's raw return on day d where d1 represents the day following the IPO, and t denotes the number of months subsequent to the initial offering. A monthly market portfolio return series (i.e., Rmt) is constructed for each IPO firm corresponding to the firm's "event time" monthly periods. This monthly market return series is calculated in the same manner as the Rit series, with the exception that the daily returns utilized in the construction of the monthly returns are the returns on the NASDAQ Composite Index, rather than an individual IPO firm's raw returns. Since each of the sample firms examined in this study is traded overthecounter, the NASDAQ Composite Index appears to be a more 10 It is important to note that the return on the first day of public trading is the commonly examined oneday underpricing return. Therefore, the empirically documented shortterm underpricing phenomenon is not included in the longterm returns examined here. appropriate measure of the "market" return than either the CRSP Value Weighted or CRSP EquallyWeighted Indices (both of which consist of New York and American Stock Exchange firms) which are often utilized in capital markets research. The INDUSTRYit terms present in Equation (12) represent the average equallyweighted return of all firms on the NASDAQ daily returns file possessing the same 2digit SIC code as IPO firm i. Again, for each sample IPO firm, this INDUSTRYit series is calculated over the same "event time" months as the Ri series. The RFt terms in Equation (12) represent the prevailing "riskfree rate" at time t. The proxy employed for the unobservable true "risk free rate" is the return on United States Treasury Bills. The monthly total return series on U.S. Treasury Bills from January 1980 through December 1988 was collected from Ibbotson Associates [1989]. Then this monthly riskfree rate series was aligned with the Rit series in order to create the RFt series for each IPO firm. Equation (12) is a generalization of the twofactor CAPM utilized in the derivation of Hypotheses 1 through 3 in Chapter 3. There are two principal reasons the multifactor specification of security pricing found in Equation (12) is employed in the empirical testing rather than the traditionally utilized twofactor CAPM or singlefactor market model. First, prior research suggests that factors other than the return on the market portfolio are needed to explain equilibrium security returns.11 In this study, industry returns as well as overall 11 For example, see Ross [1976] for a theoretical development of the multifactor Arbitrage Pricing Theory (APT). 45 market returns are employed in the specification of equilibrium security returns. Industry returns are included in Equation (12) to control for abnormal return performance arising from business risk changes over time. Assuming firms in the same industry face similar production, investment, and financing environments, then INDUSTRYij represents an appropriate proxy for business risk and 65 reflects return performance attributable to changing business risk conditions. Thus, Equation (12) abstracts from both overall market performance and unexpected industry performance (i.e., business risk) in the generation of firmspecific abnormal returns (i.e., the 6ji parameters). Since the effects of both overall market risk and firmspecific business risk are taken away, it is assumed that the abnormal returns captured by the 6ji terms are attributable to accounting risk. The second reason why the traditional CAPM or market model formulation is not utilized in this study is due to the institutional features of the IPO market. In traditional capital markets "event" studies, the CAPM or market model is estimated over a period of time prior to the "event" in question. Then the estimated parameters are utilized in the determination of abnormal returns during the event period. Such an approach is infeasible for IPOs since return data are unavailable prior to the event in question (i.e., the initial offer). Thus, the relevant parameters must be estimated during the event period. The time period specific dummy variables included in Equation (12) allow the researcher to pick up abnormal returns over specified periods during the estimation period.12'13 Specifically, Dlt is coded 1 during the first 12 months following the IPO and 0 otherwise. Similarly, D2t equals 1 only for months 1324, while D3U is coded 1 for months 2536 and 0 otherwise. Thus, the parameters on the dummy variables (i.e., 911 to 63) represent an IPO firm's abnormal returns over the first, second, and third year of public trading, respectively. Equation (12) is estimated for each firm on a timeseries basis over the first 36 months of public trading. Cumulative abnormal performance over the entire 36 month estimation period may be computed by simply adding the parameter estimates on the three time period specific dummy variables. In other words, 3 CAR36i Z ji, (14) jl where CAR361 firm i's 36 month cumulative abnormal return.14 This 12 In a typical event study, the error terms (i.e., the it terms) represent riskadjusted abnormal returns. However, when the model's parameters must be estimated over the event period, the sum of the error terms is constrained to equal zero by the OLS regression procedure and thus the errors cannot be utilized as measures of abnormal performance. 13 See Schipper and Thompson [1983] for an example of the utilization of time period specific dummy variables to measure abnormal performance surrounding different events. 14 The return generating process described by Equation (12) does not include an intercept term. Instead, it includes a time period specific dummy variable for each of the first three years of public trading. Alternatively, Equation (12) could be estimated by utilizing an intercept term along with dummy variables relating only to the first two years of public trading. The model would then be written as follows: Rit RFt ai + i(RMt RFt) + 6i(INDUSTRYt RMt RFt) 2 + Z jiDjt + Eit, j=1 where ai is an intercept term and all other terms are as previously defined. In this alternative specification, ai picks up abnormal cumulative abnormal return, as well as the individual abnormal returns pertaining to the first three years of public trading, is utilized in a crosssectional format to test Hypothesis 1. Further, the variances of these abnormal return measures are used to crosssectionally test Hypotheses 2 and 3. Univariate Tests Univariate Tests of Hypothesis 1 Hypothesis 1 states that market participants do not misprice IPOs, on average, regardless of the level of accounting risk they face. In other words, average aftermarket abnormal returns are predicted to equal zero (i.e., they are unbiased) and are hypothesized not to be influenced by accounting risk. Univariate tests of Hypothesis 1 consist of constructing portfolios of IPOs based on the firms' levels of the three previously described accounting risk measures and then comparing the average abnormal returns of each portfolio over various holding periods (1) to the hypothesized value of zero and (2) to the average abnormal returns performance over the period of estimation not accounted for by the dummy variables (i.e., year 3) and the parameter estimates on the two dummy variables measure differences between the abnormal performance of years 1 and 2, respectively, and year 3. Thus, the abnormal returns relating to year 1 would be calculated by adding the dummy variable parameter estimate pertaining to year 1 (i.e., 81i) to aj. The abnormal returns of year 2 would be calculated in a similar manner. The cumulative 36 month return would then be calculated by adding the measured abnormal returns over years 1 through 3. See Maddala [1977, Chapter 9] for a more detailed discussion of the interpretation of the estimated dummy variable coefficients in instances when an intercept term (1) is and (2) is not included in the model. of the remaining portfolios. This procedure will now be described in detail for the case of the auditor quality accounting risk surrogate.15 Two portfolios of IPO firms are constructed based on the reputation (quality) of the auditor utilized in the IPO process. Specifically, portfolios are devised according to whether an IPO firm was audited by a Big Eight (i.e., low accounting risk) or nonBig Eight (i.e., high accounting risk) auditing firm. The mean abnormal return for a given accounting risk portfolio over a specific time horizon is calculated as N MARIt Z ARit / N, (15) i=l where MARpt the mean abnormal return of portfolio p over time period t, ARit firm i's abnormal return over time period t as generated from Equation (12),16 and N the number of firms in portfolio p. Portfolio mean returns are calculated for the first, second, and third year of public trading, as well as a three year cumulative return. Two sets of statistical tests are conducted on these mean portfolio abnormal returns to test Hypothesis 1. First, each MARpt is tested via a t test to determine whether it differs significantly from the hypothesized value of zero. Second, a twopopulation t test is then conducted to see if MARpt differs across the auditor quality based 15 A similar analysis is also conducted for the other two accounting risk surrogates. 