Bias in the post-IPO earnings forecasts of affiliated analysts: Evidence from a Chinese natural experiment

Author’s Accepted Manuscript Bias in the post-IPO earnings forecasts of affiliated analysts. Evidence from a Chinese natural experiment Nancy Huyghebaert, Weidong Xu www.elsevier.com

PII: DOI: Reference:

S0165-4101(15)00069-5 http://dx.doi.org/10.1016/j.jacceco.2015.10.002 JAE1087

To appear in: Journal of Accounting and Economics Received date: 6 February 2015 Revised date: 10 September 2015 Accepted date: 5 October 2015 Cite this article as: Nancy Huyghebaert and Weidong Xu, Bias in the post-IPO earnings forecasts of affiliated analysts. Evidence from a Chinese natural e x p e r i m e n t , Journal of Accounting and Economics, http://dx.doi.org/10.1016/j.jacceco.2015.10.002 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting galley proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Bias in the post-IPO earnings forecasts of affiliated analysts. Evidence from a Chinese natural experiment Nancy Huyghebaerta, Weidong Xub a

KU Leuven, Faculty of Economics and Business, Department of Accountancy, Finance and Insurance, Naamsestraat 69, 3000 Leuven, Belgium. phone: +32 16-326737, fax: +32 16-326737, e-mail: [email protected]. b Fonds Wetenschappelijk Onderzoek (11C7912N) and KU Leuven, Faculty of Economics and Business, Department of Accountancy, Finance & Insurance, Naamsestraat 69, 3000 Leuven, Belgium. phone: +32 16-326660, fax: +32 16-326737, e-mail: [email protected].

Abstract Investment banks and issuers of Chinese domestic IPOs became fully responsible for IPO offer prices only on June 10, 2009. Before this regulatory reform, the optimistic bias in post-IPO earnings forecasts is highly comparable across affiliated and unaffiliated analysts. Afterward, the forecasts of affiliated analysts are 33 percentage points more positively distorted on average. In the first 90 days after an IPO, this relative forecast bias even increases to 63 percentage points and enlarges further when the issuer’s stock price drops in the aftermarket. Affiliated analysts distort especially their forecasts for fiscal years further away from the forecast release date.

1. Introduction In the literature on the role of financial analysts in capital markets, a stream of research has focused on the investment advice of affiliated analysts, that is, analysts working at an investment bank/brokerage that maintains another relationship with the firm being covered.1 This research generally shows that affiliated analysts issue more positively biased investment recommendations, earnings forecasts, and target prices than unaffiliated analysts do. (For a review, see Mehran and Stulz, 2007, or Ramnath et al., 2008). However, this research has not yet been able to determine the exact mechanism driving this


























































1

Sell-side financial analysts work at an investment bank/brokerage that offers investment advice to external clients. Buyside financial analysts work for institutional investors at the investor’s own account. Like most of the literature, we consider only sell-side financial analysts in this study. In the remainder of the paper, the term “financial analyst” or “analyst” refers to sell-side financial analysts.

0


relative bias (e.g., Kolasinski and Kothari, 2008; Lin and McNichols, 1998; Mehran and Stulz, 2007; Ramnath et al., 2008). Affiliated financial analysts may be genuinely more optimistic about the firms they follow. Or they may intentionally and strategically distort their investment advice because of improper incentives. To explore what drives this relative bias, we examine a regulatory change in China, using it as a natural experiment to focus on the effects of analyst affiliation arising from IPO underwriting. Several aspects of this relationship may induce an optimistic bias in the investment advice of affiliated analysts. First, an investment bank that values a listing candidate more highly faces a better chance of being chosen as lead manager for that firm’s IPO (i.e., a self-selection bias arises). The corporate finance department and the research department of an investment bank often collaborate in IPOs to generate synergies. Financial analysts working for the IPO lead manager could thus be intrinsically more optimistic about the issuer, given that their employer obtained the IPO underwriting mandate. Those analysts may then also release more positive research reports shortly after the firm’s listing. We refer to this explanation as the “analyst-optimism view.” Once their employer obtains the IPO underwriting mandate, analysts affiliated with the IPO lead manager may also be able to set up and maintain more extensive contacts with firm management and hence can accumulate a deeper understanding of the issuer’s business and financial position. If those analysts subsequently rely on their better and private information to develop their forecasts, their forecast errors should be small and random, that is, not in one direction. In contrast, the forecasts of unaffiliated analysts will be based less on such firm-specific inputs and could thus be influenced more by prevailing stock market conditions (e.g., Bergman and Roychowdhury, 2008; Hribar and McInnis, 2012; Walther and Willis, 2013). We thus expect the sentiment-induced bias in earnings forecasts to apply particularly to unaffiliated analysts. As a result, unaffiliated analysts might exhibit positive

1


forecast errors especially in bullish stock markets, while their forecast errors could be much smaller or even negative when stock markets are bearish. We label this idea as the “information-advantage view.” Finally, as noted in the work of Michaely and Womack (1999) and James and Karceski (2006), analysts affiliated with the IPO lead manager might intentionally release a more positively distorted opinion as to the issuer to achieve some other goals. Post-IPO price support—that is, underwriter activities to keep the aftermarket stock price above the offer price—is a common practice in the United States (e.g., Aggarwal, 2000; Hanley et al., 1993). The IPO underwriting contract between issuers and their investment banks often provides for an overallotment option, penalty bids, etc., to stabilize the issuer’s stock price in the aftermarket; in the United States, this is usually done over the first 30 days after listing (Aggarwal, 2000). Alternatively, investment banks may also take it upon themselves to support an issuer’s stock price, possibly over a window longer than 30 days. Indeed, primary-market investors could blame the IPO lead manager for having set a too high offer price once the issuer’s stock price plummets in the aftermarket and hence lose their desire to buy into any of its future IPOs (e.g., Dunbar, 2000). To preserve their reputation among primary-market investors, IPO lead managers may provide (extra) post-IPO price support via optimistic research reports on the issuer (see also Lewellen, 2006). Arguably, this wish/need of providing price support is stronger when the issuer’s stock price comes under pressure in the aftermarket.2 We refer to this as the “post-IPO price-supporting view.” We test the empirical validity of these three views by relying on the annual earnings forecasts for IPO firms that were released in the first year after the firm’s listing in Mainland China. China proves interesting for this kind of research because a major regulatory change took place on June 10, 2009:


























































2

We do not say that unaffiliated analysts face no incentives whatsoever to issue a positively biased investment opinion. For example, as pointed out by James and Karceski (2006) and Mehran and Stulz (2007), both affiliated and unaffiliated analysts may inflate their advice to attract future corporate finance and brokerage business, to build and maintain good relationships with firm managers, etc. However, those incentives are common among affiliated and unaffiliated analysts. If the behavior of affiliated analysts is driven only by those incentives, we do not expect to find any differences between the forecasts of affiliated versus unaffiliated analysts. What we argue is that the post-IPO price-supporting incentive is unique to affiliated analysts, which could motivate those analysts to release a more positively biased opinion on the IPO firm.

2


investment banks in Chinese domestic IPOs became—for the first time—fully responsible for fixing IPO offer prices, together with issuers. Before, IPO offer prices were largely set by the Chinese regulator, by applying a maximum P/E ratio to the issuer’s earnings. As of June 10, 2009, this P/E ratio cap was abolished, and IPO lead managers henceforth set final offer prices after a book-building procedure among institutional investors. This regulatory reform provides a natural experiment to test the causal relation between the need for post-IPO price support and the forecast bias of affiliated analysts. If analysts affiliated with the IPO lead manager only exhibit optimism or are only better informed than other analysts, they should not behave differently before versus after this regulatory change. In contrast, IPO lead managers’ incentives to sustain an issuer’s stock price in the aftermarket were hugely affected by the IPO pricing reform of June 10, 2009. Obviously, the pre-condition for providing post-IPO price support is that investors perceive investment banks to be responsible for fixing IPO offer prices. We examine the annual earnings forecasts issued by affiliated and unaffiliated analysts in the first year after IPO for the firms becoming listed in Mainland China during 2004–2011. Although the stock exchanges in Shanghai and Shenzhen had already been re-opened at the beginning of the 1990s, the business of financial analysis has a much shorter history but has grown exponentially. The first ranking of financial analysts appeared only at the end of 2003. Analysts in China usually release their opinions on listed companies by means of annual EPS forecasts and investment recommendations; they do not circulate target prices. In our dataset, over 90% of investment recommendations were either “buy” or “strong buy.” This lack of variation in investment recommendations prevented us from meaningfully analyzing this type of analyst output, which is the main reason why we focus on EPS forecasts. Yet the huge number of buy and strong buy recommendations already suggests that the investment advice of financial analysts in China could be hugely distorted.

3


Our findings overwhelmingly confirm the view that the desire to provide post-IPO price support motivates analysts affiliated with the IPO lead manager to produce more positively distorted earnings forecasts than unaffiliated analysts do. Specifically, we find that, over the full sample period, the EPS forecasts of both affiliated and unaffiliated analysts are severely biased upwards. However, before June 10, 2009, this positive forecast bias does not differ significantly across the two groups. But after the regulatory reform, the average forecast error of affiliated analysts is 33 percentage points higher. Moreover, during the 90-day institutional lock-up period, this relative forecast bias even increases to a stunning 63 percentage points and enlarges further when the issuer’s stock price drops below the IPO offer price in the aftermarket. Arguably, our results cannot be reconciled with the view that affiliated analysts unintentionally release more positively distorted EPS forecasts because of their relative optimism about issuing firms. Under that view, affiliated analysts should always release more positively biased forecasts, whether before or after the regulatory change, in or out the lock-up period, or with a high or low issuer stock price in the aftermarket. Our results also refute the view that the presumed information advantage of affiliated analysts enables them to issue less biased and more accurate forecasts. In fact, our forecast-accuracy regressions reveal that affiliated analysts produce less precise forecasts after the regulatory reform. Finally, we find that affiliated analysts distort especially their earnings forecasts for fiscal years further away from the forecast release date. As stock market investors may find it harder to pinpoint the distortion of longer-term forecasts, we infer that affiliated analysts try to strategically mask their intentional post-IPO price-supporting activities. Interestingly, using US data, Michaely and Womack (1999) find that, shortly after a firm’s listing, analysts affiliated with the IPO lead underwriter issue more positive investment recommendations than unaffiliated ones. Likewise, James and Karceski (2006) show that affiliated analysts release more favorable recommendations but find no effect on target prices, except in broken IPOs. When explaining their observations, they intuitively assume that affiliated analysts aim to 4


provide a booster shot to the aftermarket stock price of issuing companies. However, they do not further explain why nor do they test this assumption. To the best of our knowledge, we are the first to explicitly put forward the post-IPO price-supporting view to explain affiliated analysts’ forecast distortion, differentiating it from other rationales. Thereby, the regulatory reform in the Chinese domestic IPO market acts as a valuable natural experiment, allowing an examination of the idea of aftermarket price support in its various facets. In Section 2, we briefly review the institutional aspects of Chinese domestic IPOs as well as the business of financial analysis. In Section 3, we develop our hypotheses. In Section 4, we empirically examine those hypotheses and discuss our main findings. In Section 5, we conclude.

