Unexpected Information , Investor Attention , and the Contagion Effect of IPO Withdrawal *

We investigate whether firms’ withdrawals of their initial public offering (IPO) applications, as announced by the China Securities Regulatory Commission (CSRC) on 3 April 2013, lead to a contagion effect on the market reaction to other listed firms sponsored by the same agent. We find that listed firms that have the same sponsor as an “IPO withdrawing firm” face a negative market reaction, suggesting that firms’ IPO withdrawals make investors worried about the quality of sponsors and cause negative abnormal equity returns for listed firms with the same sponsor. We also find that this contagion effect is concentrated among listed firms with good accounting information because this event provides more unexpected negative information for investors. In addition, the contagion effect is concentrated among listed firms with more media or analyst coverage. This is because media and analyst coverage attracts investors’ limited attention, making them more pessimistic and finally leading to an increase in the contagion effect. Our study provides support for the contagion theory and investigates the effectiveness of the CSRC’s special checks on financial statements, which may give some suggestions for the healthy development of the IPO market.


I. Introduction
The securities issuance sponsorship system is a significant institutional arrangement aiming to assure the information quality of firms applying for initial public offerings (IPOs) in China. However, the frequent performance deterioration of IPO firms has resulted in widespread queries about the credibility of the sponsoring institutions. On 28 December 2012, the China Securities Regulatory Commission (CSRC) published the Notice of Conducting Special Checks on 2012 Financial Statements of IPO Firms, which provided that all IPO firms should complete self-inspection by 31 March 2013 and that the CSRC would conduct random inspections after that date. The IPO process of firms in which problems were detected would be terminated, and the responsible intermediaries would be punished. As a result, 166 firms withdrew their IPO applications, and this was announced by the CSRC on 3 April 2013. 2 Thereafter, the media extensively reported the withdrawals that impaired the reputation of the sponsoring institutions, as the following headlines illustrate: "IPO Withdrawal Tide: 76 firms give up within a week while 30% of GEM firms quit" (Securities Times); "IPO Financial Inspection Outcome: Minsheng, Everbright and Haitong most hurt" and "Everbright Securities' redemption from Sponsorgate" (Sina Finance). 3 Thus, this paper aims to investigate whether IPO withdrawal is contagious. Recent research defines the contagion effect as the phenomenon whereby an event that adversely affects one firm also has a similar influence on related firms . In this paper, the contagion effect of IPO withdrawal means that this event causes a negative market reaction towards the listed firms sponsored by the same agent as the "IPO withdrawing firm" (hereinafter "related firms"). This is because an IPO withdrawal probably suggests bad sponsorship quality, causing investors to worry about the accounting information quality of related firms and driving related firms' stock prices down during the window period. Unlike the research on market reaction or the signal effect of IPO withdrawal, our study focuses on related firms rather than the withdrawing firms or the sponsors. We divide the listed firms into related firms and unrelated firms on the basis of whether their sponsors are the same institution and compare their pre-and post-withdrawal cumulative abnormal return (CAR) value to identify the contagion effect. The empirical results show that the CAR of related firms declines and is significantly lower than that of unrelated firms during the event window, which confirms the contagion effect.
We further document how impairment of the sponsor's reputation, the accounting information quality of related firms, and investor attention affect the contagion effect of IPO withdrawal.
First, a sponsor's reputation plays a key role in guaranteeing information quality (Booth and Smith, 1986). Once the sponsor's reputation is damaged, the listed firms sponsored by it are undoubtedly implicated (Carter and Manaster, 1990). We use the number and fraction of IPO withdrawing firms backed by a sponsor to measure the degree of its reputation impairment. The larger the number or fraction of IPO withdrawing firms, the stronger the contagion effect faced by the related firms.
Second, the accounting information quality of related firms affects the degree of the contagion effect on them. On the one hand, when the quality of their ex ante information is high, the related firms may not be infected or may be less infected because the market trusts that the IPO firms' withdrawal would not change investors' expectations of the related firms.
On the other hand, the psychological gap and unexpected information are the main explanatory bases for the contagion effect . Sponsors' involvement in IPO withdrawal is incremental information to the investors of firms with higher information quality which may shake the market's trust and lead to negative response. We use the degree of earnings management, performance deterioration, and whether the firm is audited by a Big Ten firm to measure information quality. The results show that the contagion of IPO withdrawal mainly occurs in the firms with high information quality, indicating that sponsors involved in IPO withdrawal provide incremental information to the investors of high-quality firms.
Finally, investor attention also affects the degree of the contagion effect. Investors must follow the firms prior to making buy-in or sell-out decisions, and the level of attention will affect their decisions (Baber and Odean, 2008). In an IPO withdrawal event, the sponsor may be well known by investors if ex ante attention to the withdrawing firm is high enough, and investors may identify the related firms more easily. We use the coverage of analysts and the finance media to measure investor attention and find that the contagion of IPO withdrawal occurs more in firms that attract a high level of attention.
Our study contributes to several strands of the literature in the following ways. First, existing studies find that for firms in the same industry, having the same auditors or creditors can lead to the contagion effect among related firms. We document market reaction to related firms through the sponsor channel, indicating the information assurance function of sponsors in the IPO process. Thus, our study adds additional evidence to the literature with regard to the contagion effect by investigating a new contagion channel.
Second, we use an exogenous event -IPO withdrawal -to indirectly verify the market value of sponsor reputation, accounting information, and media coverage, which contributes to the related literature. Third, this study confirms the rationality and validity of the CSRC's special inspection of financial statements which inspired the reform of the IPO system and the orderly development of capital markets in China.