16 For example, ARij equals the estimated 811 parameter from Equation (12). Similarly, ARi2 = e2i and ARA = 631. Also, the abnormal return covering the cumulative 36 month period is the CAR36i variable defined in Equation (14). accounting risk portfolios over each of the four relevant time horizons.17 Table 41 illustrates these tests. Univariate Tests of Hypothesis 2 Hypothesis 2 states that the dispersion (i.e., variance) of IPO firms' aftermarket abnormal returns is positively associated with accounting risk. In other words, the precision of market participants' pricing decisions is hypothesized to decrease with accounting risk. Univariate tests of Hypothesis 2 utilize the accounting risk portfolios constructed for testing Hypothesis 1. However, a portfolio's mean abnormal return variance over various holding periods is now the statistic of interest, rather than the crosssectional mean abnormal return employed in the empirical testing of Hypothesis 1. An individual IPO firm's variance of period t abnormal returns is defined as VARit = ARit2, (16) where ARit is the abnormal return over time period t as provided by Equation (12). Then, an accounting risk portfolio's mean abnormal return variance over a specific time period is defined as 17 A twopopulation t test assumes that the data being tested comes from two independent populations. This assumption appears reasonable in this case since there is no a priori reason to hypothesize that high accounting risk and low accounting risk portfolios are not independent. Each of these t tests is the mathematical equivalent of a oneway analysis of variance (ANOVA) since each accounting risk proxy only possesses two levels (i.e., high and low). Even though the statistical tests are equivalent, Neter et al. state that "the t test generally is to be preferred since it can be used to conduct both twosided and one sided tests" (Neter et al. [1985, p. 544]). By contrast, the F test employed in a oneway ANOVA "can only be used for twosided tests" (Neter et al. [1985, p. 544]). Since many of this study's hypotheses are onedirectional, the t test procedure is employed in this research. Table 41 Univariate (Portfolio) Tests of Hypothesis 1 Time High Risk Low Risk Difference Horizon Portfolioa Portfoliob HighLow Year 1 MARic, d MARLld MAR 1MARLje Year 2 MARCH2 MARL2 MAR2 MARL2 Year 3 MARH3 MARL3 MARCH3 MARL3 3 Year Cumulative MARHc MARLC MARc MARLC a When auditor quality is utilized as a proxy for accounting risk, the high risk portfolio consists of IPO firms audited by nonBig Eight auditing firms. For the SEC registration statement form proxy, the high risk portfolio consists of IPOs which filed an S18 registration. Finally, the high accounting risk firms based on the accounting procedure choice proxy are those firms which utilized consistently income increasing (i.e., "liberal") methods for accounting for inventory and depreciation. b When auditor quality is utilized as a proxy for accounting risk, the low risk portfolio consists of IPO firms audited by Big Eight auditors. For the registration statement type proxy, the low risk portfolio consists of IPOs which filed S1 registration statements. Finally, the low accounting risk firms based on the accounting procedure choice proxy are those firms which utilized at least one income decreasing (i.e., "conservative") method for accounting for inventory and depreciation. MARpt portfolio p's mean abnormal return over time horizon t. Thus, MARHi the mean abnormal return of the high accounting risk portfolio over the first year of public trading, MARH2 = the high risk portfolio's second year mean abnormal return, etc. d A twotailed t test is conducted to determine whether each MARt in this column differs significantly from the hypothesized value of zero. e A twotailed twopopulation t test is conducted to determine whether each difference statistic in this column differs significantly from the hypothesized value of zero. N MVARt Z VARit / N, (17) i=1l where MVARt the mean abnormal return variance over time period t of the firms comprising portfolio p, and the remaining variables are as previously defined. This mean portfolio variance figure is computed separately for the first, second, and third years of public trading, as well as over a 36 month cumulative period. Then, the empirical tests of Hypothesis 2 merely involve assessing via twopopulation t tests18 whether MVARPt is significantly greater for high accounting risk portfolios than for low accounting risk portfolios over each of the four time horizons. Table 42 illustrates these tests. Univariate Tests of Hypothesis 3 Hypothesis 3 states that the variance of IPO firms' aftermarket abnormal returns decreases over time. This decrease in abnormal return variance is hypothesized to be attributable to decreasing accounting risk over time as more information becomes known about these newly publicly held firms. Univariate tests of Hypothesis 3 employ the same accounting risk portfolios constructed for the testing of Hypotheses 1 and 2. The empirical tests of the third hypothesis seek to determine whether the mean abnormal return variance of each portfolio decreases over the three year time horizon. Hypothesis 3 implies a monotonic decrease in the 18 See footnote 17. Table 42 Univariate (Portfolio) Tests of Hypothesis 2 Time High Risk Low Risk Difference Horizon Portfolioa Portfoliob HighLow Year 1 MVARHl MVARi MVARHiMVARL Id Year 2 MVARH2 MVARL2 MVARH2 ARL2 Year 3 MVARH3 MVARL3 MVARH3 MVARL3 3 Year Cumulative MVARHC MVARLc MVARHcMVARLc a When auditor quality is utilized as a proxy for accounting risk, the high risk portfolio consists of IPO firms audited by nonBig Eight auditing firms. For the SEC registration statement form proxy, the high risk portfolio consists of IPOs which filed an S18 registration. Finally, the high accounting risk firms based on the accounting procedure choice proxy are those firms which utilized consistently income increasing (i.e., "liberal") methods for accounting for inventory and depreciation. b When auditor quality is utilized as a proxy for accounting risk, the low risk portfolio consists of IPO firms audited by Big Eight auditors. For the registration statement type proxy, the low risk portfolio consists of IPOs which filed S1 registration statements. Finally, the low accounting risk firms based on the accounting procedure choice proxy are those firms which utilized at least one income decreasing (i.e., "conservative") method for accounting for inventory and depreciation. MVARpt portfolio p's mean abnormal return variance over time horizon t. Thus, MVARH, the mean abnormal return variance of the high accounting risk portfolio over the first year of public trading, MVARH2 the high risk portfolio's second year mean abnormal return variance, etc. d A twopopulation t test is conducted to determine whether each difference statistic in this column differs significantly from zero. Hypothesis 2 predicts these differences to be significantly greater than zero. Thus, onesided significance tests are called for in this case. 53 variance over the three years. For example, the hypothesis implies that the mean abnormal return variance of Big Eight audited IPO firms is greater in the first year of trading than in the second year of trading, and that the second year variance exceeds that of the third year. Each of the remaining accounting risk portfolios is predicted to behave in a similar fashion. Even though Hypothesis 3 implies that the mean abnormal return variances decrease monotonically (however not necessarily in a linear fashion) over time, the hypothesis may be testing by examining whether the variances exhibit the presence of a decreasing linear trend over the three year period.19 Dawes and Corrigan [1974] demonstrate the usefulness of linear models in cases such as this. They state that "linear models are good approximations to all multivariate models that are conditionally monotone in each predictor variable" (Dawes and Corrigan [1974, p. 98]). Since an individual IPO firm's abnormal return variance over year 1 is not independent of its return variance in other years, a statistical procedure which accounts for these dependencies is required. An often utilized procedure in cases such as this is a repeated measures ANOVA, where time represents the repeated measure. A test for the presence of a significant linear trend over the time horizon may be conducted in such an analysis. In instances such as this one in which the repeated measure (i.e., time) takes on only three values, the linear 19 Keppel [1982, Chapter 7] provides a detailed description of trend analysis, and provides details of empirical procedures which test for the presence of a significant linear trend. trend test is mathematically equivalent to a test of the difference in the variable of interest (i.e., MVAR) between time period 1 and time period 3.20 In these specific circumstances in which the repeated measure only takes on three possible values, a matchedpair (i.e., paired comparison) t test of the difference between year 1 and year 3 mean abnormal return variances is mathematically equivalent to the repeated measures ANOVA test of linear trend. Since Hypothesis 3 is a onedirectional hypothesis, the t test procedure is implemented since it allows one sided hypothesis tests, whereas the F statistics produced by the ANOVA procedure are solely twosided tests. Table 43 illustrates these tests. Multivariate Tests Multiple Regression Test of Hypothesis 1 Hypothesis 1 may alternatively be tested via the following cross sectional regression models: AR(t)i = a + PiAUDITORj + fi2FORMi + 03APCHOICEi + y1SIZEi + y2LNRISKSi + Y3UE(t)j + ei, (18) 20 Keppel states that "trend analysis requires a different set of special coefficients for each of the orthogonal trend components to be extracted from the data" (Keppel [1982, p. 135]). The special coefficients required to test for a decreasing linear trend when the repeated measure takes on only three values are +1, 0, and 1. Specifically, a portion of the calculated test statistic consists of the sum of +1 times the observation for year 1, 0 times the year 2 observation, and 1 times the year three observation. The resulting statistical test is mathematically equivalent to a test of the difference between period 1 and period 3 observations. Table 43 Univariate (Portfolio) Tests of Hypothesis 3 Difference Portfolio Year 1 Year 2 Year 3 YearlYear3 High Riska MVARlc MVARH2 MVARg3 MVARHiMVARHsd, Low Riskb MVARLI MVARL MVARL3 MVARLi M 4VARL3 a When auditor quality is utilized as a proxy for accounting risk, the high risk portfolio consists of IPO firms audited by nonBig Eight auditing firms. For the SEC registration statement form proxy, the high risk portfolio consists of IPOs which filed an S18 registration. Finally, the high accounting risk firms based on the accounting procedure choice proxy are those firms which utilized consistently income increasing (i.e., "liberal") methods for accounting for inventory and depreciation. b When auditor quality is utilized