2. Institutional background 2.1. The development of the Chinese IPO mechanism The Shanghai and Shenzhen stock exchanges were re-established in 1990 and 1991, respectively. By the end of 2011, 2,342 firms were listed in these markets, with a total market capitalization of RMB 21.5 trillion (USD 3.4 trillion). Since June 24, 1993, with the publication of “The circulation on enhancing the role of securities underwriters and professional intermediaries in stock offerings” by the China Securities Regulatory Commission (CSRC), every IPO issuer has to assign a qualified investment bank as lead manager for its IPO. Also in 1993, the CSRC mandated that investment banking and commercial banking should be separated. From 2004 to 2011, about 70 investment banks underwrote IPOs in China. They competed for underwriting business, which generated fees equal to 4.2% of gross IPO proceeds on average during 2004–2011. The IPO lead manager plays the most crucial role in the process of IPO. First, the lead manager performs the due diligence on the listing candidate and hence its analysts may obtain highly accurate information about the issuer. In China, the IPO lead manager assumes full responsibility for preparing 5


the IPO application documents and for submitting them with the CSRC (Huyghebaert and Xu, 2015). Second, the lead manager usually fixes the final offer price and also provides price support to stabilize the issuer’s stock price in the aftermarket. In China, a major regulatory change as regards IPO pricing took place on June 10, 2009, with the publication of “The guiding advice on further reform of the IPO pricing method.” On that date, the CSRC announced that it would no longer interfere in the pricing of IPO stock; IPO offer prices would instead by determined by the IPO lead managers after a bookbuilding exercise among institutional investors. Before June 10, 2009, the CSRC had largely set IPO offer prices, by applying a fixed P/E ratio cap to each issuer’s earnings. To attract the interest of the general public for IPOs, the CSRC deliberately set the P/E ratio cap—which was fixed annually— considerably below the prevailing market P/E ratio. As an example, P/E ratio caps were within the range of 13 to 16 during 1994–1999, much below the secondary-market P/E ratio of 15 to 58 (Francis et al., 2009). After Dec. 31, 2004, with the publication of Circulation No. 162, this official P/E ratio cap was abolished. However, the CSRC continued to manage IPO offer prices to some extent, relying on an implicit P/E ratio cap of 30 for most IPOs (Gao, 2010; Tian, 2010). Only on June 10, 2009 did investment banks, together with issuers, become fully responsible for IPO offer prices. The above IPO pricing regulation has produced extremely high first-day returns. According to the numbers compiled on Jay Ritter’s website, the first-day return in Chinese domestic IPOs is the third largest in the world, averaging to 133% in 1990–2010. Yet studies show different numbers in different periods. Su and Fleisher (1999) obtain 949% in 1987–1995. Chi and Padgett (2005) find 129% in 1996–2000. And Guo and Brooks (2008) report 93.5% between 2001 and 2005. Those extremely high first-day abnormal returns have created a “new-issue fetish,” as it was called by the Chinese media. Investors blindly bought any new issues, paying little attention to the quality of the issuer, as the CSRC almost guaranteed that they would make money. Artificially low IPO offer prices also reduced the odds of an issuer’s stock trading below the offer price shortly after listing. According to our analysis, out of 6


the 383 firms that became listed between Jan. 1, 2004 and June 9, 2009, only 60 (15.7%) saw their stock price drop to a value less than the offer price at least once in their first 90 listing days. More importantly, issuers and primary-market investors had no reason to connect stock performance in the aftermarket to the pricing ability of investment banks, as IPO lead underwriters had no influence on the final offer price before June 10, 2009. In line with this idea, Huyghebaert and Xu (2015) find that the (in)correct underpricing of IPO stock never significantly affected the subsequent market share of investment banks in Chinese domestic IPOs during those years. Correspondingly, investment banks had little incentives to provide post-IPO price support. The real change came with the IPO pricing reform of June 10, 2009, which acts as a natural experiment in our study. The reform indeed aimed to improve the market mechanism in the primary market; analyst behavior in the secondary market clearly was not an issue. As a result, investors started to face bigger chances of losing money when buying shares in IPOs after June 10, 2009. Arguably, the reform may have engendered an exogenous shock to the behavior of financial analysts. Moreover, this natural-experiment opportunity is unique to China. To the best of our knowledge, IPO underwriters in other major IPO markets have long taken full responsibility for offer prices, which likely has induced strong and persistent post-IPO price-supporting incentives for them. Interestingly—and unlike Western IPO markets—the green-shoe mechanism was only scarcely available to stabilize an issuer’s aftermarket stock price in China.3 Finally, investment banks in China never have had any discretionary allocation rights. On average, they apportion about 80% of IPO stock in the primary market to retail investors. Unlike retail investors, who can flip their initially allocated IPO shares immediately after listing, institutional


























































3

The “Regulation on security issuing and underwriting” published by the CSRC on Sep. 17, 2006, stipulates that, starting on Oct. 1, 2006, only IPOs of more than 400 million shares are entitled to rely on the green-shoe mechanism. Only 48 issuers after Oct. 1, 2006 (5.1% of the population) met that size requirement. In our sample, only three issuers relied on and fully exercised this overallotment right. Remarkably, not only the size (15%) but also the period during which the right can be exercised (30 days) are similar to Western IPO markets.

7


investors face a 90-day lock-up period for the IPO stock they obtain in an offering. 4 Institutional investors, which include mutual funds, insurance companies, and pension funds, are thus exposed to considerable price risk. Retail investors also play a vital role in the secondary market. A report issued by the International Organization of Securities Commissions reveals that retail investors held 83% of total market capitalization and contributed to 86% of total trading volume in 2010.
The 2011 year book of the China Securities Depository and Clearing Corporation similarly shows that, by the end of 2010, individual investors held 151 million stock accounts in Shanghai and Shenzhen, while institutional investors held only 0.58 million accounts.

2.2. Development of the business of financial analysis In China, the business of financial analysis developed with the stock exchanges. In 1999, a licensing system for financial analysts was established. Since then, the job of financial analyst emerged as a specific career path. In July 2001, a disciplinary committee was created under the auspices of the Securities Association of China, to supervise those analysts. In the same year, this committee published “The code of conduct for Chinese financial analysts,” to direct the activities of analysts. The guideline stresses that analysts should guard their integrity; their advice should be distorted neither by the interests of other departments within their organization, nor by those of securities issuers or institutional investors. According to the data we compiled, the number of financial analysts and their earnings forecasts have grown exponentially since 2004. In 2004, only 360 financial analysts issued 2,595 EPS forecasts on 430 listed firms (31% of the population of listed companies). By 2012, 1,630 analysts released 112,945 EPS forecasts on 1,931 listed firms (84% of the population). In total, from 2004 to 2012, 3,484


























































4

In this 90-day window, only retail investors can trade their initially allocated IPO shares. Trading volumes are nonetheless high. For IPOs during 2004–2011, 52% of shares sold in the IPO were already traded on the first listing day.

8


financial analysts issued 421,221 earnings forecasts; the average analyst followed eight firms and issued 40 EPS forecasts per year. For comparison, the average US analyst covered only six firms and released only 20 EPS forecasts per year during the period 1983–2002 (Fang and Yasuda, 2009). Most financial analysts work for an investment bank or a brokerage. In 2012, 90 investment banks or brokerages engaged at least one analyst. The average investment bank/brokerage employed 20 analysts. Independent research firms are still scarce in China. In our dataset, we find only seven. The largest employs 34 analysts releasing 2,406 earnings forecasts in 2012, that is, 2.1% of the number of EPS forecasts in that year. The first ranking of financial analysts was published only in 2003, by the magazine The New Fortune, considering institutional investors’ assessments of analyst performance in the previous year. The New Fortune has since published this list at the end of every year. The first list included only 24 star analysts, while the 2012 list grew to 127 star analysts. The list has become the most influential ranking of financial analysts in China.5 However, turnover on the list is quite high. Among the 326 analysts who appeared at least once in the rankings in 2003–2012, 190 analysts (58.3%) did so for fewer than two years. Only 58 analysts (17.8%) appeared in the list for more than five years.

3. Hypotheses The literature on financial analysts highlights the existence of self-interested incentives on the part of analysts’ employers, which may impact the behavior of both affiliated and unaffiliated analysts. In general, prior research shows that investment banks may compel their analysts to write overly optimistic investment reports to attract lucrative corporate finance business (Dugar and Nathan, 1995; Lin and McNichols, 1998). Analysts also may be under pressure to produce bullish reports to stimulate stock trading and generate extra brokerage commissions for their employers (Agrawal and Chen, 2008;


























































5

Two other rankings of financial analysts exist: Sky Eye and 21Century. Both started after 2004 and have limited influence.

9


Cowen et al., 2006). In many countries, including China, some investors face short-selling constraints, and thus pessimistic reports cannot produce the same trading volumes as optimistic ones. Analysts may also use favorable reports to develop or maintain good relationships with firm managers, so as to ease their access to information about the listed company (Das and Saudagaran, 1998; Lim, 2001). However, if these incentives are the only ones affecting the behavior of financial analysts, we may not observe any differences in the EPS forecasts of analysts affiliated vs. unaffiliated with the IPO lead manager. As James and Karceski (2006) contend, the desire to generate future business or cultivate a good relationship with firm management exists equally among affiliated and unaffiliated analysts. In other words, regardless of whether investment banks established a prior relationship with a listed company, they need to compete for future corporate finance and trading business and thus may wish to please firm management by overly optimistic research reports. Ramnath et al. (2008) therefore stress that more academic research is needed to find out what is special about investment banking relationships and how this might affect the output of affiliated analysts. Using a unique naturalexperiment opportunity in China, we propose three hypotheses as to how and why the IPO underwriting relationship might influence the post-IPO earnings forecasts of affiliated analysts. Eccles and Crane (1988) point out that the corporate finance department and the research department of investment banks often collaborate in IPOs, in order to share their expertise and reduce costs. 6 An investment bank’s view on a listing candidate could thus be largely shaped by how its analysts perceive the issuer. IPO candidates, when choosing their underwriters in a beauty contest, consider the contract terms offered by various investment banks. Banks that value an issuer more


























































6

However, in the 2003 Global Settlement between the SEC and 10 major investment banks in the USA, the SEC imposed a clear separation between the corporate finance department and the research department of investment banks (SEC release 2003-54). Correspondingly, affiliated analysts in the United States can no longer participate in pitches and road shows for IPOs and receive compensation that is related to the investment bank’s corporate finance business. But the new regulation did not prohibit those analysts from conducting research on issuing companies. Affiliated analysts can thus still put a value on IPO candidates and share their research results with internal and external clients. Moreover, in the case of China, the CSRC has never forbidden that an investment bank’s corporate finance department and its research department work together in IPOs.

10


highly likely are willing to endorse a higher IPO offer price and to accept a lower IPO fee rate, ceteris paribus. Correspondingly, the investment bank that holds a more optimistic view about the issuer has a better chance of obtaining the lead underwriter mandate. Because of such a self-selection effect, analysts affiliated with the IPO lead manager may release more positive EPS forecasts after an IPO without an intention to bias their estimates. We refer to this idea as the “analyst-optimism view.” If this view holds true, we should observe that, with or without an incentive to support the issuer’s stock price in the aftermarket, affiliated analysts always release more positive EPS forecasts than unaffiliated ones. China’s June 10, 2009 regulatory reform should thus not have changed the (potential) gap between the EPS forecasts of affiliated vs. unaffiliated analysts. Moreover, under the analyst-optimism view, this relative forecast bias should not be associated with the issuer’s stock performance after listing. Affiliated analysts should thus not release more positively biased forecasts when the issuer’s stock price comes under pressure after the IPO. We thus propose the following hypothesis:

Hypothesis 1: If the analyst-optimism view holds true: 1A. Regardless of whether the firm became listed before or after June 10, 2009, the EPS forecasts of affiliated analysts should be more positively biased than those of unaffiliated analysts. 1B. Regardless of whether the firm became listed before or after June 10, 2009, this relative bias in earnings forecasts should not be associated with the issuer’s stock price in the aftermarket.