Literature on the Contagion Effect
The contagion effect is defined as the situation where events that adversely affect a firm also have a similar influence among firms related to it .  find that accounting restatements induce share price declines among non-restating firms in the same industry and that investors impose a larger contagion penalty on the stock prices of peer firms with high earnings and high accruals when peer and restating firms use the same external auditor.  find that bankruptcy announcements cause negative abnormal equity returns and increases in credit default swap (CDS) spreads for creditors and that this credit contagion becomes stronger when the bankruptcy firm holds larger leverage. By revealing information on a common fundamental factor and thereby affecting the behaviour of creditors, the failure of a single firm can trigger the failure of another firm .  finds that a bank's default may spread to other banks through inter-bank linkages and that the inter-bank market is conducive to financial contagion.  shows that in the case of the contagion of a liquidity crisis between two non-financial institutions that have the same creditor, the contagion effect of a liquidity crisis faced by a firm with a lower failure point is likely to be severer than that faced by a firm with a higher failure point. Bereskin and Cicero (2013) document that Delaware-incorporated firms with staggered boards and no outside blockholders increased the compensation of chief executive officers (CEOs) following the mid-1990s Delaware legal cases that strengthened their ability to resist hostile takeovers and that non-Delaware firms subsequently increased CEO compensation when the rulings affected a substantial number of firms in their industries. Chiu et al. (2013) find evidence of earnings management contagion in firms with interlocked boards. A firm is more likely to manage earnings when it shares a common director with a firm that is currently managing its earnings, and the contagion is stronger when the shared director has a leadership position. Francis and Michas (2013) document that audit offices with client restatements in the past are more likely to have new client restatements in the future, suggesting a contagion of audit failures over time, and that this effect can be mitigated by large office size and a relatively high use of industry expertise. Liu et al. (2014) find that the Peizhong Bai corruption case led to a stock price drop not only for the firm but also for other firms in the same industry due to information transmission. Moreover, government ownership has a leverage effect which strengthens the contagion in government-controlled firms. Elliot et al. (2014) find that integration strengthens the connection among business sections and the contagion effect, while the possibility of adverse events, such as bankruptcy and default, declines due to less dependency on the assets of a certain section. Diversification intensifies the independence of each section and reduces the contagion effect but increases the possibility of an adverse event occurring in each section.  document that peer firms begin managing earnings after an earnings restatement is announced by target firms in their industry or in their metropolitan statistical area.
In addition, there is a body of literature on a macro contagion effect, such as the contagion due to the significant increases in cross-market linkages (Caporale et al., 2005) and international investors co-holding cross-market assets (Boyer et al., 2006); therefore, a financial crisis can be contagious from one country or region to another Zhao et al., 2013). To be specific, Cespa and Foucault (2014) suggest that illiquidity contagion among stocks will lead to a significant decline in the whole market because stocks in the market are correlated. Wang et al. (2014) find that the contagion effect is dominant in  Cespa and Foucault (2014); Garleanu et al. (2015); Zhao et al. (2013); Wang et al. (2014) Economic crisis in one country or region Globalisation, cross-market transaction, or same investors Economic crisis in related country or region the Chinese credence goods market, indicating that ineffective supervision and public distrust in regulation are the primary reasons for the credence crisis. Bekaert et al. (2014) document that the US market risk has great contagion effect on stock portfolios in one industry, where the effect gets strengthened as the quality of economic fundamentals declines. Garleanu et al. (2015) find that events that adversely affect investors in one financial market induce similar movements in other markets across the world because the markets are related by investors and firms.
Similar to the literature on medical contagion, the existing literature describes the sources, channels, consequences, and determinants of contagion. In Table 1, contagion sources are defined as the adverse events occurring in a firm, which are usually financial restatement, debt crisis, operational risk, and so forth; a contagion channel is defined as the connection between the firm and the other firms, such as having the same director or auditor or being in the same industry; and contagion consequences are defined as adverse events occurring in related firms, in which the common indicators of consequence include earnings management, stock price reaction, and risk volatility. Such research investigates, through channel analyses, the circumstances under which the contagion effect can be strengthened.
Besides, the usual explanation of contagion is the market's concern about adverse events or the market panic effect.  claim that share price declines among non-restating firms in the same industry reflect investors' concerns about accounting quality.  conclude that the bankruptcy announcement of a debtor induces liquidity risk concerns in the market, resulting in negative reactions to creditors. Oh (2013) considers that creditors will reduce loans to other debtors because of the extensive liquidity risk when one borrower encounters a debt crisis, leading to other debtors falling into crisis due to limited credit availability. Bereskin and Cicero (2013) argue that other Delaware intra-industry firms and non-Delaware firms subsequently increased CEO compensation because they were concerned about the risk of losing excellent CEOs. Thus it can be seen that market concern is one of the driving factors of contagion.