as a proxy for accounting risk, the low risk portfolio consists of IPO firms audited by Big Eight auditors. For the registration statement type proxy, the low risk portfolio consists of IPOs which filed S1 registration statements. Finally, the low accounting risk firms based on the accounting procedure choice proxy are those firms which utilized at least one income decreasing (i.e., "conservative") method for accounting for inventory and depreciation. SMVARpt portfolio p's mean abnormal return variance over time horizon t. Thus, MVARH, the mean abnormal return variance of the high accounting risk portfolio over the first year of public trading, MVARH2 = the high risk portfolio's second year mean abnormal return variance, etc. d A matchedpair (i.e., paired comparison) t test is conducted to determine whether each difference statistic in this column differs significantly from zero. Hypothesis 3 predicts these differences to be significantly greater than zero. Thus, onesided significance tests are called for in this case. S This matchedpair t test of the difference between the MVAR of years 1 and 3 is mathematically equivalent to a test for the presence of a significant linear trend over time in a repeated measures ANOVA analysis. where AR(t)i firm i's abnormal return over holding period t as generated from Equation (12) (i.e., Ot), AUDITOR1 the quality of auditor utilized in the IPO process, FORMi the registration statement type filed by IPO firm i, APCHOICE1 the level of conservatism surrounding firm i's accounting procedure choices for inventory and depreciation, SIZE1 firm size at the initial offering, LNRISKS1 the natural logarithm of one plus the number of risk factors listed in firm i's prospectus, UE(t)i firm i's unexpected earnings for period t, and ej is an error term. The coefficients a, 01 to 03, and 71 to 73 are regression parameters. In Equation (18), AUDITORJ, FORMi, and APCHOICE1 are empirical proxies for accounting risk, while SIZE1, LNRISKS1, and UE(t)i are included as control variables. The control variables are designed to control for potential differences among firms' abnormal returns which are unrelated to differences in accounting risk. Thus, Equation (18) allows one to test the impact of accounting risk on longterm mispricing, while controlling for potential confounding influences. SIZEi and LNRISKS1 are included in the regression models to control for business risk (i.e., uncertainty related to a firm's production, investment, and financing activities). It is important to control for firm size since it has been empirically documented21 that larger IPO firms are less speculative than small initial offering firms. The number of risk factors supplied by an IPO firm in its prospectus may also proxy for the inherent cash flow riskiness of the firm. The SEC 21 See Beatty [1989b]. requires firms undertaking IPOs to enumerate important "risk factors" which will aid prospective investors in assessing the riskiness of the offering. The SEC may choose to delay an offering through a "deficiency letter" until the firm provides adequate details concerning the riskiness of its operations, its financing obligations, etc. Hence, Beatty [1989b] argues that riskier IPO firms provide greater disclosure in the "risk factors" section of the prospectus than less risky firms. Finally, UE(t)i is included in the regression models to control for the widely documented empirical result that abnormal security returns are positively associated with "unexpected" accounting earnings. Equation (18) is estimated crosssectionally for four different specifications of the dependent variable (i.e., longterm mispricing). Specifically, the equation is separately estimated for abnormal returns calculated over the first, second, and third year of public trading, as well as a 36 month cumulative return. AUDITORj is measured as a dichotomous variable. It is coded 1 in cases where nonBig Eight auditing firms are utilized in the IPO process and 0 for Big Eight firms. FORMi is similarly measured in a categorical fashion. The variable is coded 1 for S18 registrations and 0 for S1 registrations. APCHOICEj is assigned a value of 1 for firms which employ consistently income increasing (i.e., "liberal") inventory and depreciation accounting procedure choices and 0 otherwise. SIZEi is calculated as the natural logarithm of IPO firm i's total book value of assets as provided in the prospectus, and LNRISKSi is defined as the natural logarithm of one plus the number of risk factors listed in firm i's prospectus. 58 Each of the previously described independent variables is measured based on the most recently available data at the time of the IPO (i.e., that available in the prospectus). In other words, these independent variables remain constant across the four crosssectional regressions. Thus, the regressions are intended to determine the impact of accounting risk at the time of the IPO on various specifications of longterm IPO mispricing. Hypothesis 1 predicts zero coefficients on the three accounting risk variables in each regression.22 In contrast to the other independent variables, UE(t)i is measured over the same time horizon as the dependent variable in each of the four crosssectional regressions. Unexpected earnings is calculated using a variant of the commonly utilized random walk earnings expectation model. An alternative specification of the random walk model is required in this study due to the misalignment of the abnormal security return series and the earnings series. Specifically, an IPO's abnormal returns are generated via Equation (12) based on monthly returns beginning on the day following the IPO date. In other words, the return series is based on "event time," with the event in question being the initial offering. By contrast, the annual earnings series is based on each firm's fiscal year end (FYE). Therefore, unless a firm went public on (or very near) its fiscal year end date, the returns and earnings series utilized in this study are not properly aligned. 22 It is easily verified that the model presented in Chapter 3 also predicts nonsignificant coefficients for the business risk control variables (i.e., SIZEi and LNRISKSi) since the abnormal security returns are already riskadjusted. 59 To alleviate this potential misalignment problem, the random walk earnings expectation model is adjusted in order that the earnings series becomes aligned with the security returns series. Specifically, IPO firm i's unexpected earnings for period t is calculated as follows: UE(t)i [(EPSit EPSit1) (Nlit / 12) + (EPSit+1 EPSi) (N2it / 12)] / PRICEit, (19) where EPS signifies annual earnings per share, PRICE designates the security price per share,23 and N1 and N2 are adjustment factors utilized to bring the earnings series into alignment with the returns series. As an example of how Equation (19) is used in the determination of unexpected earnings, the calculation of firm i's unexpected earnings over the first year of public trading (i.e., UElI) will be illustrated. In the calculation of UEli, the time subscript t equals 1 throughout Equation (19). EPSil signifies firm i's annual earnings per share as of the first fiscal year end subsequent to the IPO. For example, if a December 31 FYE firm went public on August 1, 1980, EPSil would represent the firm's earnings per share for the year ended December 31, 1980. The period of time (rounded to the nearest month) which elapses between the IPO date and this first FYE subsequent to the initial offer is designated Nlil months. In the above example, Nlil equals the five month period from August 1, 1980 to December 31, 1980. N2i1 then represents the number of months from this first FYE following the IPO to the date one year subsequent to the IPO (i.e., the end of the first year 23 Both the earnings per share and the price per share series are adjusted for stock dividends and stock splits. of trading in "event time"). In the above example, N2j, equals the seven month period from December 31, 1980 to August 1, 1981. Therefore, Nli1/12 (i.e., 5/12 in the example) of the traditionally calculated random walk based "unexpected earnings" accruing from the FYE preceding the IPO to the first FYE subsequent to the IPO (i.e., from December 31, 1979 to December 31, 1980) is included in the calculation of UEli. Further, N2j/12 (i.e., 7/12 in the example) of the traditionally calculated "unexpected earnings" based on the random walk model accruing from the first FYE following the IPO to the second FYE subsequent to the IPO (i.e., from December 31, 1980 to December 31, 1981) is included in the calculation of UEli. Thus, the N1/12 and N2/12 factors serve to transform the FYE based earnings series into a series compatible with the "event time" returns series. Therefore, as is customary in a random walk annual earnings expectation model, UEli is calculated based on 12 months of earnings. However, this 12 months of earnings represents a weighted average of the "unexpected earnings" pertaining to two fiscal years (i.e., the fiscal years ended December 31, 1980 and December 31, 1981 in the above example).24 Consistent with the widely documented result that unexpected accounting earnings and abnormal security returns are positively associated, the parameter estimate on UE(t)i is predicted to be positive in each of the four crosssectional regressions. 