Jacob et al. (2008) contend that, by conducting the due diligence on a listing candidate, analysts affiliated with the IPO lead manager obtain a better understanding of the issuer than other analysts. The lead manager’s analysts also can establish a closer relationship with the issuer’s management. That relationship may give the analysts access to more accurate and timelier information about the firm once 11


listed. Using US data from 1995 to 2003, Jacob et al. (2008) indeed find that affiliated analysts release more precise forecasts than unaffiliated ones. Although forecast accuracy differs from forecast bias,7 we contend that affiliated analysts may also produce less biased forecasts, as they may better understand the issuer’s business and financial position and thus can base their forecasts to a larger extent on the company’s fundamentals. Correspondingly, the forecast errors made by those affiliated analysts can be expected to be small and random, that is, not in one direction. In contrast, unaffiliated analysts, because of their lack of precise information, likely rely more on judgment, which could be influenced more by prevailing market conditions (e.g., Bergman and Roychowdhury, 2008; Hribar and McInnis, 2012; Walther and Willis, 2013). We thus expect the sentiment-induced bias in earnings forecasts to apply particularly to unaffiliated analysts. As a result, unaffiliated analysts might exhibit positive forecast errors in times of bullish stock market conditions and much smaller or even negative errors when stock markets are bearish. We label this idea as the “information-advantage view.” China’s June 10, 2009 regulatory reform did not influence the information advantage of affiliated analysts. Hence, if affiliated analysts issue less biased and more accurate EPS forecasts because of an information advantage, this effect should endure regardless of whether a firm became listed before or after June 10, 2009. We thus put forward the following hypothesis:

Hypothesis 2: If the information-advantage view holds true: 2A: Regardless of whether the firm became listed before or after June 10, 2009, the earnings forecasts of affiliated analysts should be less biased than those of unaffiliated analysts. 2B: Regardless of whether the firm became listed before or after June 10, 2009, the earnings forecasts issued by affiliated analysts should be more accurate than those of unaffiliated ones.


























































7

An analyst who makes big but random errors issues less biased forecasts than an analyst who constantly makes small but positive forecast errors. However, in terms of forecast accuracy (no matter whether it is measured by the absolute forecast error or by the mean squared forecast error), the former analyst is less accurate than the latter.

12


Aggarwal (2000) shows that IPO underwriting contracts between US issuers and their investment banks often provide for aftermarket price support for the issuer’s stock price, typically during the first 30 days after listing. Lewellen (2006) points to two other rationales for lead underwriters to provide post-IPO price support: to reward institutional investors for revealing their private information and thus help investment banks to price the offering and to disguise IPO overpricing. Ljungqvist et al. (2006) emphasize that institutional investors need time to sell out their initially allocated IPO shares. If the issuer’s stock price declines too fast in the aftermarket, they may incur a loss on their IPO investment. In the case of China, this concern cannot be ignored, as institutional investors face a 90-day lock-up period. Likewise, when the IPO lead manager is ultimately responsible for fixing IPO offer prices, primary-market investors could blame that bank for having set too high a price once the issuer’s stock price drops in the aftermarket and thus could lose interest in buying into any of its future IPOs. Likewise, IPO candidates may no longer want their offering to be underwritten by a bank that cannot fix IPO offer prices (Dunbar, 20000). Investment banks may thus also take it upon themselves to support an issuer’s stock price in the aftermarket. Hence they could put pressure on their analysts to inflate issuer EPS forecasts, thereby provoking a relative forecast bias vis-à-vis unaffiliated analysts. We refer to the above arguments as the “post-IPO price-supporting view.” To examine this post-IPO price-supporting view, it would be ideal if we could randomly assign price-supporting incentives to some IPO lead managers and take those away from others. While we cannot set up such an experiment in real life, we contend that China’s regulatory reform of June 10, 2009, created—for the first time—strong incentives for IPO lead managers in the country to provide post-IPO price support. Indeed, when IPO underwriters have no major influence on IPO offer prices, they have no specific reason to sustain an issuer’s stock price in the aftermarket, as primary-market investors are unlikely to blame them when the issuer’s stock price drops. In contrast, after the 13


regulatory reform, investment banks became fully responsible for setting IPO offer prices. IPO lead managers—as specialists and repeat participants in the IPO market—could then be held liable by primary-market investors for a declining issuer stock price shortly after the IPO. Losing the interest of institutional investors is problematic for an investment bank, as those professional investors play a crucial role in book-building by providing their tentative price indications. Some scholars even argue that an investment bank’s network of institutional investors is its most valuable asset (e.g., Eccles and Crane, 1988; Morrison and Wilhelm, 2007). However, investment banks in China can also not ignore retail investors, given their massive stake in the primary market. We therefore expect analysts affiliated with the IPO lead manager to issue more positively distorted earnings forecasts than unaffiliated ones only after June 10, 2009. Moreover, we expect those price-supporting incentives to be strongest in the first 90 days after IPO, when institutional investors face a lock-up for their IPO shares. The other direct implication of this post-IPO price-supporting notion is that the upward bias in affiliated analysts’ EPS forecasts should become even larger when the issuer’s stock starts to trade below the offer price. Finally, analysts affiliated with the IPO lead manager may prefer to distort their forecasts on later-year rather than current-year EPS, so as not to tarnish too much their reputation among investors. More information on the issuer usually becomes available as the end of the forecast horizon approaches (Lin and McNichols, 1998; Bessler and Stanzel, 2009). If affiliated analysts wish to strategically disguise their price-supporting behavior, they likely prefer to manipulate their later-year forecasts. Over time, they can then slowly adjust downward their initial upwardly biased EPS forecasts for the later years. With an upward bias in later-year EPS forecasts, investors, especially retail ones, may not even remember what affiliated analysts had forecasted shortly after IPO by the time that the issuer releases its realized EPS. In sum, we propose the following hypothesis:

14


Hypothesis 3: If the post-IPO price-supporting view holds true: 3A. For the firms that became listed before June 10, 2009, the earnings forecasts of affiliated analysts are no more positively biased than those of unaffiliated analysts. 3B. For the firms that became listed after June 10, 2009, the earnings forecasts of affiliated analysts are more positively biased than those of unaffiliated analysts. This relative bias is stronger in the 90-day lock-up period and when the issuer’s stock price dropped after IPO. Also, affiliated analysts will bias especially their later-year rather than their current-year forecasts.

These three views are not mutually exclusive. An affiliated analyst with an optimistic opinion on an issuer can also obtain access to more accurate information about this firm during IPO underwriting and, at the same time, be required by his employer to provide price support in the aftermarket. However, a fundamental difference between the post-IPO price-supporting view and the other two views is that the former assumes that analysts intentionally distort their forecasts, whereas the other two presume that analysts honestly report their forecasts. It is here that the regulatory reform can help to identify what is driving the bigger relative forecast bias of affiliated analysts, if any. Indeed, this regulatory reform affected neither affiliated analysts’ relative optimism nor their access to information but likely did hugely impact their price-supporting incentives.

4. Empirical study 4.1. Data and methodology 4.1.1. Data collection We relied on the CSMAR database to collect the annual earnings forecasts on the 1,095 firms that became listed in Mainland China between 2004 and 2011. We limit our sample to forecasts issued in 15


the first year after IPO, as other corporate finance activities (e.g., SEOs, M&As) may take place if the sample period is extended for too long after IPO (see also James and Karceski, 2006; Ljungqvist et al., 2007; Michaely and Womack, 1999).8 This could then complicate a clear test of the effects of the IPO underwriting relationship. Moreover, both the information advantage obtained in IPO underwriting and post-IPO price-supporting incentives tend to diminish as time elapses after the IPO. All forecasts relate to the issuer’s annual rather than quarterly earnings, given that the latter forecasts are only scarcely disclosed in China. While some of the sample forecasts relate to the EPS of the IPO year itself, others relate to the EPS of subsequent years. The latter feature of our dataset enables us to compare EPS forecasts across various forecast horizons. From 2004 to 2011, 969 issuers (88.5% of the total number of IPO firms) were covered by at least one analyst in their first listing year, corresponding with 319 firms in the subperiod of 2004–June 9, 2009 and 650 firms in the subperiod of June 10, 2009–2011. In total, 1,627 financial analysts employed by 90 investment banks/brokerages released 36,885 earnings forecasts on these 969 firms in their first post-IPO year. Next, as in prior research (e.g., Jacob et al., 2008; Lin and McNichols, 1998), we restrict our sample to the issuers that were covered by at least one affiliated analyst, that is, an analyst employed by the IPO lead manager.9 For the 969 issuers on which we have analyst coverage data, 384 firms (39.6% of this subsample) were followed by an affiliated analyst, corresponding with 80 firms before June 10, 2009 (25.1% of this subsample) and 304 firms thereafter (46.8% of the subsample). Our final sample thus reduces to 21,499 earnings forecasts (58.3% of the initial sample). Arguably, the IPO firms in our sample are among the most widely covered issuers. On average, a sample firm is followed by 10


























































8

Affiliated financial analysts in the United States cannot release any investment recommendations, earnings forecasts, and target prices in the quiet period, now lasting 40 days after IPO. In China, such regulation did not exist during our sample period. 9 For all earnings forecasts in our sample, 2,076 were issued by an analyst employed by the IPO lead manager, while only 106 were released by an analyst employed by an IPO co-underwriter. In the main test, we classify the latter analysts as unaffiliated; in an unreported robustness test, we classified them as affiliated, finding similar but somewhat less significant results. This outcome thus confirms our earlier conjecture that IPO lead managers assume much larger responsibilities in IPOs, thereby generating stronger incentives to provide post-IPO price support after June 10, 2009.

16


analysts (median of eight). Anyway, as the decision of the IPO lead manager’s analyst(s) to cover an issuer is not a random one, this sample selection criterion might introduce a sample selection bias. We therefore also adopt Heckman’s (1976) two-step regression model to deal with this potential problem.

4.1.2. The measurement of forecast error and forecast accuracy We define an analyst’s forecast error FEj,i,t as (FEPSj,i,t – AEPSi,t)/|AEPSi,t|, where FEPSj,i,t is the forecast of earnings per share by analyst j on firm i for year t. AEPSi,t is the reported earnings per share by firm i for year t. We define forecast accuracy (FAj,i,t) as the absolute value of FEj,i,t.10 A bigger FAj,i,t thus means a less precise forecast. We winsorize FE at 5% and 95% and FA at 95% to deal with outliers.11 Table 1 reports summary statistics on FE and FA. For the full sample, the average forecasted EPS is 130% bigger than the reported EPS number (median of 40%). This positive bias applies to affiliated as well as unaffiliated analysts. t-tests and Wilcoxon rank-sum tests confirm that the average and median FE are significantly larger than zero. Next, Table 1 reveals that the mean and median FE are 112.9% and 16.5% for the firms that became listed before June 10, 2009. Thereafter, they equal a significantly larger 133.5% and 54%. Similar results arise as to forecast accuracy: the magnitude and significance level of FA resemble those of FE. Also, only 3,843 out of 21,499 EPS forecasts (17.9% of the full sample) turn out to be smaller than the reported EPS number, which renders the summary statistics on FE and FA highly comparable. In sum, analysts in Chinese domestic stock markets did not correctly forecast issuers’ EPS throughout the sample period; their forecasts were severely positively biased on average. Finally, the average and median FE (and FA) of affiliated analysts are significantly bigger than those of unaffiliated ones, in both subsamples.12




























































10

In an unreported robustness check, we also measured FA as the mean squared forecast error. Results were not affected. We also winsorized FE at 1% and 99% and FA at 99%; results proved similar. 12 In an unreported robustness test, we also examined the forecasts released in the second year after IPO. Interestingly, we find that for IPOs after June 10, 2009, the mean and median FE of affiliated analysts tend to decline in the second post-IPO 11

17


To improve our comparison of EPS forecasts of affiliated vs. unaffiliated analysts, we should control for those firm characteristics that may influence the forecast error (and forecast accuracy) as well as for the forecast horizon. Hence we decided to compare the FE of affiliated and other analysts by IPO firm and by forecast year. For every issuer and (post-)IPO year, we compute the difference between the average FE of affiliated vis-à-vis unaffiliated analysts. We repeat this procedure for all 384 sample firms. Table 2 shows summary statistics on the relative forecast bias for the firms that became listed up till June 10, 2009 and for the firms that became listed thereafter. For each subsample, we further divide the forecasts into those for the IPO year (IPO year) and those for the years thereafter (Later years). From a t-test and a Wilcoxon rank-sum test, we conclude that the average and median relative bias are not significantly different from zero before June 10, 2009. However, they do become significantly positive after the regulatory reform. Those results thus indicate that the regulatory reform has induced analysts affiliated with the IPO lead manager to produce more severely inflated EPS forecasts. Interestingly, we also detect that this positive relative bias after June 10, 2009 arises particularly for the later-year EPS forecasts. The latter outcome suggests that affiliated analysts strategically distort their forecasts in a way to mask their self-interested behavior. To examine the stock market effects of those positively distorted EPS forecasts, we conduct an event study (see also Lin and McNichols, 1998; Michaely and Womack, 1999). We calculate the threeday cumulative abnormal return (CAR) from one day before till one day after the forecast release date. Table 3 shows the results. We find a significant positive CAR for the full sample as well as for each subsample, thereby confirming that an upward-biased EPS forecast can boost an issuer’s stock price






































































































































































































year, while unaffiliated analysts’ FE is not affected. Moreover, we find regression results that are comparable but weaker than those reported for the first listing year.