Sponsors and their Reputation
Sponsors can act as an agency of information disclosure and certification (Booth and Smith, 1986;Carter and Manaster, 1990): that is, guidance from sponsors can assist IPO firms to disclose high-quality information, and sponsors can guarantee the reliability of firms' public information through their reputation. After China adopted the IPO sponsorship However, the sponsorship system does not seem to have been as effective as expected.
Declines in corporate performance after IPO have frequently occurred, leading to the public questioning the sponsors' guarantee. Empirical evidence shows that the sponsorship system in China has only changed the timing of the adoption of a corporate accounting policy and has not raised information quality as anticipated (Wang and Lian, 2010). The release of the Notice of Conducting Special Checks on 2012 Financial Statements for IPO firms by the CSRC aims to change this situation.
Reputation is an important determinant that influences the effectiveness of information production and the certification agency function by sponsors. Using their industry publicity, expertise, and relations with the monitoring department, sponsors are capable of seeking superior issuing firms whose values are well matched with their reputation (Beatty and Ritter, 1986;Dewenter and Field, 2001). In this situation, sponsors generate a high income as well as avoiding the risk of reputation impairment. Additionally, IPO firms can reduce the issue discount on the first day and may have superior long-term performance by choosing sponsors with high service quality and reputation (Carter and Manaster, 1990;Tinic, 1998;Guo and Zhao, 2006). Therefore, the impairment of sponsors' reputation not only affects their own financial benefits but also creates doubt in the market about their ability and concern about their recommended IPO firms, which further results in negative valuations of these firms. however, the qualifications for listing in those markets are no less restrictive than those in China and sponsors have to be qualified, which is costly, while the spending on IPOs in China would become sunk costs. Thus, investors would have been more likely to consider IPO withdrawal as a negative signal from these firms.

Hypothesis
2. Contagion channel: same sponsors. Besides research on market reaction and signal transmission, we also investigate the contagion effect on the listed firms related to the withdrawing firms. Withdrawing firms have a contagion effect on listed firms because their sponsors are the same institution, which makes the sponsor the contagion channel.
3. Contagion consequence: stronger negative market reaction to listed firms with the same sponsor. We verify the contagion effect by the CAR, consistent with  and  4. Contagion determinants: whether the firms share the same sponsor and its reputation impairment. The negative market reaction to related listed firms is stronger than that to unrelated listed firms, and the negative reaction is further increased as the sponsor's reputation is impaired more severely.