24 The previously described method of estimating unexpected earnings is utilized to determine unexpected earnings over the first, second, and third year of public trading. However, unexpected earnings corresponding to the cumulative three year holding period is calculated as the sum of the undeflated unexpected earnings calculated for each of the first three years deflated by the security price at the end of year three. Multiple Regression Test of Hypothesis 2 Hypothesis 2 may alternatively be examined via the following crosssectional regression models: VAR(t)i a + PIAUDITORj + A2F0ORM + A3APCHOICEi + 1SIZEi + 721!NRISKSi + 73ABSUE(t)i + ej, (20) where ABSUE(t)i the absolute value of firm i's period t unexpected earnings, and all remaining variables are as previously defined. The absolute value specification of unexpected earnings is utilized in Equation (20) since the absolute magnitude (rather than the signed magnitude) of unexpected earnings should impact the variance of returns. As was the case for Equation (18), Equation (20) is estimated separately across all sample firms for four different specifications of the dependent variable (i.e., abnormal return variances over four different time horizons). Specifically, an IPO firm's abnormal return variance is measured over the first, second, and third years of public trading, as well as over the cumulative 36 month period. As previously indicated by Equation (16), an IPO's variance of period t abnormal returns (i.e., VARit or equivalently VAR(t)i) is calculated as the abnormal return over time t squared (i.e., AR 2). According to Hypothesis 2, the estimated coefficients on the accounting risk variables (i.e., 01 to P3) are predicted to be positive for each of the four crosssectional regressions.25 Similarly, the 25 However, Hypothesis 3 predicts that the ability of these variables to explain the variation in abnormal return variances decreases over time. Empirical tests of Hypothesis 3 utilizing a multiple regression approach are detailed in the following section. 62 parameter estimate on the unexpected earnings variable is also expected to be positive. If business risk is adequately controlled for in the return generating process described by Equation (12), the coefficients on the business risk control variables (i.e., SIZEj and LNRISKSj) should not differ significantly from zero. However, 71 and 72 may be significantly different from zero if INDUSTRYit does not adequately control for business risk. In such a case, the coefficient on SIZEit is predicted to be negative (since smaller firms correspond to a higher level of business risk and hence should experience a higher variance of returns), while the coefficient on LNRISKS is predicted to be positive (since the number of risk factors enumerated in the prospectus is positively associated with business risk). Multiple Regression Test of Hypothesis 3 Hypothesis 3 may alternatively be tested via the following pooled crosssectional timeseries regression model: VARit + 6iAUDITORit + fi'AUDITORit*Ilit + 3i' 'AUDITORit*I2it + 2FORMit + /2'FORMit*Ilit + 2' 'FORMit*I2it + 83APCHOICEit + 13'APCHOICEit*Ilit + 03' 'APCHOICEit*I2it + 7iSIZEit + 71'SIZEjt*Iljt + 7y' 'SIZEit*I2jt + 72LNRISKSjt + 72'LNRISKSit*Iljt + 72' 'LNRISKSit*I2it + 73ABSUEjt + 73'ABSUEit*Ilit + 73' ABSUEjt*I2it + eit, (21) where Ilit is an indicator variable which is coded 1 for observations relating to the second year following the IPO and 0 otherwise, I2jt is an indicator variable which takes on the value of 1 for third year 63 observations and 0 otherwise, and all other variables are as previously defined. As stated earlier, Equation (21) is a pooled crosssectional time series model. Specifically, the equation is estimated with three observations per sample firm. These three observations per firm pertain to the first, second, and third year of public trading, respectively. Thus, Equation (21) is estimated with three times as many observations as the separate year by year crosssectional regressions utilized in the testing of Hypotheses 1 and 2. The interaction terms employing the indicator variables llit and 12t are included in the model so that tests concerning changes in the regression coefficients over time may be conducted.26 Specifically, since both the indicator variables are coded 0 for observations relating to the first year following the IPO, the regression coefficients 01 to 03 and 71 to 73 represent the respective coefficients for year 1. Similarly, since 1lit equals 1 for year 2 only and 12it equals 0 for year 2, then the regression coefficients #I' to 03' and 71' to 73' represent the changes in the respective coefficients for year 2 as compared to the base year (i.e., year 1). In a similar fashion, since llit is coded 0 for year 3 and 12it takes on the value 1 only for year 3, then the regression coefficients 81' to 083'' and yj'' to 73'' represent changes in the respective coefficients for year 3 in comparison to year 1. 26 It should be noted that the model specified in Equation (21) assumes that the intercept does not vary across sample years but that the slopes differ according to whether the first, second, or third year of public trading is being examined. Hypothesis 3 implies that the accounting risk variables explain less of the variation in abnormal return variances as more information is revealed as time progresses from the IPO date. Thus, the 01' to 03' and 1' to P3'' coefficients predicted to be negative.27 Even though the coefficients on the interaction terms in Equation (21) represent differences in the coefficients for year 2 and year 3, respectively, from the coefficients pertaining to year 1, they provide no indication of the change in the parameters from year 2 to year 3. The change in the regression coefficients from year 2 to year 3 is tested by examining the significance levels of the following linear hypotheses regarding the coefficients: 01' i1'' 0, 02' P2'' 0, 03' 03' 0, 71' 7' 0'' 72' 72'' 0, and 73' 73'' 0. Each of these tests seeks to determine whether a particular year 2 regression 27 A priori, it is difficult to predict the sign of the coefficients on the interaction terms which contain the control variables. On the one hand, it could be postulated that the business risk control variables (i.e., SIZEi and LNRISKSj) will decrease in significance over time since these variables are measured as of the IPO date and remain constant across the three time periods examined. However, it could also be hypothesized that the business risk proxies will increase in significance as the accounting risk variables lose explanatory power. Similarly, one could reasonably argue that the unexpected earnings control variable could either increase, decrease, or remain constant over time. Specifically, it can be hypothesized that unexpected earnings will increase in explanatory power over time as the accounting risk variables decrease in importance. However, it could also be argued that the unexpected earnings variables will decrease in explanatory power over time since earnings may play a greater role in firm valuation early in the life of a public firm (i.e., before other competing sources of information become readily available). Finally, one could also reasonably argue that the unexpected earnings variable should not vary in significance over time since the relationship of earnings and returns should not change significantly from year to year. Therefore, predictions are not made relating to the hypothesized signs of the 71' to 73' and y7' to 73'' coefficients. coefficient is significantly greater than the corresponding year 3 coefficient.28 Summary of the Empirical Methodology This chapter describes both univariate and multivariate approaches employed in the empirical testing of Hypotheses 1 through 3. The univariate tests allow a straightforward assessment of the three hypotheses. The univariate (portfolio) approach is particularly useful in the examination of Hypothesis 1 since it permits a direct determination of whether mean abnormal returns differ significantly from zero for (and across) portfolios of IPOs which differ according to accounting risk. However, the portfolio approach suffers from the limitation that it is univariate in nature. In other words, the portfolio approach is limited in that the analysis must be conducted separately for portfolios based on each of the three accounting risk proxies. Thus, the approach does not allow an assessment of the incremental impact of various measures of accounting risk on longterm IPO mispricing. A potentially more important limitation is that the univariate tests do not control for factors other than accounting risk which may influence mean aftermarket abnormal returns and mean abnormal return variances. Thus, it is not possible to conclude from these 28 According to Hypothesis 3, the regression coefficients on the three accounting risk variables should be greater in year 2 than in year 3. Thus, the differences in the estimated year 2 and year 3 regression parameters pertaining to AUDITOR, FORM, and APCHOICE are expected to be significantly greater than zero. As explained in footnote 26, no predictions are made in regards to the signs of the differences between the regression coefficients of years 2 and 3 for the .control variables (i.e., SIZE, LNRISKS, and ABSUE). 66 univariate tests that it is accounting risk that is responsible for any observed empirical relationships, rather than some other factor such as IPO firm size or inherent cash flow riskiness. In contrast, the multiple regression techniques are multivariate analyses which allow the determination of the incremental effects of this study's three measures of accounting risk on longterm mispricing. The regression approach also allows for the controlling of potentially confounding influences on the relationships of interest. The technique also allows a direct assessment of Hypothesis 3, the change in explanatory power of the accounting risk variables over time. However, the regression method provides only an indirect test of the validity of Hypothesis 1. Specifically, while the regression technique indicates whether accounting risk and abnormal returns are significantly correlated, it provides no indication of whether IPOs are mispriced on average. Thus, there are advantages and disadvantages associated with both the univariate (portfolio) and multivariate (regression) approaches. Due to the complementary nature of the two approaches, both are employed in the testing of the three hypotheses. CHAPTER 5 DATA, SAMPLE SELECTION CRITERIA, AND RESULTS Data Sources and Sample Selection Criteria The major source of data employed in this study consists of a collection of microfiche IPO registration statements (which include preliminary prospectuses) which were obtained from Disclosure Inc.1 This database contains registration statements for a number of firm commitment2 IPOs of nonfinancial firms which occurred between January 1980 and December 1983. The database contains registration statements and preliminary prospectuses for 511 firms which made initial public offerings of equity securities (unit offerings which include warrants are excluded) during this 19801983 time period. Only firms for which at least three years of security price data are available on the CRSP 1 I wish to thank Professor Chris James for allowing me access to these registration statements. Please see James and Wier [1989] for a complete description of the criteria utilized in assembling this microfiche database. 