18


after IPO. This CAR is not significantly different for forecasts released by affiliated vs. other analysts. If anything, after June 10, 2009, the average CAR is somewhat larger (but not significantly so; p-value = 0.275) for forecasts issued by an affiliated analyst. The latter findings thus indicate that stock market investors do not question the credibility of EPS forecasts released by affiliated analysts.

4.1.3. Regression model on forecast error and forecast accuracy To compare the forecast error and forecast accuracy across affiliated and unaffiliated analysts, we rely on multivariate regression analyses, for the IPOs before and after June 10, 2009, respectively. To that end, we run OLS regressions for both subsamples, using FE and FA as dependent variables. As regards the explanatory variables, we first create a dummy equal to one if the forecast was issued by an analyst employed by the IPO lead manager and zero otherwise (Affiliated dummy). The coefficient on this dummy should then capture whether affiliated analysts are associated with a bigger forecast error (or a better forecast accuracy) than unaffiliated ones. Like James and Karceski (2006), we also construct a variable that captures the issuer’s stock performance in the aftermarket. We subtract the IPO offer price from the stock’s closing price at the day before the forecast was released and divide it by the offer price (Relative price). From the primary-market investors’ point of view, Relative price thus reflects how much they have gained or lost from having bought one IPO share at the IPO offer price up till the forecast release date. To examine whether the positive relative bias of affiliated analysts increases as the issuer’s stock price drops in the aftermarket, we also interact Affiliated dummy and Relative price. As EPS forecast errors can be influenced by many other factors, we include a number of control variables. We account for the analyst’s reputation by means of a dummy that equals one if he was included in The New Fortune’s ranking in the year before the forecast release date and zero otherwise

19


(Analyst ranking).13 Jackson (2005) and Fang and Yasuda (2009) argue that reputation is extremely important to financial analysts and that their concern about it may help them to resist conflicts of interest. To separate analyst reputation from analyst experience, we also add the analyst’s total number of forecasts, on IPO firms as well as on already-listed firms, from the start of our dataset (i.e. 2001) till the current forecast. We thereafter take the log of one plus this number of forecasts, to account for the declining marginal benefit of an extra forecast (Analyst experience). Second, reputational concerns on the part of the analyst’s employer could also help to mitigate conflicts of interest. In line with Fang and Yasuda (2009), we include the employer’s market share in IPO underwriting in the previous year (Market share). The market share of investment bank i in year t–1 is calculated by dividing the gross proceeds of IPOs in which bank i served as lead manager in year t–1 by the gross proceeds raised in all IPOs in year t–1.14 As some investment banks may focus their business model on offering valuable research coverage, they may employ a larger team of analysts, ensuring that their research staff can produce independent investment reports. We therefore also add the total number of financial analysts employed by an analyst’s investment bank at the start of the forecasting year (Number of analysts). We further control for a number of issuer-specific characteristics. We include three variables that relate to the firm’s equity issuance: the fee rate that it paid to its investment banks on its IPO (Fee rate), the first-day abnormal return on its IPO (Abnormal return), and a dummy equal to one when it offered equity in the three years after IPO (SEO dummy).15 The issuer that accepted a larger IPO fee rate might also consent to bigger fees on future corporate finance transactions. To compete for future business with this firm, both affiliated and unaffiliated analysts could then issue overly optimistic EPS forecasts. For this same reason, we may also find a positive relation between the first-day abnormal


























































13

In an unreported robustness test, we set the Analyst ranking dummy equal to one when the analyst was included at least once in The New Fortune’s star-analyst ranking in the year(s) before the forecast release date. Results prove similar. 14 We also used the average market share of investment bank i in years t–1, t–2, and t–3, finding that results did not change. 15 We do not have any data on the firm’s M&A activity nor on the brokerage trading business it generates for investment banks. So we cannot control for an investment bank’s incentives to release positively distorted EPS forecasts to generate those two types of future business.

20


return and FE. Conversely, when IPO lead managers underpriced the offering to a larger extent, thereby allowing for a bigger first-day abnormal return, it might reflect that the IPO was surrounded by more uncertainty (e.g., Beatty and Ritter, 1986; Rock, 1986). Then Abnormal return and FA might correlate positively. We expect SEO dummy to associate positively with FE and FA when investment banks compete for future corporate finance business by means of upwardly biased EPS forecasts. Implicitly, this variable specification assumes that banks can anticipate at least in part which IPO firms will implement an SEO (see also Cliff and Denis, 2004; Lin and McNichols, 1998). Next, the EPS of larger issuers may be easier to forecast as bigger firms usually have more stable earnings and release more information. We therefore add the log of the issuer’s total assets at the end of the pre-IPO year (Assets).16 We further include a dummy equal to one if the Chinese state controls more than 50% of the issuer’s direct and indirect voting rights before IPO and zero otherwise (SOE dummy). Private owners, who usually have invested much of their personal wealth in their firm, may care more about the firm’s stock performance in the aftermarket than government-related owners; the latter might indeed also value nonfinancial objectives (e.g., Huyghebaert and Wang, 2012). On the other hand, as a repeat participant in the IPO market, the Chinese state may want to avoid falling stock prices on its privatization IPOs, as this could jeopardize the success of its future privatizations. Furthermore, the earnings of SOEs likely are easier to predict, as SOEs have a longer history and usually benefit from monopoly power. As a result, the room left for EPS forecast distortions that go unnoticed by stock market investors tends to be smaller for SOEs. In sum, we consider the relation between state ownership (SOE dummy) and FE/FA as an empirical question. Lin and McNichols (1998) and Bessler and Stanzel (2009) find that a forecast that is released much before fiscal-year end is associated with a larger forecast error. We therefore also add Forecast horizon, which captures the number of days between the forecast release and the end of the fiscal year


























































16

Results did not change when using the log of the firm’s market capitalization at the day before the forecast was released.

21


to which the forecast relates.17 To account for potential industry effects, we include Industry dummies based on the CSRC industry classification code. Finally, we control for year fixed effects by means of Year dummies.18 We run the same model to explain the forecast error (FE) and forecast accuracy (FA):

 E    E      E    E  ! E" #  E$ % & E' ()*  E+ 
 E, )* E - ./ E

 E  ./









E  
*&* 0* 1 T 2  1 J 3  H Table 4 presents summary statistics on the above variables. Because of missing values for some variables, we could retain only 3,098 (86.2%) out of 3,596 forecasts in the first subsample and 12,838 (71.7%) out of 17,903 forecasts in the second subsample. However, the average and median FE and FA of the dropped-out forecasts are not significantly different from those of the retained forecasts. When comparing the other variables in Table 4 across the two subsamples, we find that Relative price and Abnormal return are much lower for IPOs after June 10, 2009. The average Relative price is about 100% in the first subsample, while it equals only 30% in the second subsample. The average first-day abnormal return is 87.6% in the first subsample, while it is only 28.8% in the second subsample. The declining values of Relative price and Abnormal return may thus indicate that the need for post-IPO price support has grown over time. Table 4 also shows that 82.6% of IPO firms are state-controlled in the first subsample; this fraction drops to 22.6% in the second subsample. Finally, from checking the correlations among all explanatory variables in Table 4, we infer that multicollinearity is unlikely to be an issue in our study. This is also confirmed by each model’s VIF statistics (not shown in the tables).


























































17

Ideally, we would like to calculate Forecast horizon using the official EPS release date rather than the fiscal-year end, as firms announce their annual EPS numbers at the time that they publish their audited financial statements, which usually is only in March or April of the subsequent year. However, we do not have the data on the annual reports’ release date. 18 Although not conventional in the literature, we also ran the models after including the aftermarket stock-return volatility and stock turnover, measured from the second trading day after IPO till the day before the forecast release date. Those variables were never significant. Also, they did not affect the significance of the other explanatory variables in the models.

22


4.1.4. The Heckman (1976) two-step regression model As we restricted our sample to the issuers that were covered by at least one affiliated analyst in their first listing year, questions may arise about whether analysts’ follow-up decisions confound with their forecast errors. This problem has been largely ignored in the literature, maybe because most IPO firms in Western economies are covered by their lead underwriter’s analysts after listing. As an example, Cliff and Denis (2004) show that 80% of IPO firms in the United States during 1993–2000 were followed by an affiliated analyst. However, in China, for the 969 firms on which we have analystcoverage data in their first listing year, only 384 firms (39.6%) are followed by an affiliated analyst. We thus cannot ignore a potential sample selection bias. We adopt Heckman’s (1976) two-step method to deal with this potential problem. In a first step, we run a probit model for the probability that the issuer is covered by an affiliated analyst. 19 The Follow-up dummy is set equal to one in this case and zero otherwise. As explanatory variables, we include the total number of analysts employed by the issuer’s IPO lead manager at the start of the IPO year (Underwriter analyst number). We expect that IPO lead underwriters engaging more financial analysts are more likely to provide post-IPO research coverage, ceteris paribus. Next, we add the issuing P/E ratio, from which we subtract the market P/E ratio at IPO (Pricing aggressiveness). Stocks issued at a higher P/E ratio may need more research coverage to sustain their stock price once listed. We expect this effect to arise only after June 10, 2009, when the issuing P/E ratio cap was abolished. We further control for a number of IPO-related factors, by means of Fee rate and Abnormal return. If, as argued by Loughran and Ritter (2004), issuers esteem research coverage and thus are willing to pay a higher IPO fee rate for it as well as to accept more IPO underpricing, the odds of underwriter-analyst


























































19

In our sample, all issuers that are covered by their IPO lead manager’s analysts are also followed by at least one unaffiliated analyst. So we can exclude the possibility that the issuer is covered only by its lead manager’s analysts.

23


coverage may increase with those two factors. Next, we include SEO dummy, as underwriters may prefer to cover issuers they expect to raise extra equity once listed. As in the main regression model, we account for firm-specific characteristics by means of Assets and SOE dummy. Bigger firms likely provide more opportunities for future investment banking business. Likewise, investment banks may face stronger incentives to cover SOEs, as the Chinese state is a repeat participant in the IPO market. We again include industry and year fixed effects. Our first-step model thus looks as follows:
**45 E 64 ) E 7  !!!  E 
 E )* E" ./ E$  E' ./









1 T 2  1 J 3  H

From the above regression, which is run separately for each of the two subsamples, we calculate Heckman’s (1976) inverse Mills ratio and add it as an extra regressor in the second-step regression.