Explanation
: widespread concern about accounting information in the market. IPO withdrawal implies the systematic risks of the sponsor service (Francis and Michas, 2013), such as collusion with clients or incapability. Due to the learning effect, investors are likely to distrust listed firms sharing the same sponsor and to evaluate their information quality negatively. Therefore, we formulate the following hypothesis: H1: Since related firms share the same sponsor with IPO withdrawing firms, the related firms encounter a stronger market reaction than unrelated firms when the sponsor's reputation is questioned, and this contagion effect of IPO withdrawal is strengthened by severe impairment of the sponsor's reputation.

Information quality, unexpected information, and the contagion of IPO withdrawal
According to the literature on market reaction and signal transmission, markets respond positively to high-quality information and good news (Ball and Brown, 1968). In the case of firms that have previously provided high-quality information, these firms are credible and reliable to investors even if their sponsors are involved in IPO withdrawals. Thus these firms may not be affected or may be less affected by contagion as high-quality information could play a governance role. However, the contagion effect may be strengthened if a firm has better fundamental factors.  finds that the failure of a single firm can trigger the failure of another firm, and this effect is strengthened as the relevance of fundamental information increases and weakened as the firm gets close to bankruptcy.  documents that the lower the operational risk of the liquidity crisis firms, the stronger the contagion on the liquidity of debtors sharing the same creditors resulting from the incremental information and psychological gap. Furthermore, a great discrepancy leads to unexpected information resulting in contagion when a well-behaved firm suffers an adverse event. In  design, it is out of the market's expectation that a firm with highly relevant fundamentals goes wrong, leading to stronger negative reaction, while the market reacts indifferently to a firm with lowly relevant fundamentals that becomes distressed, because this is within the market's expectation. Similarly, a firm considered to be less risky in its operation being in trouble indicates intensive unexpected information. Zhao et al. (2013) use unexpected return to investigate contagion during the US subprime mortgage crisis and the European debt crisis between the Shanghai market and 27 other markets over the world; they find that a nonlinear change in unexpected return can forecast crisis.
From the perspective of accounting information quality, high-quality firms withdrawing IPO applications may release negative unexpected information to the market, inducing an intense negative reaction. The market only reacts to unexpected information that lasts for a while (Ball and Brown, 1968), or even overreacts (Bondt and Thaler, 1985); therefore, events that are unexpected may trigger an intense market reaction. Yang et al. (2008) find that the information disclosure violations of special treatment (ST) firms do not induce a strong reaction from the market because of the psychological expectation from investors while the violations of non-ST firms do. Similarly, an IPO withdrawal by a firm with high information quality is not expected by investors, which leads to a more intense negative market reaction, and thus the contagion effect is stronger. Accordingly, we formulate the following alternative hypotheses: H2a: The higher the firm's accounting information quality, the weaker the contagion effect of IPO withdrawal.
H2b: The higher the firm's accounting information quality, the stronger the contagion effect of IPO withdrawal. Barber and Odean (2008) point out that investors must follow firms prior to making buy-in or sell-out decisions, and the level of their attention will affect their decisions. Since the media and analysts are the primary sources of investors' attention, the more heavily a firm is followed by analysts or covered by the press, the stronger the contagion effect of its IPO withdrawal could be. The reasons are as follows:

Investor attention, limited attention, and the contagion of IPO withdrawal
1. The press and analysts can attract limited attention from the public. Attention is a scarce recognition resource , and the press can influence the decisions of the public by drawing their attention to news reports .
Information released by the press can trigger overall market volatility leading to significant changes in corporate value ; thus the earnings announcement effect will be intensified when investors' limited attention is attracted to the announcement (Quan and Wu, 2010).
2. Investor sentiment can be influenced by media or analyst reports. For example, media reporting can lead to large amounts of buy-in transactions by making investors optimistic, and this affects market response . Optimistic investors make stock prices more sensitive to good news, while pessimistic investors make stock prices more sensitive to bad news (Baker and Wurgler, 2006).
3. Media or analyst reports can affect investor recognition. A firm is well known by investors when it is covered more heavily by analysts or the media, and investor recognition of this firm, which affects market reaction to earnings information directly, becomes higher (Barber and Odean, 2008). A firm's earnings information is reflected into its stock price more quickly as short-term attention increases (Fang and Peress, 2009).
Even though an IPO withdrawal indicates the systematic risk of the firm and sponsor, not all investors pay attention to the withdrawal event; for those who have paid attention to the event, they may not understand it; and for those who understand the event, they may not dominate the market sentiment. However, media and analyst coverage can attract limited market attention to the event and the withdrawing firm and can additionally induce market pessimism about accounting information quality. Under these circumstances, media and analyst coverage amplifies the contagion effect of IPO withdrawal by attracting limited attention from the market. Thus, we develop the following hypothesis: H3: The more attention is drawn from the market to the firm, the stronger the contagion effect of IPO withdrawal.

Sample Selection
The sample selection process is as follows: 1. We collect data from the IPO waiting and withdrawal lists up to 3 April 2013 announced by the CSRC. 3. We define firms as "related firms" if they have the same sponsor as the IPO withdrawing firm and as "unrelated firms" otherwise.
4. Excluding observations in the financial industry, with missing main variables, and delivering annual reports on 3 April 2013, we finally obtain 2,161 firm observations.
Because of unavailable data for information quality and market attention indicators, some of these observations are excluded in the regression. Table 2 presents the overall information on Chinese sponsors. We find the following: 1. 934 firms sponsored by 74 institutions applied for IPOs; of these firms, 166 sponsored by 54 (i.e. 73%) institutions withdrew their applications (i.e. the withdrawal rate is 17.78%). Moreover, among the total of 233 sponsoring institutions, 20 were not involved in IPO withdrawals and 159 did not sponsor any IPO applications during the sample period. 4. On the basis of market shares calculated by the number of sponsored listed firms, we find that the number of withdrawing clients is not positively related to the existing market share.

Model Design
Following  and , we use the event study method introduced by Fama et al. (1969) to investigate the contagion effect of IPO withdrawal using CAR. In equation (1), RepuDown, measuring the expected reputation impairment of the sponsor, has a negative correlation β i1.
, α ε (1) We apply quantile regression on equation (1) to test H2a and H2b by the median of Quality and to test H3 by the median of Focus, expecting that the correlation β i1 will be more significant in the high Focus group.

Calculation of CAR
We define the event date as 3 April 2013 and calculate CAR, which indicates the short-term market reaction during the withdrawal announcement, by choosing several short-term event windows. First, we regress equation (2) using stock indicators during the event window period [-155, -6] (Li et al., 2010) to estimate α i and β i : where R it represents actual return for firm i on date t, while R mt represents market return on date t and ε it represents random error.
Second, using estimated α i and β i , we estimate the expected normal return of each stock on every single trading day during the event window period in equation (3). (3) Then, we calculate the abnormal return of stock i on date t during the event window period by equation (4). (4) Finally, we generate the cumulative abnormal return (CAR) of stock i during [t 1 , t 2 ].