2 The two major types of underwriting contracts utilized in the going public process are "firm commitment" and "best efforts" contracts. In a firm commitment offering, the underwriter purchases the entire security issue from the issuer and then must resell the shares to investors. Thus, the underwriter makes a "firm commitment" to purchase the offering at a specified price (i.e., the IPO price less the underwriter's discount), and bears the risk of not being able to sell the entire issue at the IPO price. Conversely, in a best efforts offering, the underwriter's only commitment is to make his "best efforts" to sell the issuer's security offering to investors. In this case, the issuer bears the risk of not being able to sell the entire issue at the IPO price. NASDAQ file are included in the final sample.3 This three year of security return availability criterion is the only criterion utilized in the sample selection process and results in the exclusion of 139 firms. Thus, the final sample consists of 372 firms. Table 51 summarizes the sample selection process employed in this study. Table 51 Sample Selection Procedures Total Available Microfiche Prospectuses of NonUnit IPOs 511 Number of Firms Removed Due to a Lack of 3 Years of Data on the 1988 Daily CRSP NASDAQ File 139 Final Sample 372 Much of the data required for this study was gathered from this collection of microfiche registration statements. For example, the data required to calculate the AUDITOR, FORM, and SIZE variables have been collected from these registration statements and accompanying prospectuses. Further, this microfiche collection was utilized in conjunction with the Compustat FullCoverage file in the collection of the sample firms' accounting procedure choices for inventory and depreciation. Another significant portion of the required data was gathered from "Ritter's 19751984 IPO Database."4 This database contains extensive 3 The entire empirical analysis was repeated using a more restrictive five years of CRSP NASDAQ data availability criterion. This more restrictive criterion resulted in a final sample of 299 firms. Summary results of this additional analysis are reported in Appendix B. 4 I wish to express thanks to Professor Jay Ritter for providing me this extensive amount of data on IPO firms. Please see Ritter [1984b] and Beatty and Ritter [1986] for detailed descriptions of the criteria utilized in building this database. 69 detail on 2,567 initial public offerings which occurred between 1975 and 1984. Each of the 372 sample initial offerings is contained in Ritter's IPO database. This database was utilized to collect information concerning the number of risk factors listed in an IPO's prospectus, the date of the initial offering, the IPO price, the first available closing bid price following the IPO, the number of shares sold in the offering, and the year that the firm was first organized. Thus, this data allowed the calculation of the LNRISKS variable, the amount of shortterm underpricing, the gross proceeds of the offering, and the age of the firm at the time of the IPO. Descriptive statistics on these and other variables will be provided in the following section. The required security returns data were collected from the CRSP NASDAQ file. Specifically, the NASDAQ daily returns file was utilized in order to gather raw returns for each sample firm, the returns on the NASDAQ Composite Index, and the returns required in the calculation of the industry return variable. Finally, the information required to calculate the unexpected earnings variables was gathered from the COMPUSTAT FullCoverage file in conjunction with Standard and Poor's Corporation Records. Descriptive Statistics Sample Firm and Offering Characteristics Table 52 provides descriptive statistics relating to certain characteristics of the 372 sample firms and their respective initial offerings. Panel A provides information about the sample as a whole, Table 52 Descriptive Statistics Panel A Sample Firm and Initial Offering Descriptive Statistics Variable Mean Median Standard Deviation Nb PROCEEDS 15.62 9.06 18.34 372 BKASSETS 15.56 6.92 25.78 372 SIZE 8.66 8.84 1.65 372 RISKS 4.98 0.00 6.50 366 LNRISKS 1.09 0.00 1.22 366 AGE 10.80 7.00 13.57 370 Panel B Summary Firm and Offerings Statistics By Year of Offering Mean Mean Year Number of Offers PROCEEDS BKASSETS 1980 35 10.11 13.89 1981 114 10.81 12.31 1982 25 12.70 12.01 1983 198 19.72 18.18 a The variables examined in this table relate to certain characteristics of the 372 sample firms and their initial offerings. The variables of interest are defined as follows: PROCEEDS the gross proceeds of the initial offering measured in millions of dollars. Gross proceeds are calculated as the actual number of shares sold in the offering multiplied by the offering price per share. BKASSETS the book value of total assets (measured in millions of dollars) prior to the IPO. This variable is collected from the latest audited balance sheet supplied in the prospectus. SIZE = the natural logarithm of BKASSETS. RISKS the number of risk factors listed in the prospectus. LNRISKS the natural logarithm of (1 + RISKS). AGE the age of the firm in years prior to the initial offering. b N = the number of firms involved in the calculation of the mean, median, and standard deviation statistics. while Panel B provides descriptive statistics by year of initial offering. Panel A demonstrates that the average offering gross proceeds (i.e., PROCEEDS) raised by the 372 sample firms equals $15.62 million. This very closely approximates the sample firms' mean book value of total assets (i.e., BKASSETS) prior to the offering. Thus, on average, the sample firms raised an amount approximately equal to their currently prevailing book value of assets via an IPO. Panel A also reveals that the mean (median) number of risk factors (i.e., RISKS) enumerated in the offering prospectus is approximately equal to 5 (0).5 One other variable reported in Panel A of Table 52 merits mention. The average age of the sample firms prior to the IPO is slightly less than 11 years. Thus, it is apparent that the sample is not made up entirely of startup firms (i.e., firms with less than two years of operations). Panel B reports the number of offers, the mean gross proceeds, and the mean book value of assets (prior to the IPO) by sample year. It is apparent that the number of offers is not spread evenly across the sample years. In fact, the sample is dominated by firms undertaking IPOs in 1981 and 1983. This is consistent with evidence provided in 5 Panel A also indicates that the logarithmic transformations of BKASSETS and RISKS which are employed as business risk control variables (i.e., SIZE and LNRISKS, respectively) in the crosssectional regressions reported later have greatly reduced standard deviations in comparison to the original variables. However, even though these logarithmic transformations are utilized in the regression models rather than the original variables, the error terms of the various regressions are heteroscedastic. Later sections will provide (1) details of the severity of the heteroscedasticity and (2) the means utilized in correcting the problems caused by heteroscedastic error terms. Ibbotson et al. [1988] that more initial offers were undertaken in the years 1981 and 1983 than in 1980 or 1982. Thus, the makeup of the sample does not appear to differ significantly from the population as a whole. Panel B also illustrates that the average offering gross proceeds increases over the sample period. The same relationship also generally holds for the sample firms' book value of assets prior to the offering. LongTerm Mispricing. ShortTerm Mispricing, and Unexpected Earnings Table 53 provides additional descriptive statistics relating to various measures of longterm and shortterm IPO mispricing, as well as to the unexpected earnings variables employed in this study. The mean shortterm underpricing (i.e., UP) of the 372 sample firms is 11.70%. As is evidenced by Table 21, this 11.70% level of underpricing approximates findings in other research. Descriptive statistics pertaining to the sample firms' risk adjusted abnormal returns over the first, second, and third year of public trading, as well as a 36 month cumulative return are provided in Table 53. These measures of longterm IPO mispricing are generated from the return generating process described in Equation (12), and are designated as AR1, AR2, AR3, and CAR36 respectively. The table indicates that only AR1 (i.e., first year abnormal returns) exceeds 1% (in absolute value) on average. Further, AR1 is the only one of the four riskadjusted longterm mispricing measures which differs significantly from zero at conventional significance levels, with a Table 53 Descriptive Statistics Relating to Measures of LongTerm and ShortTerm Mispricing, and Unexpected Earnings Variable Mean Median Standard Deviation Nb UP 11.70% 2.78% 0.27 372 AR1 1.21% 1.49% 0.05 372 AR2 0.35% 0.29% 0.05 372 AR3 0.35% 0.63% 0.05 372 CAR36 0.50% 0.39% 0.08 372 RIMRM1 2.12% 0.30% 0.59 372 RIMRM2 4.19% 3.15% 0.61 372 RIMRM3 6.35% 6.90% 0.59 372 RIMRM36 8.42% 10.95% 0.97 372 UE1 0.44 0.00 8.29 371 UE2 0.04 0.01 0.11 366 UE3 0.01 0.00 0.39 337 UE36 0.95 0.02 15.05 337 a The variables examined in this table relate to various measures of shortterm and longterm IPO mispricing, as well as measures of unexpected accounting earnings for the 372 sample firms. The variables of interest are defined as follows: UP shortterm underpricing of the initial offering price by the underwriter. UP is calculated as: (PlPIPO)/PIPO, where PI refers to the first available closing bid price following the IPO and PIPO equals the initial offering price set by the underwriter. AR1, AR2, and AR3 represent the riskadjusted abnormal returns as generated from Equation (12) for the first, second, and third years of public trading, respectively. CAR36 = the cumulative riskadjusted abnormal return over the first 36 months of public trading as generated from Equation (12). CAR36 is defined as the sum of ARI, AR2, and AR3. RIMRM1, RIMRM2, and RIMRM3 represent market adjusted security returns over the first, second, and third years of public trading, respectively. These variables are calculated by subtracting the return on the NASDAQ Composite (Market) Index from an individual IPO's raw security returns over the appropriate time periods. RIMRM36 the sum of RIMRMI, RIMRM2, and RIMRM3. Thus, the variable represents the cumulative marketadjusted return over the first 36 months of trading. UE1, UE2, UE3, and UE36 are defined as unexpected accounting earnings over the first year, second year, third year, and 36 month cumulative time horizon, respectively. b N = the number of firms involved in the calculation of the mean, median, and standard deviation statistics. 