4.2. Empirical results Table 5 shows the FE regression results, while Table 6 displays the FA regressions results. In Panel A, the output relates to the IPOs before June 10, 2009; Panel B then shows the output for the IPOs thereafter. For each subperiod, we use all EPS forecasts issued in the first listing year. We also run the regressions using only the forecasts released in the 90-day institutional lock-up period. For each set of analyses, the first column reports the simple OLS results, while the subsequent two columns show the Heckman (1976) two-step results. Although the inverse Mills ratio is usually significant, thereby indicating that the choice of the IPO lead manager’s analysts to cover an issuer generally correlates with their FE and FA, the second-step regression results are very similar to the OLS results. Our conclusions thus do not just arise from the follow-up decision of the IPO lead manager’s analysts. 24


As regards the forecast-error regressions, the coefficient on Affiliated dummy is never significant for IPOs before June 10, 2009. So we find no evidence whatsoever that affiliated analysts issue more positively biased EPS forecasts than unaffiliated ones when IPO offer prices were set by the CSRC. Likewise, the interaction between Affiliated dummy and Relative price is never significant for those early IPOs. However, after June 10, 2009, regardless of the window over which the forecasts were released (one year or 90 days), the EPS forecasts of affiliated analysts are always significantly larger. For forecasts issued in the first post-IPO year, the average FE of affiliated analysts is 33 percentage points bigger than that of unaffiliated ones (34 percentage points in the Heckman model). For forecasts released in the 90 days after IPO, this difference even increases to a stunning 63 percentage points (68 percentage points in the Heckman model). 20 The latter outcome thus reveals that affiliated analysts issue even more upwardly biased forecasts during the 90-day institutional lock-up period, which is when post-IPO price support likely is most valuable. Figure 1 illustrates the huge effects of institutional lock-up expiration on daily trading volumes and abnormal returns. Next, the coefficient on the interaction between Affiliated dummy and Relative price becomes negative and significant at the 5% level (1% level in the Heckman model) for IPOs after June 10, 2009 but only when considering the EPS forecasts released in the 90 days after listing. In other words, when the issuer’s stock price declined by one percentage point relative to the IPO offer price in this 90-day window after the IPO pricing reform, the FE of affiliated analysts is 0.93 percentage points more positive on average than the FE of unaffiliated analysts (1.06 percentage points in the Heckman model). The coefficient on Relative price itself is never significant, thereby indicating that unaffiliated analysts do not positively distort their EPS forecasts depending upon the issuer’s stock performance in the aftermarket.




























































20

As a robustness check on this 90-day window, we also used the forecasts released in the 30 days and 120 days after IPO for the firms listed after June 10, 2009. The coefficient on Affiliated dummy equals 0.32 for the 30-day regression model (significant at the 5% level) and 0.45 for the 120-day regression model (significant at the 1% level). Together, those findings suggest that affiliated financial analysts distort their earnings forecasts especially in the 90-day window after listing.

25


The results from the forecast-accuracy regressions resemble those from the forecast-error regressions. For IPOs before June 10, 2009, we find no significant negative coefficient on Affiliated dummy. The information advantage that affiliated analysts are presumed to enjoy because of IPO underwriting thus does not seem to result in more precise EPS forecasts. In contrast, for IPOs after June 10, 2009, affiliated analysts have significantly bigger absolute forecast errors than unaffiliated ones. This accuracy gap between affiliated and unaffiliated analysts is most pronounced during the 90-day lock-up period and grows even larger as the issuer’s stock trades below the IPO offer price after listing. Overall, our results indicate that, in the absence of any post-IPO price-supporting incentives (i.e. before June 10, 2009), affiliated analysts did not release more optimistic EPS forecasts than unaffiliated ones. Also, their forecasts were not more accurate. However, once such price-supporting incentives emerged (i.e. after June 10, 2009), the upward bias in affiliated analysts’ EPS forecasts started to soar. In addition, this relative forecast bias proves even larger when price-supporting incentives are particularly strong (i.e. during the 90-day institutional lock-up period and when the issuer’s stock price drops after listing). Our Chinese natural experiment therefore allows us to conclude that the relative optimism of affiliated analysts is hugely influenced by their incentives to provide post-IPO price support. Next, our findings clearly refute the analyst-optimism view, which predicts that the relative forecast bias should neither change with the regulatory reform nor depend on the issuer’s stock performance in the aftermarket (Hypotheses 1A and 1B). They also indicate that the informationadvantage view may not apply for Chinese IPOs. Table 1 indeed revealed that both affiliated and unaffiliated analysts issue severely positively biased EPS forecasts, in both subsamples. According to the information-advantage view, when affiliated as well as unaffiliated analysts issue overly optimistic forecasts, a less biased forecast would mean that the FE of affiliated analysts is less positive on average. The coefficient on Affiliated dummy should then have a negative sign. However, we find an

26


insignificant coefficient for IPOs before June 10, 2009, and a significant but positive coefficient for IPOs thereafter. From those results, we can thus also refute Hypothesis 2A. Next, we briefly discuss our results for the control variables from the FE regressions.21 For IPOs before June 10, 2009, star analysts do not seem to issue less biased forecasts. However, for IPOs after June 10, 2009, we do find some evidence that their forecasts are less distorted but only for forecasts released in the first listing year (and when using the Heckman model). As the ranking of financial analysts was not yet very stable during our sample period, this outcome might not be so surprising. An analyst’s experience cannot help to reduce forecast errors, either. However, by the end of 2012, the average analyst in our sample only had 3.3 years of experience (median of three). Next, investment banks with larger market shares in IPO underwriting are associated with less positively distorted EPS forecasts. This outcome thus suggests that particularly the banks with smaller market shares may rely on their analysts’ forecasts to build their market share in IPOs. Investment banks employing many analysts also issue less biased forecasts, though not always significantly so. Combining our findings on analyst-specific characteristics (Analyst ranking, Analyst experience) and bank-specific traits (Market share, Number of analysts), we conclude that the behavior of financial analysts in China is determined mostly by their employers rather than by the analysts themselves, i.e. the boss talks. We further find that the issuers that agreed to a larger IPO fee rate are rewarded by more optimistic EPS forecasts but only in the 90 days after IPO and after June 10, 2009. While we detect no association between the first-day abnormal return and subsequent FE, we do find that the forecasts of firms that initiate an SEO in the three years after IPO are more positively distorted. The latter result supports the notion that investment banks may also rely upon their analysts’ EPS forecasts to compete for future SEO underwriting business. Next, bigger issuers (Assets) are associated with less inflated


























































21

As the results for the control variables from the FA regressions are largely comparable to those from the FE regressions, we do not separately discuss them.

27


forecasts, in line with the idea that these firms have more stable earnings and release more information. However, the coefficient on Assets is never significant during the 90-day institutional lock-up period. The forecast error of SOEs is smaller than that of private-controlled firms for IPOs before June 10, 2009, though this effect declines over time. This finding supports the idea that SOEs have more stable earnings, which hampers an analyst’s ability to increase the EPS forecasts of SOEs without making investors wary of such distortions. Finally, like Lin and McNichols (1998) and Bessler and Stanzel (2009), we find that Forecast horizon has a significant positive effect on FE, in both subsamples. The longer the time interval between the forecast release date and the end of the fiscal year to which the forecast relates, the less reluctant analysts are to inflate their forecasts. As a further direct test of Hypothesis 3B, we divide the EPS forecasts for each subsample into forecasts on the EPS of the same fiscal year as the forecasting year and forecasts on the EPS of fiscal years thereafter. We focus on the forecasts released in the 90-day institutional lock-up period, which is when post-IPO price-supporting incentives likely are strongest. We run a Heckman (1976) two-step regression model for this purpose. Table 7 shows the results. To simplify the output, we no longer include the first-step results in the table; they resemble those in Tables 5 and 6. We also no longer report the results as to forecast accuracy, as they are again comparable to those for the forecast error. The regression output once more confirms our conjectures. For IPOs before June 10, 2009, we cannot find any difference in the EPS forecasts of affiliated vs. unaffiliated analysts, no matter whether those forecasts relate to the forecasting year or to a later year. For IPOs after June 10, 2009, both the same-year and later-year forecasts issued by affiliated analysts are significantly more positive, yet the magnitude of the effect is hugely different. The later-year EPS forecasts of affiliated analysts are 106 percentage points bigger on average, while this relative bias equals only 19 percentage points for the nearby EPS forecasts. Besides, the significant negative coefficient on the interaction between Affiliated 28


dummy and Relative price in the later-year regressions for IPOs after June 10, 2009 emphasizes once again that affiliated analysts strongly distort their EPS forecasts especially when the issuer’s stock price dropped after the IPO. This effect on nearby forecasts is also significant, yet is generally smaller. Next, we perform a split-sample regression analysis on the subsamples of issuers whose fraction of IPO stock allocated to institutional investors is above vs. below the sample median (not shown). Our earlier FE/FA results arise in both subsamples. And our results are not affected after adding the variable Institutional allocation to the initial models, while Institutional allocation is never significant. Finally, we re-estimate all models for the subsamples of private-controlled and state-controlled issuers, respectively. Our earlier findings for Affiliated dummy and its interaction with Relative price again arise in both subsamples. Interestingly, as information asymmetries (captured by Assets and Abnormal return) grow larger, analysts’ forecast errors now tend to increase as well, particularly for the private-controlled issuers. For those private firms, larger IPO underpricing thus leads to more upward-distorted EPS forecasts after all in the first listing year, yet only after June 10, 2009.

5. Conclusions Prior research has shown that financial analysts may face incentives to distort their earnings forecasts to help their employers to attract future corporate finance business, generate trading revenue, cultivate good relationships with firm management, or a combination of these. Yet this earlier research is vague as to how and why affiliated analysts distort their forecasts even more than unaffiliated analysts do. We address this research question by examining how IPO underwriting in China influences the output of analysts affiliated with the IPO lead manager. In so doing, we answer the call of Ramnath et al. (2008, p. 62): “Further research is needed to sort out the effects of affiliation and investment banking on analysts’ optimism/pessimism.”

29


We find support for the view that analysts affiliated with the IPO lead manager strategically distort their earnings forecasts upward so as to support the issuer’s stock price in the aftermarket. In contrast, we find no evidence that affiliated analysts’ optimism or the information advantage that they are presumed to enjoy can explain their forecast bias/accuracy. To examine the above ideas, we relied on a major regulatory change in the Chinese IPO market, which offered a natural experiment. To the best of our knowledge, such an exogenous shock did not arise in any other major IPO market, where underwriters have long taken full responsibility as to IPO offer prices. We find that affiliated analysts exhibit no difference in their forecast error and forecast accuracy when compared to unaffiliated analysts for IPOs before June 10, 2009, when the Chinese regulator largely set IPO offer prices. However, for IPOs after that date, affiliated analysts produce more positively biased forecasts, and their forecast accuracy is also significantly lower. Their relative bias is even bigger during the 90-day institutional lock-up period, when aftermarket price support likely is most valuable. Affiliated analysts’ relative accuracy is also worse in this timeframe. We further note that, when distorting their EPS forecasts, affiliated analysts, in particular, tend to inflate their forecasts for fiscal years further away from the forecast release date. Our study contributes to the financial analysts’ literature by presenting and verifying self-serving behavior on the part of affiliated analysts when releasing their opinion, that is to offer post-IPO price support. We also add to the IPO literature by pointing out an alternative mechanism underwriters can use to deliver aftermarket price support, especially in a setting without or with only limited quiet-period regulation. The results from our research have several policy implications. First, like in the United States before the Global Settlement was reached on April 28, 2003, the impact of affiliated analysts’ selfserving behavior on their EPS forecasts proves substantial in China. The Chinese regulator should therefore consider following the example of the SEC, investigating such behavior, and enforcing a 30


clearer separation between the corporate finance department and research department of investment banks. It might also consider introducing quiet-period regulation. Second, we notice that, by the end of 2012, independent research firms remain scarce in China. As analysts working at independent research institutes likely produce more impartial investment advice, the regulator might therefore design policies that support the development of these firms. We would also advise investors to check analysts’ affiliations before reading and acting on the recommendations in their research reports. Likewise, since affiliated analysts tend to inflate primarily their long-term EPS forecasts, we suggest that investors should be particularly cautious when relying on those forecasts. Finally, we recommend further academic research on the compensation scheme of analysts employed by investment banks in China. As we find, the behavior of financial analysts in China seems largely determined by their employers. It is therefore imperative to investigate whether and how the compensation scheme used by Chinese investment banks induces wrong incentives on the part of their analysts. Future research might also examine under what conditions investment banks prefer positively biased investment advice over other instruments to support an issuer’s stock price in the aftermarket, not only in the Chinese context.