Explanatory variable
RepuDown represents the degree of sponsor reputation impairment. Most sponsors are also the underwriters for a stock issuance. The common measures of underwriter reputation impairment are the underwriter's ranking in the IPO tombstone announcement Carter and Manaster, 1990) and the underwriter's market share . Therefore, following Huang (2005) and prior studies on underwriter reputation, we measure sponsor reputation impairment according to whether the sponsor is involved in the IPO withdrawal or not, the number of withdrawing clients, and the withdrawal rate. The indicators are as follows: Sgiveup1c represents the number of IPO withdrawing clients of a sponsor for this event; Sgiveup1d is a dummy variable which equals 1 if a listed firm is sponsored by the same institution as the withdrawing firm and 0 otherwise; Rgiveup1c is calculated as the ratio of the number of withdrawing clients to the total number of firms sponsored by the institution; Rgiveup1d is calculated as the ratio of the number of withdrawing clients of the sponsor to the number of total withdrawing firms; and Rank1c is the sponsor ranking by the number of withdrawing clients, which gets larger as the number of withdrawing clients increases. As the values of the indicators become larger, the possibility of the sponsor's reputation being impaired and the contagion spreading to related firms both increase. According to H1, the negative coefficient of RepuDown indicates the contagion effect of IPO withdrawal.
Quality represents accounting information quality. Earnings management  and audit quality (Francis and Michas, 2013) are usually used to measure information quality, while consistency of analysts' forecasts is also used as a proxy variable. However, as analysts cannot provide assurance on the quality of accounting information, we refer to the first two measurements and design the indicators for information quality as follows: abs_DTAC follows the Jones (1991) model to calculate the absolute value of abnormal accruals, which indicates poor information quality when its value is large. Following Xu and Luo (2007), Change is a dummy variable which equals 1 for performance deterioration after IPO (i.e. net profit of the first quarter in 2013 is lower than that of the first quarter in 2012) and 0 otherwise to indicate good information quality. Bigten is a dummy variable which equals 1 for auditing by the Big Ten accounting firms, following Wang et al. (2009), and 0 otherwise.
Focus is the measure of coverage by the media and analysts. Media represents the quantity of news reports, measuring media exposure (Luo, 2012), for each firm from 1 March to 30 May 2013 5 in the Genius Finance database. Analyst measures analyst coverage . We regress equation (1) separately by high and low Focus and expect a significantly negative coefficient of RepuDown in the high Focus group.
Additionally, following Ball and Brown (1968), we control for other variables that affect our explained variables, including Size, Lev, ROA, Beta, 6 and Industry. 7 Bigten statistics show that 56.6% of our sample firms are audited by the Big Ten accounting firms, in which the fraction of related firms is significantly higher than that of unrelated firms but there is no obvious difference in the level of earnings management. For investor attention, the average analyst coverage is 44 firms, with a minimum of 1 and a maximum of 478, and the coverage of related firms is higher than that of unrelated firms. Meanwhile, the average quantity of news reports for all listed firms is 56 pieces, with a maximum of 1,348 pieces, indicating that the degree of attention paid to sample firms varies greatly, but there is no significant difference in the number of news reports between the compared groups.

Descriptive Statistics
Furthermore, the average Beta is 1.185, the average ROA is 4%, and the average leverage is 5 We do not use longer pre-and post-event window periods due to the timeliness of media reporting of hot issues and because reports over a long period may be irrelevant to the IPO withdrawal. 6 We calculate the coefficients of daily return for single stocks and daily return for the market, backtracking one year starting from 26 March 2013. 7 We adopt 20 dummy industry variables according to the two-digit CSRC industry classification code, in which we further subdivide the manufacturing industry by the four-digit code. 8 This percentage is higher than that in Table 1 because we have deleted some samples with missing  financial data. 42.5%, indicating that for the related firms, size and leverage are significantly lower while risk and return are significantly higher.

CAR Analysis Pre-vs. Post-IPO Withdrawal
In order to observe the short-term market reaction around the IPO withdrawal announcement, we divide the sample firms into two groups, related and unrelated. A run chart for CAR value is generated by calculating the CAR for each group from event day -5 to event day 5, as shown in Figure 1. We can see that the CAR values of both groups increase without obvious difference between the two during the window period [-5, -2].
From day -2, the CAR values go down and the CAR of related firms goes below zero.
During the window period [0, 5], the CAR of the unrelated group gets back to the original level, while the CAR of the related group is still negative and even declines in the window period [2, 5] to a level which is far lower than its prior level.
For a more clear observation on contagion, we present the average abnormal return (AR) on every single trading day and the average CAR of each event window period for the related and unrelated groups separately and also conduct difference tests, as shown in Table   4. The results show that the average AR on the event day is negative and that the average AR of related firms is lower than that of unrelated firms. The average CARs of related and unrelated firms during the period [-5, -2] are both positive, and there is no significant difference between the two groups. From event day -1, the average CAR of unrelated firms is still positive; however, the average CAR of related firms becomes negative and significantly different from the former, which indicates the negative effect of IPO withdrawal on related firms and initially verifies the contagion effect.   (-5,5) 0.002 -0.009 0.010***

IPO Withdrawal Contagion Analysis
On the basis of Table 4 and Figure 1, we use the CAR of event window period [-3, 3] for our main regression. Table 5 presents the IPO withdrawal contagion results, which show that the coefficients of the five proxy variables of sponsor reputation impairment are all negative. The results indicate that the market reaction to related firms is more negative than the market reaction to unrelated firms, suggesting that the contagion effect of related firms is stronger than that of unrelated firms. Moreover, the contagion effect of related firms is strengthened as sponsor reputation impairment increases. Overall, the results provide evidence to support H1.