74 PValue of less than .01. Thus, the assertion of Hypothesis 1 that mean abnormal returns should not differ from zero is generally supported for the sample as a whole. However, the hypothesis is not supported for the first year of public trading time horizon.6 Table 53 also lists alternative measures of longterm mispricing based on simple marketadjusted (rather than riskadjusted) returns. Specifically, RIMRMl to RIMRM3 and RIMRM36 represent marketadjusted security returns over the first, second, and third years of public trading, as well as a 36 month cumulative marketadjusted return. The market adjustments consist of subtracting the return on the NASDAQ Composite Index from an IPO's unadjusted raw return over the corresponding time period in the construction of monthly returns. Then, yearly marketadjusted returns are constructed from these monthly returns via an additive process. Table 53 shows that the absolute magnitudes of these marketadjusted returns are much greater than those of the riskadjusted returns previously described. However, only RIMRM3 is significantly different from zero at the .05 level, while the 36 month cumulative return is marginally significant at .10.7 6 More formal statistical tests of Hypothesis 1 will be provided in later sections. 7 Marketadjusted returns are employed by Ritter [1989] in the examination of longterm IPO mispricing. Ritter finds average market adjusted returns of 24.33% over the first three years of trading. This compares with the 8.42% cumulative return of the sample examined in this study. One potential reason for the much more negative mean marketadjusted returns of Ritter's sample firms is that the market index employed differs between the two studies. Ritter uses an index based on NYSE and AMEX firms, while the current study uses the NASDAQ Composite Index. Average marketadjusted returns for the current sample were also estimated using the market index employed by Ritter (i.e., the CRSP ValueWeighted Index based on NYSE and AMEX firms). The results of such an estimation process indicate three year cumulative mean market 75 Empirical tests of Hypotheses 1 through 3 were performed utilizing both the riskadjusted and marketadjusted definitions of longterm mispricing, with virtually no qualitative differences in results.8 Hence, only the results pertaining to the riskadjusted measures are reported in subsequent sections. Unexpected accounting earnings over the first three years of public trading, as well a 36 month cumulative unexpected earnings measure are also provided in Table 53. With the exception of year 2 which has a PValue of less than .01, the mean unexpected earnings measures do not differ significantly from zero. Makeup of Accounting Risk Portfolios Table 54 reports the number of firms comprising each accounting risk category (portfolio). For example, the table illustrates that 294 out of the 372 firms employed Big Eight auditing firms in the going public process. Thus, the low accounting risk portfolio (based on auditor quality) consists of 294 firms. Conversely, the high accounting risk portfolio consists of the 78 firms which utilized nonBig Eight auditors. Thus, the sample is dominated by Big Eight audited IPO firms. adjusted returns of 25.14%. Thus, it appears that the differences between the marketadjusted returns reported by Ritter and those reported in Table 53 of the current study are largely due to differences in the market indices employed in the two studies. 8 The Pearson correlation coefficients between the corresponding marketadjusted and riskadjusted longterm mispricing measures are in the .75.95 range. Thus, the two methods of calculating longterm mispricing appear to be capturing the same phenomena. Table 54 Number of Sample Firms Comprising Each Accounting Risk Category Number of Firms in Number of Firms in High Accounting Risk Low Accounting Risk Accounting Risk Proxy Categorya Category AUDITOR 78 294 FORM 78 294 APCHOICE 264 95 a When auditor quality is utilized as a proxy for accounting risk, the high risk portfolio consists of IPO firms audited by nonBig Eight auditing firms. For the SEC registration statement form proxy, the high risk portfolio consists of IPOs which filed an S18 registration. Finally, the high accounting risk firms based on the accounting procedure choice proxy are those firms which utilized consistently income increasing (i.e., "liberal") methods for accounting for inventory and depreciation. b When auditor quality is utilized as a proxy for accounting risk, the low risk portfolio consists of IPO firms audited by Big Eight auditors. For the registration statement type proxy, the low risk portfolio consists of IPOs which filed S1 registration statements. Finally, the low accounting risk firms based on the accounting procedure choice proxy are those firms which utilized at least one income decreasing (i.e., "conservative") method for accounting for inventory and depreciation. Similarly, Table 54 shows that the sample is dominated by IPO firms which filed S1 SEC registration statements. Again, 294 out of the 372 sample firms are classified in the low accounting risk category (based on the SEC registration statement type filed), and only 78 are classified as high accounting risk firms. However, when accounting procedure choices are utilized in the classification of high versus low accounting risk sample firms, the sample is dominated by high risk firms. Specifically, 264 out of the 359 sample firms which had data available on the accounting methods chosen utilized consistently income increasing (i.e., "liberal") accounting procedures for inventory and depreciation. By contrast, only 95 out of these 359 firms employed at least one income reducing method. Therefore, the common practice among the sample firms was to utilize straightline depreciation and an inventory method other than LIFO for financial reporting purposes. Results of Univariate Tests Results of Univariate Tests of Hypothesis 1 Hypothesis 1 predicts that the mean abnormal return (MAR) of each accounting risk portfolio will not significantly deviate from zero over any of the four time horizons. Further, the hypothesis states that the mean abnormal return of the high risk portfolios should not differ significantly from the MAR of the low risk portfolios over each time period examined. Table 55 provides the results of the univariate (portfolio) tests of Hypothesis 1. Panels A, B, and C supply the Table 55 Univariate (Portfolio) Tests of Hypothesis 1 Time Horizon Year 1 Year 2 Year 3 3 Year Cumulative Time Horizon Year 1 Year 2 Year 3 3 Year Cumulative Panel A Accounting Risk Proxy Variable AUDITOR High Risk Low Risk Portfolio MARa Portfolio MAR 0.94% 1.28%** 0.48% 0.57%* 0.56% 0.30% 0.86% 0.41% Panel B Accounting Risk Proxy Variable FORM High Risk Low Risk Portfolio MAR Portfolio MAR 1.53% 1.12%** 0.09% 0.47% 0.35% 0.35% 1.27% 0.30% Difference HighLow 0.34% 1.05% 0.26% 0.45% Difference HighLow 0.41% 0.56% 0.00% 0.97% Panel C Accounting Risk Proxy Variable = APCHOICE Time High Risk Low Risk Horizon Portfolio MAR Portfolio MAR Year 1 1.02%** 1.12%* Year 2 0.26% 0.08% Year 3 0.48% 0.40% 3 Year Cumulative 0.29% 0.64% a MAR signifies mean abnormal return. ** significant at .01 level, based on twotailed tests * significant at .05 level, based on twotailed tests Difference HighLow 0.10% 0.18% 0.08% 0.35% results of the univariate tests when AUDITOR, FORM, and APCHOICE, respectively, are employed as proxies for accounting risk. Panel A indicates that the mean abnormal return of the high accounting risk portfolio based on auditor quality (i.e., nonBig Eight audited IPO firms) does not differ significantly from the hypothesized value of zero over any of the four time horizons. By contrast, the mean abnormal return of the low accounting risk auditor portfolio (i.e., Big Eight audited IPOs) is significantly greater than zero in year 1 and significantly less than zero in year 2. However, the low risk portfolio's MAR conforms to the predictions of Hypothesis 1 in year 3 and over the 36 month cumulative period. Finally, consistent with Hypothesis 1, the differences between the mean abnormal returns of the high risk and low risk portfolios do not differ significantly from zero for any of the time periods studied. Thus, the prediction of Hypothesis 1 that accounting risk should not influence mean abnormal returns is supported for the case of the auditor quality accounting risk surrogate. Panel B provides similar results for the SEC registration statement type accounting risk proxy. Again, the hypothesis that the MAR of the high risk portfolio (i.e., IPOs which filed S18 registrations) equals zero in each time period cannot be rejected at conventional levels. The hypothesis can be rejected for the low risk group (i.e., Form S1 filers) only for the first year of public trading, when the mean abnormal return is significantly greater