31


References Aggarwal, R. (2000). Stabilization activities by underwriters after initial public offerings. J. Financ., 55(3), 1075−1103. Agrawal, A., & Chen, M. A. (2008). Do analyst conflicts matter? Evidence from stock recommendations. J. Law Econ., 51(3), 503−537 Beatty, R. P., & Ritter, J. R. (1986). Investment banking, reputation, and the underpricing of initial public offerings. J. Financ. Econ., 15(1), 213−232. Bessler, W., & Stanzel, M. (2009). Conflicts of interest and research quality of affiliated analysts in the German universal banking system: Evidence from IPO underwriting. Eur. Financial Manage., 15(4), 757−786. Bergman, N. K., & Roychowdhury, S. (2008). Investor sentiment and corporate disclosure. J. Acc. Res., 46(5), 1057−1083. Chi, J., & Padgett, C. (2005). Short-run underpricing and its characteristics in Chinese initial public offering (IPO) markets. Res. Int. Bus. Financ., 19(1), 71−93. Cliff, M. T., & Denis, D. J. (2004). Do initial public offering firms purchase analyst coverage with underpricing? J. Financ., 59(6), 2871−2901. Cowen, A., Groysberg, B., & Healy, P. (2006). Which types of analyst firms are more optimistic? J. Acc. Econ., 41(1), 119−146. Das, S., & Saudagaran, S. M. (1998). Accuracy, bias, and dispersion in analysts’ earnings forecasts: The case of cross-listed foreign firms. J. Int. Financ. Manage. Acc., 9(1), 16−33. Dugar, A., & Nathan, S. (1995). The effect of investment banking relationships on financial analysts’ earnings forecasts and investment recommendations. Contemp. Acc. Res., 12(1), 131−160. Dunbar, C. G. (2000). Factors affecting investment bank initial public offering market share. J. Financ. Econ., 55(1), 3–41. Eccles, R. G., & Crane, D. (1988). Doing deals: Investment banks at work. Boston: Harvard Business School Press. Fang, L., & Yasuda, A. (2009). The effectiveness of reputation as a disciplinary mechanism in sell-side research. Rev. Financ. Stud., 22(9), 3735−3777. Francis, B. B., Hasan I., & Sun, X. (2009). Political connections and the process of going public: Evidence from China. J. Int. Money Financ., 28(4), 696–719. Gao, Y. (2010). What comprises IPO initial returns: Evidence from the Chinese market. Pac. Basin Financ. J., 18(1), 77–89. Guo, H., & Brooks, R. (2008). Underpricing of Chinese A-share IPOs and short-run underperformance under the approval system from 2001 to 2005. Int. Rev. Financ. Anal., 17(5), 984–997. Hanley, K. W., Kumar, A. A., & Seguin, P. J. (1993). Price stabilization in the market for new issues. J. Financ. Econ., 34(2), 177−197. Heckman, J. J. (1976). The common structure of statistical models of truncation, sample selection and limited dependent variables and a simple estimator for such models. In Annals of Economic and Social Measurement, 5(4), 475−492. Hribar, P., & McInnis, J. (2012). Investor sentiment and analysts’ earnings forecast errors. Manage. Sci., 58(2), 293–307. Huyghebaert, N., & Xu, W. D. (2015). What determines the market share of investment banks in Chinese domestic IPOs? China Econ. Rev., forthcoming. Huyghebaert, N., & Wang, L. (2012). Expropriation of minority investors in Chinese listed firms: The role of internal and external corporate governance mechanisms. Corp. Gov., 20(3), 308–332. Jackson, A. R. (2005). Trade generation, reputation, and sell side analysts. J. Financ., 60(2), 673−717. 32


Jacob, J., Rock, S., & Weber, D. P. (2008). Do non-investment bank analysts make better earnings forecasts? J. Acc. Audit Financ., 23(1), 23−61. James, C., & Karceski, J. (2006). Strength of analyst coverage following IPOs. J. Financ. Econ., 82(1), 1−34. Kolasinski, A. C., & Kothari, S. P. (2008). Investment banking and analyst objectivity: Evidence from analysts affiliated with mergers and acquisitions advisors. J. Financ. Quan. Anal., 43(4), 817–842. Krigman, L., Shaw, W. H., & Womack, K. L. (2001). Why do firms switch underwriters?. J. Financ. Econ., 60(2), 245−284. Lewellen, K. (2006). Risk, reputation, and IPO price support. J. Financ., 61(2), 613−653. Li, L., & Fleisher, B. M. (2004). Heterogeneous expectations and stock prices in segmented markets: Application to Chinese firms. Q. Rev. Econ. Financ., 44(4), 521−538. Lim, T. (2001). Rationality and analysts’ forecast bias. J. Financ., 56(1), 369−385. Lin, H. W., & McNichols, M. F. (1998). Underwriting relationships, analysts’ earnings forecasts and investment recommendations. J. Acc. Econ., 25(1), 101−127. Ljungqvist, A. P., Nanda V., & Singh, R. (2006). Hot markets, investor sentiment, and IPO pricing. J. Bus., 79(4), 1667–1702.
Ljungqvist, A. P., Marston, F., Starks, L. T., Wei, K. D., & Yan, H. (2007). Conflicts of interest in sellside research and the moderating role of institutional investors. J. Financ. Econ., 85(2), 420−456. Loughran, T., & Ritter, J. R. (2004). Why has IPO underpricing changed over time? Financ. Manage., 33(3), 5–37. Mehran, H., & Stulz, R. M. (2007). The economics of conflicts of interest in financial institutions. J. Financ. Econ., 85(2), 267−296. Michaely, R., & Womack, K. L. (1999). Conflict of interest and the credibility of underwriter analyst recommendations. Rev. Financ. Stud., 12(4), 653−686. Morrison, A. D., & Wilhelm Jr, W. J. (2007). Investment banking: Institutions, politics, and law. Oxford University Press. Ramnath, S., Rock, S., & Shane, P. (2008). The financial analyst forecasting literature: A taxonomy with suggestions for further research. Int. J. Forecasting, 24(1), 34−75. Rock, K. (1986). Why new issues are underpriced. J. Financ. Econ., 15(1), 187−212. Su, D., & Fleisher, B. M. (1999). An empirical investigation of underpricing in Chinese IPOs. Pac. Basin Financ. J., 79(2), 173−202. Tian, L. (2011). Regulatory underpricing: Determinants of Chinese extreme IPO returns. J. Empir. Financ., 18(1), 78–90. Walther, B. R., & Willis, R. H. (2013). Do investor expectations affect sell-side analysts’ forecast bias and forecast accuracy?. Rev. Account. Stud., 18(1), 207−227.

33


Figure 1 Trading volume and cumulative abnormal return around the expiration of the lock-up period





























Average cumulative abnormal return

Average relative daily trading volume

This figure shows the average relative daily trading volume and the average cumulative abnormal return from ten days before to ten days after the expiration of the institutional lock-up period, for the 304 firms that became listed after June 10, 2009 and that were followed by at least one affiliated analsyt. The Relative daily trading volume is the daily number of shares traded divided by the average daily number of shares traded from the 2d trading day to the 11th day before the end of the institutional lock-up period. Likewise, the Daily abnormal return is calculated using the daily return minus the average daily return on the stock from the 2d trading day to the 11th day before the end of the institutional lock-up period.

                 Trading days; 0 = End of institutional lock-up period Average relative daily trading volume Average cumulative abnormal return

34


360 3,209 3,569 1,744 16,186 17,930 2,104 19,395 21,499

Number of observations

1.5942 1.0567 1.1289 1.6143 1.2867 1.3354 1.6108 1.2486 1.3012

Mean FE

0.2043 0.1517 0.1650 0.7142 0.5228 0.5398 0.6823 0.3809 0.4012

Median FE

3.2011 2.2890. 2996 2.1062 2.4893 2.1016 2.1454 2.6012 2.1433 2.1457

Std. Dev.

<0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001

p-value of t-test

p-value

<0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001

test

Wilcoxon

1.6520 1.1254 1.1972 1.6716 1.3465 1.3956 1.6682 1.3099 1.3626

Mean FA

0.2043 0.2187 0.2197 0.7142 0.5228 0.5398 0.6823 0.3809 0.4012

Median FA

3.0141 2.1089 2.0236 2.3564 2.0101 2.0568 2.5873 2.0952 2.0530

Std. Dev.

<0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001

p-value of t-test

Table 2 Comparison of forecast error by affiliated and unaffiliated analysts on the same IPO firm and forecast year

Affiliated Unaffiliated Total Affiliated Unaffiliated Total Affiliated Unaffiliated Total

Type of analyst

<0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001

test

p-value of Wilcoxon




IPOs in June 10, 2009–2011

IPOs in 2004–June 9, 2009

Period

IPO year Later years Total IPO year Later years Total

Forecast year

23 76 99 153 347 500

Number of observations

0.0390 -0.0061 0.0044 0.0180 0.2141 0.1540

Mean relative bias

0.0192 0.0053 0.0092 0.0020 0.0511 0.0132

Median relative bias

0.4715 0.3644 0.3429 0.2891 1.4622 1.2210

Std. Dev.

-0.5740 -0.9721 -0.9721 -1.0320 -5.0573 -5.0573

Min.

0.9621 1.0530 1.0530 2.6064 6.2486 6.2486

Max.

0.4719 0.8787 0.8842 0.4394 0.0077 0.0031

p-value of t-test

35

0.8282 0.3752 0.4527 0.7195 0.0200 0.0531

p-value of Wilcoxon test

This table presents summary statistics for the comparison of the forecast error across affiliated and unaffiliated financial analysts (Relative bias). We categorize earnings forecasts by IPO firm and by forecast year. For each IPO firm and forecast year, we subtract the average forecast error made by unaffiliated analysts from the average forecast error made by affiliated analysts to obtain the Relative bias. We also split the full sample into firms that became listed before June 10, 2009 and firms that became listed thereafter. For each subsample, we further divide the sample into forecasts on the EPS of the IPO year (IPO year) and forecasts on the EPS of fiscal years thereafter (Later years). We use a t-test and a Wilcoxon rank-sum test to examine the null hypothesis that the mean and median Relative bias equal zero, respectively.




IPOs in 2004–2011

IPOs in June 10, 2009–2011

IPOs in 2004–June 9, 2009

Period

This table presents summary statistics on forecast error (FEi,j,t) and forecast accuracy (FAi,j,t). FE,j,i,t is defined as (FEPSj,i,t – AEPSi,t)/|AEPSi,t|, where FEPSj,i,t is the forecast of earnings per share on firm i in year t by analyst j. AEPSi,t is the reported earnings per share by firm i for year t. FE is winsorized at 5% and 95%, while FA is winsorized at 95%. All forecasts are released in the first year after IPO. We split the full sample into firms that became listed before June 10, 2009 and firms that became listed thereafter. We use a t-test and a Wilcoxon ranksum test to examine the null hypothesis that the mean and median FE and FA equal zero, respectively.