Information Quality and IPO Withdrawal Contagion
On the one hand, high-quality accounting information may mitigate the contagion effect. When the information quality is good, investors may not change their expectation of the related firms in the case of IPO withdrawal, leading to no or less contagion effect on the high-quality related firms due to investors' trust. On the other hand, IPO withdrawal may be incremental information to the investors of related firms with high-quality information and may strengthen the contagion effect by shaking investors' expectations. Thus, we divide all the firms into subsamples by information quality to test the effect.
First, we divide our sample by level of earnings management into high and low groups.

Constant
-0.057*** -0.024 -0.057*** -0.023 -0.056*** -0.024 -0.057*** -0.024 -0.057*** -0.023 (-4.861) (-1.635) (-4.807) (-1.550) (-4.742) (-1.636) (-4.861) (-1.635) (-4.805) (-1 Finally, we divide our sample firms by whether they are audited by the Big Ten accounting firms or not. In the existing auditing literature, auditing by the Big Four or Big Ten is frequently used as a proxy of auditing quality or accounting quality, the supposition being that the quality of the auditing or accounting provided by larger firms is higher. The results in Table 10 show that the coefficients in the Big 10 group are significantly negative while those in the non-Big 10 group are insignificantly negative. Therefore, firms audited by the Big Ten are affected by the contagion effect, which supports H2.

Investor Attention and IPO Withdrawal Contagion
When a listed firm gets more attention, investors are more likely to become familiar with its sponsor and be able to recognise the related firms in the event of IPO withdrawal.
To test the hypothesis of a stronger contagion effect on firms that get more attention, we measure Focus by analyst coverage and media exposure.
First, we use the median of analyst coverage to divide the sample firms into two groups. Table 11 shows that the coefficients of RepuDown are significantly negative for firms which are followed by more analysts but not significant in the other group. These results are consistent with H3.

Previous market reaction and the contagion effect
First, we apply the test by further distinguishing between positive and negative attention given by investors that affect the IPO withdrawal contagion. The attention investors give to firms may be positive or negative, which may influence market reaction to the firms. Thus, following , we measure the direction of investor attention to a firm by the sign of the CAR value for window period [-35, -6] (i.e. CAR_pre). 9 We define attention to a firm as positive when the sign of CAR_pre is positive and as negative otherwise for the reason that the attitude of media or analyst reports will be reflected in the stock prices. Compared to text analysis, this approach is more objective because we refer to market participants' rather than researchers' understanding of market or analyst reports. Moreover, it reflects the market's overall understanding or evaluation of a firm rather than any single piece of news or report.
In Table 13, we find that IPO withdrawal contagion affects firms with a better market reaction, while those with a worse market reaction are not affected, which is consistent with the influence of accounting information quality on the contagion effect. IPO withdrawal events shock investors who had high expectations of firms with previous good performance and shake their trust and confidence in the firms, leading to the contagion effect.

Other robustness tests
To validate our findings, we further conduct an empirical analysis for H2 and H3 using the CAR value during different window periods and find that the results are consistent with those during window period [-3, 3].
We also conduct a regression for related firms only following  and find that the results are robust, but with a lower significance.

V. Conclusion
The objective of this study is to examine the contagion effect of IPO withdrawal. We find that IPO withdrawals have a contagion effect, which is more pronounced for related firms when sponsor reputation is impaired more severely, when the related firms have higher accounting information quality, and when they receive more attention from the market. Our findings complement the existing research by suggesting a new contagion channel and confirm the validity of the reform of the IPO process conducted by the CSRC to some extent. The results suggest that the financial check reform and the intense supervision by the CSRC not only deter IPO firms and sponsors but also affect the listed related firms, which indicates the continuous certification function of sponsors before and after IPO periods. IPO firms should select sponsors with a high reputation to avoid the contagion effect, thereby promoting professional competence in sponsors.        以及 Jorion and Zhang(2009) 的做法，本文采用 Fama et al.(1969)