than zero. However, the differences between the MARs of the high and low risk portfolios do not differ significantly from the hypothesized value of 80 zero. As was the case for the AUDITOR proxy, Hypothesis 1 is supported when FORM is utilized as a proxy for accounting risk. Panel C illustrates that Hypothesis 1 is also supported when APCHOICE is utilized as a proxy for accounting riskiness. Specifically, the mean abnormal returns of both the high risk (i.e., firms utilizing consistently "liberal" accounting methods for inventory and depreciation) and low risk (i.e., firms utilizing at least one "conservative" method) groups differ significantly from zero in only the first year of public trading. Further, consistent with the predictions of Hypothesis 1, none of the MAR differences between the accounting risk groups differs significantly from zero. In summary, the results reported in Table 55 strongly support the prediction of Hypothesis 1 that IPO firms' mean abnormal returns do not vary according to the level of accounting risk faced by potential investors, since none of the difference statistics in the table differs significantly from zero. Thus, the results are consistent with the hypothesis that investors do not systematically misprice IPOs because of excessive accounting riskiness. However, Table 55 also reveals that in certain isolated cases a given portfolio's MAR differs from the zero hypothesized value, even though none of the portfolios possess statistically significant 36 month cumulative mean abnormal returns. Therefore, Hypothesis l's prediction of zero abnormal returns over all time horizons for each portfolio (i.e., no systematic mispricing of IPOs by investors) is not uniformly supported even though the MARs do not differ significantly from zero in the vast majority of cases. Results of Univariate Tests of Hypothesis 2 Hypothesis 2 predicts that the mean abnormal return variance (MVAR) of high accounting risk firms will exceed that of low accounting risk firms in each of the time horizons examined. Panels A, B, and C of Table 56 report the results of examining Hypothesis 2 using the AUDITOR, FORM, and APCHOICE accounting risk proxy variables, respectively.9 Panel A reveals that when AUDITOR is utilized in the classification of firms into high and low accounting risk categories, the mean abnormal return variance of the high accounting risk group exceeds that of the low risk group in each of the first three years. The difference is significant at the .05 level only for year 2. However, Hypothesis 2 is marginally supported in years 1 and 3. Specifically, the differences in mean abnormal return variances are significant at the .06 and .13 levels in years 1 and 3, respectively. Thus, Hypothesis 2 is supported in year 2 and marginally supported in years 1 and 3 when auditor quality is utilized as a surrogate for accounting risk. Panel B demonstrates that Hypothesis 2 is supported in years 1 and 2, as well as over the cumulative 36 month period, when the FORM accounting risk proxy variable is utilized to classify firms into high 9 It is important to note that in Table 56 a given portfolio's cumulative 36 month MVAR does not equal the sum of year 1 through year 3 MVARs. As indicated by Equation (16), abnormal return variances over a given time period are calculated as the abnormal return over that same time period squared. Therefore, the cumulative abnormal return variance equals the cumulative abnormal return squared. Hence, for each firm (and portfolio) VAR36 = CAR362 rather than AR12 + AR22 + AR32. Time Horizon Year 1 Year 2 Year 3 3 Year Cumulative Time Horizon Year 1 Year 2 Year 3 3 Year Cumulative Table 56 Univariate (Portfolio) Tests of Hypothesis 2 Panel A Accounting Risk Proxy Variable AUDITOR High Risk Low Risk Portfolio MVAR Portfolio MVAR 0.00347 0.00240 0.00313 0.00222 0.00254 0.00207 0.00642 0.00649 Panel B Accounting Risk Proxy Variable FORM High Risk Low Risk Portfolio MVAR Portfolio MVAR 0.00501 0.00199 0.00436 0.00190 0.00229 0.00213 0.01013 0.00550 Difference HighLow 0.00107 0.00091* 0.00047 0.00007 Difference HighLow 0.00302** 0.00246** 0.00016 0.00463** Panel C Accounting Risk Proxy Variable APCHOICE Time High Risk Low Risk Horizon Portfolio MVAR Portfolio MVAR Year 1 0.00212 0.00279 Year 2 0.00250 0.00175 Year 3 0.00237 0.00136 3 Year Cumulative 0.00614 0.00645 a MVAR signifies mean abnormal return variance. ** significant at .01 level, based on onetailed tests * significant at .05 level, based on onetailed tests Difference HighLow 0.00067 0.00075* 0.00101** 0.00031 versus low accounting risk portfolios. Specifically, the MVAR of the high risk firms exceeds that of the low risk firms in each year, and this difference is statistically significant in all but the third year. Thus, Hypothesis 2 is strongly supported when FORM is used as a surrogate for accounting risk. Panel C provides evidence that supports Hypothesis 2 for only two of the four time horizons when APCHOICE is employed to proxy for accounting riskiness. Specifically, the high risk group's mean abnormal return variance exceeds that of the low risk group at conventional significance levels for years 2 and 3. In contrast, the MVAR difference over year 1 and over the 36 month cumulative period is insignificantly different from zero. Thus, Hypothesis 2 can be supported for only years 2 and 3 when APCHOICE is used to proxy for accounting risk. In summary, the univariate tests of Hypothesis 2 reveal that the hypothesis can be supported very strongly for the FORM proxy, and can be marginally supported for the other two proxy variables. Therefore, the results generally support the prediction of Hypothesis 2 that accounting risk and the variance of IPO firms' aftermarket abnormal returns are positively correlated. Hence, it appears that accounting risk influences the absolute magnitude of longterm IPO mispricing, as predicted by Hypothesis 2. Results of Univariate Tests of Hypothesis 3 Hypothesis 3 states that the variance of IPO firms' abnormal returns will decrease over time subsequent to the IPO. This reduction in variance is attributed to a reduction in accounting risk over time as more information is learned about these newly public firms. Table 57 reveals the results of univariate (portfolio) tests of this hypothesis. Panels A, B, and C, respectively, provide the results obtained when AUDITOR, FORM, and APCHOICE are used to proxy for accounting risk. Panel A discloses that, consistent with the prediction of Hypothesis 3, the mean abnormal return variance of the high accounting risk firms (based on auditor quality) in year 1 exceeds that of the same firms' MVAR in year 3. However, this difference is not statistically significant. The same relationship holds true for the low accounting risk firms. Thus, Hypothesis 3 is not supported for auditor quality based accounting risk portfolios. By contrast, Panel B reports a statistically significant decrease in the high accounting risk (based on the FORM proxy) portfolio's MVAR from year 1 to year 3. Thus, a significant decreasing linear trend over the three year time horizon is documented for the high accounting risk portfolio's MVAR, consistent with the prediction of Hypothesis 3. However, an insignificant difference between the low accounting risk portfolio's year 1 and year 3 MVAR is documented. Thus, Hypothesis 3 is supported only for the high accounting risk firms when the SEC registration statement type proxy variable is employed. Finally, Panel C demonstrates empirical support for Hypothesis 3 only for the low accounting risk (based on APCHOICE) firms. The high accounting risk portfolio's MVAR is demonstrated not to decrease over the year 1 to year 3 time period. Thus, inconsistent results are found when the accounting procedure choice proxy variable is used to test Hypothesis 3. Table 57 Univariate (Portfolio) Tests of Hypothesis 3 Accounting Year 1 MVARa 0.00347 0.00240 Panel A Risk Proxy Variable AUDITOR Year 2 Year 3 MVAR MVAR 0.00313 0.00254 0.00222 0.00207 Difference YearlYear3 0.00093 0.00033 Accounting Year 1 MVAR 0.00501 0.00199 Panel B Risk Proxy Year 2 MVAR 0.00436 0.00190 Variable FORM Year 3 MVAR 0.00229 0.00213 Difference YearlYear3 0.00272** 0.00014 Accounting Year 1 Panel C Risk Proxy Variable APCHOICE Year 2 Year 3 Portfolio MVAR MVAR MVAR High Risk 0.00212 0.00250 0.00237 Low Risk 0.00279 0.00175 0.00136 a MVAR signifies mean abnormal return variance. ** significant at .01 level, based on onetailed tests * significant at .05 level, based on onetailed tests Difference YearlYear3 0.00025 0.00143* Portfolio High Risk Low Risk Portfolio High Risk Low Risk In summary, Hypothesis 3 cannot be strongly supported based on these univariate tests. When AUDITOR is used to surrogate for accounting risk, no significant differences are documented. When the other two proxies are analyzed, inconsistent results are found. Specifically, for both the FORM and APCHOICE proxies, the results pertaining to one of the accounting risk portfolios support Hypothesis 3, while the other portfolio provides insignificant results. Hence, the results reported in Table 57 do not provide convincing evidence that IPO firms' abnormal return variances decrease over the first three years of public trading. One possible reason for these results may relate to the short time horizon examined in this study. Specifically, it is possible that the effect predicted by Hypothesis 3 may occur gradually over time, in which case it would not be detected over a short (e.g., three year) time horizon.10 Summary of Univariate Tests The results of the univariate tests indicate that Hypothesis 1 can be strongly supported for each accounting risk proxy variable. Hypothesis 2 is strongly supported for the FORM proxy and marginally supported for the AUDITOR and APCHOICE surrogates. Finally, the empirical results support Hypothesis 3 only in isolated instances. As is implied by the name "univariate tests," these tests are univariate in 10 Appendix B reports the results of testing Hypothesis 3 over a longer five year time horizon. Over this longer time horizon, univariate test results are documented which strongly support