Table 1 Summary statistics on forecast error and forecast accuracy

Table 3 Cumulative abnormal return from releasing a positively biased earnings forecast This table presents summary statistics on the three-day cumulative abnormal return (CAR) after a positively biased earnings forecast was released. The daily abnormal return is obtained by deducting the market return in either Shanghai or Shenzhen from the issuer’s stock return on the same trading day. The three-day window starts at one trading day before the forecast release date and ends one day thereafter. If the forecast release date is not a trading day, the nearest next trading day is used as forecast release date to calculate the CAR. Panel A reports the results for the full sample as well as the subsamples of the firms that became listed before June 10, 2009 and the firms that became listed thereafter. Panel B further shows the results after splitting the sample and subsamples into forecasts released by an affiliated analyst (Affiliated CAR) and forecasts released by an unaffiliated analyst (Unaffiliated CAR). In Panel A, we use a t-test and a Wilcoxon rank-sum test to examine the null hypothesis that the mean and median CAR equal zero, respectively. In Panel B, we use these same tests to examine the null hypothesis that the difference in CAR across affiliated and unaffiliated analysts equals zero.

Panel A: Period

Obs.

Mean

Median

Std. Dev.

p-value of t-test

IPOs in 2004–June 9, 2009 IPOs in June 10, 2009–2011

3,083 15,530

0.0031 0.0068

0.0022 0.0053

0.0641 0.0573

0.0001 0.0001

p-value of Wilcoxon test 0.0050 0.0002

IPOs in 2004–2011

18,613

0.0062

0.0045

0.0450

0.0001

0.0005

p-value of t-test

p-value of Wilcoxon test

Panel B: Period

Affiliated CAR

Unaffiliated CAR

Obs.

Mean

Median

Obs.

Mean

Median

IPOs in 2004–June 9, 2009 IPOs in June 10, 2009–2011

293 1,522

0.0032 0.0080

0.0041 0.0040

2,790 14,008

0.0031 0.0067

0.0019 0.0050

0.5201 0.2752

0.4215 0.6105

IPOs in 2004–2011

1,815

0.0072

0.0042

16,798

0.0061

0.0040

0.3023

0.8401

36


Table 4 Summary statistics for the variables used in the regressions This table presents summary statistics on all variables used in the forecast-error and forecast-accuracy regressions. In Panels A and B, the summary statistics cover the two subsamples: firms that became listed before June 10, 2009 and firms that became listed thereafter, respectively. FE is the forecast error, FA is forecast accuracy. Affiliated dummy equals one if the forecast was issued by an analyst affiliated with the IPO lead manager and zero otherwise. Relative price is calculated by subtracting the IPO offer price from the aftermarket stock price on the day before the forecast release date, divided by the IPO offer price. Analyst ranking equals one if the analyst is ranked as a star analyst by The New Fortune in the year before the forecast release and zero otherwise. Analyst experience is the log of one plus the total number of forecasts issued by the analyst before the current forecast. Market share is the analyst’s investment bank market share in IPO underwriting in the previous year. Number of analysts is the total number of analysts employed by the analyst’s investment bank at the beginning of the forecasting year. Fee rate is the IPO fee rate (i.e. the underwriting fee plus sponsor fee divided by the IPO gross proceeds) charged by the lead manager in the firm’s IPO. Abnormal return is the first-day abnormal return calculated as (first-day closing price – offer price)/offer price – first-day market return. SEO dummy equals one if the IPO firm made a seasoned equity offering in the three years after listing and zero otherwise. Assets is the log of total assets before IPO. SOE dummy equals one if the Chinese state controls more than 50% of the firm’s direct and indirect voting rights before IPO and zero otherwise. Forecast horizon is the number of days between the forecast release and the end of the forecast period.

Panel A: IPOs in 2004–June 9, 2009 Variable Number of observations 3,098 FE 3,098
FA 3,098
Affiliated dummy 3,098
Relative price 3,098
Analyst ranking 3,098
Analyst experience 3,098
Market share 3,098
Number of analysts 3,098
Fee rate 3,098
Abnormal return 3,098 SEO dummy 3,098
Assets 3,098
SOE dummy 3,098
Forecast horizon Panel B: IPOs in June 10, 2009–2011 Variable Number of observations 12,838 FE 12,838
FA 12,838
Affiliated dummy 12,838
Relative price 12,838
Analyst ranking 12,838
Analyst experience 12,838
Market share 12,838
Number of analysts 12,838
Fee rate 12,838
Abnormal return 12,838 SEO dummy 12,838
Assets 12,838
SOE dummy 12,838
Forecast horizon

Mean

Median

Std. Dev.

Min.

Max.

1.0804 1.1518 0.0881 1.0043 0.1094 3.7912 0.0814 34.9435 0.0222 0.8761 0.2541 24.5726 0.8260 511.4078

0.1521 0.2295

2.0964 4.8815

-0.2996 0.0000

7.6345 7.6410

0.6392

1.3795

-0.6193

11.5296

4.1700 0.0132 34.0000 0.0180 0.7781

1.1889 0.0996 16.4920 0.0128 0.6022

0.0000 0.0000 3.0000 0.0075 -0.0123

6.6800 0.3345 86.0000 0.0741 3.7934

25.6861

3.0831

18.7133

29.4966

484.0000

375.9846

-35.0000*

1929.0000

Mean

Median

Std. Dev.

Min.

Max.

1.3474

0.5301

1.9974

-0.2996

7.6345

1.4023 0.0986 0.3042 0.1558 5.3601 0.0542 40.5027 0.0462 0.2875 0.0991 20.8798 0.2260 372.5381

0.5681

3.2071

0.0000

7.6410

0.1731

0.4864

-0.5800

4.8422

5.2900 0.0392 38.0000 0.0450 0.2251

1.1871 0.0636 19.9458 0.0187 0.3182

0.0000 0.0000 1.0000 0.0082 -0.1864

8.3101 0.3438 92.0000 0.1330 1.8101

20.3457

1.9237

18.3506

29.8148

338.0000

266.2988

-28.0000*

1218.0000


As Chinese firms publish their audited financial statements in March or April of the subsequent year, forecasts can still be made after the end of the forecast period.


37


Table 5 The forecast-error regressions This table shows the regression results on forecast error (FE) using a simple OLS regression model and using Heckman’s (1976) two-step regression model. Panel A reports the results for the firms that became listed before June 10, 2009; Panel B then reports the results for the firms that became listed thereafter. In both panels, we present the results using all forecasts issued in the first year after IPO (One year) and using all forecasts issued in the first 90 days after IPO (90 days). For each set of analyses, the first column shows the results using simple OLS, the subsequent two columns show the Heckman (1976) two-step results. The test and control variables are defined in Table 4. The regression models also include Industry dummies based on the CSRC industry classification code and Year dummies. For the first-step Heckman regressions, we run a probit regression model on whether the IPO firm is followed by at least one analyst affiliated with the IPO lead manager. The explanatory variables in this regression model include the number of analysts employed by the IPO underwriter at the beginning of the IPO year (Underwriter analyst number), the issuing P/E ratio minus the market P/E ratio (Pricing aggressiveness), Fee rate, Abnormal return, SEO dummy, Assets, and SOE dummy. We also include Industry dummies and Year dummies. We calculate the inverse Mills ratio based upon this probit model and include it in the second-step regression. The variables used in the second-step regression are the same as those in the OLS model. We always cluster the errors by financial analysts. Coefficients significant at the 1%, 5% and 10% level are indicated with ***, **, *, respectively.

Panel A: IPOs in 2004–June 9, 2009 OLS Intercept

5.7848*** (0.00)

Underwriter analyst number Pricing aggressiveness Affiliated dummy Affiliated dummy * Relative price Relative price Analyst ranking Analyst experience Market share Number of analysts Fee rate Abnormal return SEO dummy Assets SOE dummy Forecast horizon

0.2928 (0.19) -0.1248 (0.37) 0.1801 (0.15) -0.2524 (0.24) 0.0546 (0.25) -2.5542*** (0.00) 0.0014 (0.68) -4.4755 (0.55) -0.0182 (0.89) 0.9721*** (0.00) -0.1888*** (0.00) -1.6932*** (0.00) 0.0012*** (0.00)

Inverse Mills ratio Industry dummies Year dummies p-value of F-test/Chi2-test Adjusted R-square Number of observations

Yes Yes 0.00 0.32 3,098

One year Heckman First step Second step -5.3017*** 7.5885*** (0.00) (0.00) 0.0210*** (0.00) 0.0130*** (0.00) 0.2040 (0.15) -0.1278 (0.20) 0.1721 (0.16) -0.2512 (0.24) 0.0604 (0.15) -2.7718*** (0.00) 0.0015 (0.50) 12.6818*** -8.7861 (0.00) (0.40) 0.1863*** 0.0344 (0.00) (0.68) 0.3756*** 0.9904*** (0.00) (0.00) 0.3256*** -0.2551*** (0.00) (0.00) 0.2099*** -1.6821*** (0.00) (0.00) 0.0012*** (0.00) 0.3782 (0.20) Yes Yes Yes Yes 0.00 0.00 0.44 5,647 3,098

OLS 1.8927 (0.50)

-0.2002 (0.37) 0.1503 (0.65) 0.1571 (0.40) -0.2174 (0.31) 0.0074 (0.91) -2.5331** (0.04) -0.0036 (0.15) 14.9531 (0.40) -0.2684 (0.40) 0.5686** (0.04) 0.0479 (0.64) -2.0838*** (0.00) 0.0012*** (0.00)

Yes Yes 0.00 0.37 576

90 days Heckman First step Second step -7.2600*** 1.2600 (0.00) (0.62) 0.0296*** (0.00) 0.0130** (0.04) -0.4191 (0.33) 0.1702 (0.59) 0.1039 (0.61) -0.1791 (0.49) -0.0128 (0.84) -2.2160** (0.01) -0.0062 (0.47) 6.0118** 9.0780 (0.03) (0.54) 0.0169 -0.3012 (0.92) (0.15) 0.4864*** 0.7697*** (0.00) (0.00) 0.4403*** 0.0109 (0.00) (0.91) 0.8221*** -1.6842*** (0.00) (0.00) 0.0013*** (0.00) 0.5576* (0.06) Yes Yes Yes Yes 0.00 0.00 0.53 1,097 576

38


Panel B: IPOs in June10, 2009–2011 OLS Intercept

1.2973* (0.07)

Underwriter analyst number Pricing aggressiveness Affiliated dummy Affiliated dummy * Relative price Relative price Analyst ranking Analyst experience Market share Number of analysts Fee rate Abnormal return SEO dummy Assets SOE dummy Forecast horizon

0.3298** (0.03) 0.0662 (0.78) 0.3093 (0.21) -0.0835 (0.39) 0.0460 (0.75) -2.2347*** (0.00) -0.0015 (0.31) 1.7307* (0.07) -0.2259 (0.17) 0.4955*** (0.00) -0.0994*** (0.00) -0.0250 (0.83) 0.0039*** (0.00)

Inverse Mills ratio Industry dummies Year dummies p-value of F-test/Chi2-test Adjusted R-square Number of observations

Yes Yes 0.00 0.28 12,838

One year Heckman First step Second step -8.2861*** 1.8490*** (0.00) (0.00) 0.0234*** (0.00) 0.0109*** (0.00) 0.3441*** (0.00) 0.0389 (0.68) 0.3003 (0.20) -0.0749* (0.09) 0.0501 (0.70) -2.1411*** (0.00) -0.0015* (0.06) 2.0802*** 0.5090 (0.00) (0.62) 0.2237*** -0.2042 (0.00) (0.20) 0.0009 0.4510*** (0.99) (0.00) 0.2856*** -0.1260*** (0.00) (0.00) 0.1240*** -0.0818 (0.00) (0.26) 0.0039*** (0.00) 0.1112* (0.07) Yes Yes Yes Yes 0.00 0.00 0.22 20,369 12,838

OLS -3.1269** (0.05)

0.6308*** (0.01) -0.9300** (0.03) 0.3885 (0.14) -0.0386 (0.81) 0.1206 (0.15) -2.1985** (0.04) -0.0015 (0.57) 5.3303** (0.05) -0.2010 (0.36) 0.5338*** (0.00) 0.0403 (0.53) -0.2178* (0.07) 0.0042*** (0.00)

Yes Yes 0.00 0.34 2,716

90 days Heckman First step Second step -9.5669*** -2.5359** (0.00) (0.03) 0.0269*** (0.00) 0.0060*** (0.00) 0.6805*** (0.00) -1.0573*** (0.00) 0.3990 (0.21) -0.0068 (0.95) 0.1242 (0.20) -1.1951 (0.13) -0.0011 (0.51) 3.8829*** 5.0335** (0.01) (0.03) 0.3255*** -0.1820 (0.00) (0.27) 0.2446*** 0.4182*** (0.00) (0.00) 0.3300*** -0.0093 (0.00) (0.82) 0.2640*** -0.0851 (0.00) (0.21) 0.0042*** (0.00) 0.1910** (0.04) Yes Yes Yes Yes 0.00 0.00 0.23 4,381 2,716

39


Table 6 The forecast-accuracy regressions This table shows the regression results on forecast accuracy (FA) using a simple OLS regression model and using Heckman’s (1976) two-step regression model. Panel A reports the results for the firms that became listed before June 10, 2009; Panel B then reports the results for the firms that became listed thereafter. In both panels, we present the results using all forecasts issued in the first year after IPO (One year) and using all forecasts issued in the first 90 days after IPO (90 days). For each set of analyses, the first column shows the results using simple OLS, the subsequent two columns show the Heckman (1976) two-step results. The test and control variables are defined in Table 4. The regression models also include Industry dummies based on the CSRC industry classification code and Year dummies. For the first-step Heckman regressions, we run a probit regression model on whether the IPO firm is followed by at least one analyst affiliated with the IPO lead manager. The explanatory variables in this regression model include the number of analysts employed by the IPO underwriter at the beginning of the IPO year (Underwriter analyst number), the issuing P/E ratio minus the market P/E ratio (Pricing aggressiveness), Fee rate, Abnormal return, SEO dummy, Assets, and SOE dummy. We also include Industry dummies and Year dummies. We calculate the inverse Mills ratio based upon this probit model and include it in the second-step regression. The variables used in the second-step regression are the same as those in the OLS model. We always cluster the errors by financial analysts. Coefficients significant at the 1%, 5% and 10% level are indicated with ***, **, *, respectively.

Panel A: IPOs in 2004–June 9, 2009 OLS Intercept

6.7850*** (0.00)

Underwriter analyst number Pricing aggressiveness Affiliated dummy Affiliated dummy * Relative price Relative price Analyst ranking Analyst experience Market share Number of analysts Fee rate Abnormal return SEO dummy Assets SOE dummy Forecast horizon

0.2950 (0.20) -0.1278 (0.35) 0.1902 (0.17) -0.6734** (0.02) 0.0501 (0.28) -2.7621*** (0.00) 0.0010 (0.81) -4.4012 (0.50) -0.0170 (0.79) 0.9621*** (0.00) -0.1932*** (0.00) -1.7135*** (0.00) 0.0011*** (0.00)

Inverse Mills ratio Industry dummies Year dummies p-value of F-test/Chi2-test Adjusted R-square Number of observations

Yes Yes 0.00 0.31 3,098

One year Heckman First step Second step -5.3017*** 7.8404*** (0.00) (0.00) 0.0210*** (0.00) 0.0130*** (0.00) 0.2121 (0.15) -0.1330 (0.21) 0.1870 (0.17) -0.7196** (0.02) 0.0721 (0.15) -2.9300*** (0.00) 0.0012 (0.31) 12.6818*** -8.531 (0.00) (0.37) 0.1863*** -0.0314 (0.00) (0.68) 0.3756*** 0.9721*** (0.00) (0.00) 0.3256*** -0.2735*** (0.00) (0.00) 0.2099*** -1.7026*** (0.00) (0.00) 0.0011*** (0.00) 0.3610 (0.20) Yes Yes Yes Yes 0.00 0.00 0.44 5,647 3,098

OLS 2.1209 (0.35)

-0.2312 (0.30) 0.1212 (0.61) 0.1871 (0.39) -0.2520 (0.29) 0.0090 (0.80) -2.721** (0.05) -0.0021 (0.27) 14.9651 (0.29) -0.2610 (0.36) 0.5486** (0.05) 0.0370 (0.78) -2.1390*** (0.00) 0.0012*** (0.00)

Yes Yes 0.00 0.39 576

90 days Heckman First step Second step -7.2600*** 1.2932 (0.00) (0.43) 0.0296*** (0.00) 0.0130** (0.04) -0.4390 (0.28) 0.1895 (0.61) 0.1200 (0.76) -0.1681 (0.78) -0.0209 (0.55) -2.5320** (0.01) -0.0039 (0.40) 6.0118** 9.3760 (0.03) (0.32) 0.0169 -0.2917 (0.92) (0.19) 0.4864*** 0.7370*** (0.00) (0.00) 0.4403*** 0.0123 (0.00) (0.71) 0.8221*** -1.6511*** (0.00) (0.00) 0.0013*** (0.00) 0.5620* (0.06) Yes Yes Yes Yes 0.00 0.00 0.53 1,097 576

40


Panel B: IPOs in June10, 2009–2011 OLS Intercept

1.3210* (0.06)

Underwriter analyst number Pricing aggressiveness Affiliated dummy Affiliated dummy * Relative price Relative price Analyst ranking Analyst experience Market share Number of analysts Fee rate Abnormal return SEO dummy Assets SOE dummy Forecast horizon

0.3521** (0.03) 0.0632 (0.80) 0.3130 (0.21) -0.2981* (0.08) 0.0470 (0.75) -2.3210*** (0.00) -0.0012 (0.29) 1.9607* (0.06) -0.2241 (0.17) 0.4890*** (0.00) -0.0965*** (0.00) -0.0231 (0.89) 0.0038*** (0.00)

Inverse Mills ratio Industry dummies Year dummies p-value of F-test/Chi2-test Adjusted R-square Number of observations

Yes Yes 0.00 0.21 12,838

One year Heckman First step Second step -8.2861*** 1.970*** (0.00) (0.00) 0.0234*** (0.00) 0.0109*** (0.00) 0.3761*** (0.00) 0.0360 (0.79) 0.3214 (0.25) -0.3210* (0.06) 0.0520 (0.65) -2.2120*** (0.00) -0.0013* (0.07) 2.0802*** 0.5085 (0.00) (0.62) 0.2237*** -0.1760 (0.00) (0.25) 0.0009 0.4328*** (0.99) (0.00) 0.2856*** -0.1109** (0.00) (0.02) 0.1240*** -0.0790* (0.00) (0.09) 0.0038*** (0.00) 0.1087* (0.08) Yes Yes Yes Yes 0.00 0.00 0.22 20,369 12,838

OLS -3.1001** (0.05)

0.6710*** (0.01) -0.9090** (0.04) 0.4108 (0.17) -0.0371 (0.81) 0.1310 (0.17) -2.3009** (0.04) -0.0012 (0.60) 5.6908** (0.05) -0.1951 (0.38) 0.5176*** (0.00) 0.0352 (0.60) -0.2153* (0.07) 0.0042*** (0.00)

Yes Yes 0.00 0.35 2,716

90 days Heckman First step Second step -9.5669*** -2.2515** (0.00) (0.04) 0.0269*** (0.00) 0.0060*** (0.00) 0.6991*** (0.00) -0.9760*** (0.00) 0.4210 (0.19) -0.0059 (0.97) 0.1300 (0.20) -1.3012* (0.10) -0.0014 (0.31) 3.8829*** 5.1294** (0.01) (0.03) 0.3255*** -0.1654 (0.00) (0.29) 0.2446*** 0.3980*** (0.00) (0.00) 0.3300*** -0.0087 (0.00) (0.83) 0.2640*** -0.0838 (0.00) (0.28) 0.0042*** (0.00) 0.2076** (0.05) Yes Yes Yes Yes 0.00 0.00 0.23 4,381 2,716

41


Table 7 The forecast error regressions using the forecasts for different fiscal years This table shows the regression results on forecast error (FE) using Heckman’s (1976) two-step regression model. All EPS forecasts are issued in the first 90 days after IPO. We divide the full sample into firms that became listed before June 10, 2009 and firms that became listed thereafter. For each subsample, we further divide the sample into forecasts relating to the EPS of the same fiscal year as the forecasting year (Same year) and forecasts relating to the EPS of fiscal years thereafter (Later years). The test and control variables are defined in Table 4. The regression models also include Industry dummies based on the CSRC industry classification code and Year dummies. We calculate the inverse Mills ratio based upon the probit model results shown in Table 5 and include it in the second-step regression. We always cluster the errors by financial analysts. Coefficients significant at the 1%, 5% and 10% level are indicated with ***, **, *, respectively.

IPOs in 2004–June 9, 2009 Same year Intercept Affiliated dummy Affiliated dummy * Relative price Relative price Analyst ranking Analyst experience Market share Number of analysts Fee rate Abnormal return SEO dummy Assets SOE dummy Forecast horizon Inverse Mills ratio Industry dummies Year dummies p-value of Chi2-test Number of observations

1.2971 (0.68) 0.1238 (0.82) 0.0076 (0.99) 0.3011 (0.26) -0.1590 (0.64) -0.0192 (0.83) -5.0791*** (0.00) -0.0020 (0.75) 15.8215 (0.45) 0.0315 (0.92) 0.7524** (0.02) -0.0720 (0.56) -0.4451 (0.44) 0.0033** (0.01) 1.2421** (0.03) Yes Yes 0.00 169

Later years 6.5100* (0.06) -0.8457 (0.16) 0.4738 (0.34) -0.1673 (0.54) -0.1961 (0.59) 0.0033 (0.97) -1.3480** (0.03) -0.0044 (0.51) -9.8191 (0.63) -0.0774 (0.15) 1.0748*** (0.00) -0.3760** (0.05) -1.7806*** (0.00) 0.0020*** (0.00) 1.2092*** (0.01) Yes Yes 0.00 407

IPOs in June 10, 2009–2011 Same year -0.5328 (0.31) 0.1864*** (0.00) -0.3044** (0.04) 0.1885** (0.01) 0.0072 (0.87) 0.0110 (0.45) -0.8995** (0.02) -0.0008 (0.31) 5.0216*** (0.00) -0.1977 (0.21) 0.0524 (0.36) 0.0070 (0.71) -0.0068 (0.90) 0.0031*** (0.00) 0.0307 (0.69) Yes Yes 0.00 1,050

Later years -2.3752 (0.23) 1.0582*** (0.00) -1.6137*** (0.00) 0.0653** (0.01) 0.0291 (0.86) 0.2060*** (0.00) -1.9532** (0.05) -0.0012 (0.68) 5.4487** (0.02) -0.3391 (0.24) 0.7441*** (0.00) -0.0404 (0.57) -0.1327* (0.07) 0.0037*** (0.00) 0.2689** (0.07) Yes Yes 0.00 1,666

42