Hypothesis 3. Thus, there is some indication that the effect predicted by Hypothesis 3 may only be empirically verified over a long time horizon. nature and the results may not hold in a multivariate setting. In other words, the results attributed to accounting risk in this section may actually be the result of another factor (e.g., IPO firm size or business riskiness) not explicitly controlled for in these univariate analyses. The next section describes the results of multivariate examinations of the three hypotheses. Results of Multivariate Tests Results of Multivariate Tests of Hypothesis 1 The multivariate tests of Hypothesis 1 consist of estimating the crosssectional multiple regression model described by Equation (18) over four separate time horizons. Table 58 provides evidence of the correlations among the independent variables utilized in the various specifications of the model.11 The table indicates that a number of the independent variables are correlated in a statistically significant sense. However, only three of the correlations (excluding correlations among the unexpected earnings variables) exceed .50 in absolute value, and various model specification checks (to be described in more detail later) indicated no serious multicollinearity problems among the independent variables. Table 59 provides details of the estimation of Equation (18) over the four different time horizons. Specifically, Panels A, B, C, and D report the results of estimating the model over the first year, second 11 It should be noted that the correlations among the four measures of unexpected earnings are not of concern since only one of these variables is utilized in each of the four estimated models. O r CD n0 v ,i' 0 %en CM1 cn 0c~, cn r. 0 cn 4 M *en 0 I. cl1 OCY Oir Ou, 0 cn cn ocn * (" Ln r 0 ., o r *C.1 CM r 0 f1 *(T eel *" 0 Lf"O 4 Lf 0,c 0,.4 o r * %.i *c") Sc *Sc 4,c 0 0 r 'Sc cn 'Sc .N U 0 0n c 0 .ii .11 4) v ca 41 w ,r4 u S0 0 I=4 ,.4 4 .1 '1 >.o 0 .14 blo 4.J a1)W 0 0 10 co 0 0 *Xc * (N* 0v en o 0 i A cc c i *SCM CM4 r~~ *en c oa 0 0 4J 4.J 0 0 (a cc 44 0 0 u 0 4q4 4 44 0 0 4 4)
.C4 .rl v,4X rc1 ca 'ScSc 'Sc Table 59 Regression Approach to Testing Hypothesis 1 Model: ARIi a Panel A )i1AUDITORi + 02FORMi + 63APCHOICEi 71SIZEi + 72LNRISKSi + 3UEli + ei t Statistic 3.275** 0.139 1.092 0.243 2.613** 3.030** 6.458** Heteroscedasticity Adjusted X2 Test Statistic 5.277# 0.017 0.882 0.054 3.414 8.760## 16.456## Test of the Model Goodness of Fit: FValue8.535 PValue=.0001 White's [1980] Homoscedasticity Test: x2Value183.59 PValue.000l Model Adjusted R2 .1141 N352 Panel B Model: AR2i a + 01AUDITORi + P2FORMi + 83APCHOICEi + y1SIZEi + 72LNRISKSi + 73UE2i + ei Standard Error .021 .006 .007 .006 .002 .003 .022 t Statistic 0.539 1.565 0.981 0.600 0.577 1.366 6.936** Heteroscedasticity Adjusted X2 Test Statistic 0.321 2.078 0.632 0.511 0.364 2.041 33.596## Test of the Model Goodness of Fit: FValue9.441 PValue.000l White's [1980] Homoscedasticity Test: x2Value281.08 PValue=.000l Model Adjusted R2 = .1277 N=347 Variable INTERCEPT AUDITOR FORM APCHOICE SIZE LNRISKS UE1 Parameter Estimate .067 .001 .008 .001 .005 .008 .187 Standard Error .020 .006 .007 .005 .002 .003 .029 Variable INTERCEPT AUDITOR FORM APCHOICE SIZE LNRISKS UE2 Parameter Estimate .011 .010 .007 .003 .001 .004 150 Table 59continued Model: AR31 = a Panel C + fi1AUDITORi + 02FORMi + P3APCHOICEi + 7y1SIZEi + y2LNRISKSi + fy3UE3i + ei Standard Error .022 .007 .008 .006 .002 .003 .007 t Statistic 1.113 0.486 0.732 0.011 1.281 0.374 1.217 Heteroscedasticity Adjusted x2 Test Statistic 1.483 0.213 0.487 0.000 1.983 0.116 0.460 Test of the Model Goodness of Fit: FValue0.542 PValue=.7759 White's [1980] Homoscedasticity Test: x2Value195.85 PValue.000l Model Adjusted R2 .0087 N318 Model: CAR36i  Panel D a + PIAUDITORi + 02FORMi + 03APCHOICEi + y1SIZEi + y2LNRISKSi + y3UE36i + ei Standard Error .035 .011 .012 .010 .003 .005 .007 t Statistic 1.713 0.031 0.539 0.232 1.492 1.088 6.061** Heteroscedasticity Adjusted X2 Test Statistic 2.007 0.001 0.256 0.054 1.537 1.246 12.586## Test of the Model Goodness of Fit: FValue6.624 PValue=.0001 White's [1980] Homoscedasticity Test: X2Value298.73 PValue=.O00l Model Adjusted R2 .0962 N=318 ** significant at .01 level, based on twotailed t tests * significant at .05 level, based on twotailed t tests ## significant at .01 level, based on X2 tests # significant at .05 level, based on X2 tests Variable INTERCEPT AUDITOR FORM APCHOICE SIZE LNRISKS UE3 Parameter Estimate .025 .003 .006 .000 .003 .001 .008 Variable INTERCEPT AUDITOR FORM APCHOICE SIZE LNRISKS UE36 Parameter Estimate .060 .000 .007 .002 .005 .005 .041 91 year, third year, and 36 month cumulative time horizon.12 In each case, White's [1980] test for the presence of heteroscedasticity revealed that the null hypothesis of homoscedasticity could be rejected at the .0001 level. Thus, the error terms in the four specifications of the model are heteroscedastic. In such cases, the estimated parameters are unbiased but inefficient (i.e., they are not minimum variance estimators), and the resulting parameter estimate covariance matrix is inconsistent. Therefore, Table 59 reports test statistics based both on an inconsistent covariance matrix (i.e., the typically utilized t statistics) and on a heteroscedasticity consistent covariance matrix (i.e., the heteroscedasticity adjusted X2 statistics based on White's [1980] correction). The significance tests performed using the heteroscedasticity consistent covariance matrix are asymptotic tests. Unless otherwise noted, only the tests based on the heteroscedasticity adjusted test statistics will be discussed in this section.13 The four specifications of Equation (18) were also examined for the presence of multicollinearity. Diagnostic tests based on tolerance 12 It is important to note that the 36 month cumulative time horizon is not independent of the other three time horizons since cumulative abnormal returns are calculated as the sum of the abnormal returns accruing over the first three years. Therefore, the four regression models are not independent. 13 Weighted least squares (WLS) regressions were also estimated in an attempt to overcome the heteroscedasticity problem. The residuals of the estimated OLS models were examined in an attempt to determine the nature of the heteroscedasticity and thus to aid in the determination of the appropriate weighting factors. However, proper weighting factors were not identified as a result of these residual analyses. Thus, White's [1980] correction is utilized since its utilization appears to be superior to employing WLS regressions with weighting factors chosen on an ad hoc basis. 92 factors revealed no serious multicollinearity problems. An independent variable's tolerance factor is calculated as lR2k, where R2k is the coefficient of multiple correlation which results when the independent variable is regressed on all the other independent variables in the model. This tolerance factor is often used "to detect instances where an X variable should not be allowed into the fitted regression model because of excessively high interdependence between this variable and the other X variables in the model" (Neter et al. [1985, p. 393]). Neter et al. continue by stating that frequently utilized tolerance limits are .01, .001, and .0001, below which the independent variable should be discarded from the model. In other words, unless the calculated tolerance factor is below the chosen tolerance limit, the independent variable is typically not dropped from the model since it is not considered to possess excessive collinearity with any other independent variable. In the four regressions reported in Table 59, the lowest tolerance factor calculated for any independent variable is .54. Thus, multicollinearity does not appear to pose a serious threat in the four estimated regressions.14 Panel A of Table 59 provides the results of estimating Equation (18) for the first year of public trading time horizon. As predicted by Hypothesis 1, none of the accounting risk proxy variables is statistically significant at conventional levels. Thus, consistent with 14 Analysis of the estimated residuals, as well as an examination of Cook's D distance measures, was also performed in an effort to detect outlying observations. The number of identified possible "outliers" was very small in each regression. Equation (18) was then reestimated with these potentially outlying observations dropped from the analysis. The results did not qualitatively change from those reported in Table 59. Hypothesis 1, accounting risk is not significantly correlated with IPO firms' first year abnormal returns. All three control variables exhibit (at least marginally) significant relationships with the dependent variable. Consistent with prior research, year 1 unexpected earnings is positively associated with year 1 abnormal security returns. Also, firms which list a greater number of risk factors in their prospectus (i.e., firms with greater business risk) experience lower abnormal returns than firms which enumerate fewer risk factors. However, SIZE is negatively associated (at the .10 level) with first year abnormal returns. Specifically, Panel A demonstrates that smaller firms (i.e., those with greater business risk) experience greater abnormal first year returns. Thus, conflicting results concerning the relationship of business risk and first year abnormal returns are documented. These conflicting results are consistent with the theory presented in Chapter 3 which indicates that mean riskadjusted abnormal returns should not be systematically related to measures of risk. Panel B shows that the three accounting risk variables also are insignificant in the explanation of year 2 abnormal returns. Thus, Hypothesis 1 is also supported for year 2 riskadjusted returns. Again, the widely documented positive association between unexpected earnings and abnormal returns is demonstrated. In fact, unexpected earnings is the only significant variable among the three control variables. Therefore, neither SIZE nor LNRISKS contributes to the explanation of second year abnormal returns. 