Stock price crash and switching of earnings management : empirical evidence from Chinese listed companies

(1)

1

Stock price crash and switching of earnings

management

Empirical evidence from Chinese listed companies

Master Thesis

Student: Yanyu Yang

Student Number: 10654240

Date of final version: 22 June, 2015 Word count:13215

Education: MSc Accountancy and Control, Control track

Institution: University of Amsterdam, Amsterdam Business School Faculty: Economics and Business

(2)

2

Statement of originality

This document is written by Yanyu Yang (Student number: 10654240) who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(4)

4

Abstract

Stock price crash, which refers to extreme price declines that may lead to large and negative stock returns on the particular individual stock, is a significant phenomenon in stock market. Stock price crash can lead invasive financial losses of investors. Stock price crash can be explained by agency problem between managers and investors.

This paper is the first to investigate consequences of stock price crash on switching between real earnings management and accrual-based earnings management. Using sample of Chinese listed companies from 2005 to 2013, this paper finds that stock price crash is associated with prior level of accrual-based earnings management. After stock price crash, level of real earnings manageme nt increase, while level of accrual-based earnings management decline. Firms substitute accrual-based earnings management by real earnings management after stock price crash. In addition, this paper finds that abnormal discretionary expenses such as R&D investments, advertising expenses and administrative expenses are positively associated with prior stock price crash, implies that most firms are engaging in managing earnings through these discretionary expenses. However, in terms of magnitude, this paper finds that increase in the level of real earnings management is not very significant, so we do not have to worry a lot. Compared with increase in real earnings management, magnitude in decrease of accrual-based earnings management is larger, which should gain more attention.

This paper can make several contributions. First of all, it makes contribution to fill the void in stock price crash risk literature. Prior research on stock price crash focuses on explaining of the occurrence stock price crash and mechanism to release stock price crash risk. This paper is the first to investigate impact of stock price crash. Secondly, this paper makes contribution to switching between real earnings management and accrual-based earnings management by adding new factor that would lead firms to tradeoff between these two methods. Thirdly, this paper provides several important implications to investors. Investors need to increase their ability to

(5)

5

identify stock price crash risk in order to avoid significant financial losses. Moreover, with the results of relationship between earnings management and stock price crash, this paper promotes investors to be wary about accrual-based earnings management which may lead to subsequent stock price crash and pay attention to real earnings management activities after stock price crash which may harm normal operation activities and value of the firm. Finally, this paper also provides implications for regulators. Stock price crash is not good for maintaining stability of stock market and may destroy investors’ confidence. Regulators need to pay attention to stock price crash and associated earnings management activities. Accrual-based earnings management may undermine integrity of accounting information and make accounting information become less efficient. So regulators need make relevant policy to maintain quality and efficiency of accounting information.

(6)

6

1.0 Introduction

Stock price crash, which refers to extreme price declines that may lead to large and negative stock returns on the particular individual stock, is a significant phenomenon in stock market (K im et al., 2011b; Hamm et al., 2014). Recent years have seen heightened concern on stock price crash as it happened now and then in global stock market and could have invasive impact on investors. A series of financial scandals such as Enron and WorldCom, for example, are related with stock price crashes. Moreover, investors may suffer tremendous loss from stock price crash. For instance, the suddenly and extremely decline on the price of the share Guizhou Maotai which owns the largest market value in the stock market of China led to collapse of over 50 trillion. Hence, it is an important issue which has attracted considerable attention of both academic and practice.

This paper intends to investigate the relationship between stock price crash and switching of earnings management. More specific, this paper intends to study whether stock price crash would have impact on earnings management methods conducted by the company, that is, the consequences of stock price crash on switching between real earnings management and accrual-based earnings management. Particularly, research question of this paper is whether companies would trade off accrual based earnings management with real earnings management after stock price crash. This paper is the first to focus on studying consequences of stock price crash on changes in earnings management behavior.

This research has three major motivations. First of all, this research is motivated by the interests of the topic of stock price crash risk and switching between real earnings management and accrual-based earnings management. Stock price crash is an interesting phenomenon as well as a significant phenomenon in the stock market. It is an extreme outcome of the returns of stocks. Returns on the stock of firms experienced stock price crash present fat tail characteristic. Moreover, as has mentioned, stock price crash is an important phenomenon in stock market and could lead to invasive results for investors (K im et al., 2011b). There are significant risks

(7)

7

associated with stock price crash. So it is important for investors to identify this kind of risk. Tradeoff between real earnings management and accrual-based earnings management is also an interesting behavior conducted by companies. Prior studies find evidence that companies substitute between real earnings management and accrual-based earnings management (Zang, 2012). This switching behavior is identified interesting for me. Earnings management activities are also important for companies. Both accrual-based earnings management and real earnings management could have impact on shareholder’s value. Accrual-based earnings management may obscure true performance of the company and may affect financing of the company. According to Graham et al. (2005) and Cohen et al. (2008), real earnings management is more costly to companies and would have real impact on company’s value as it can affect business and operation of the company. Hence, research on the relationship between stock price crash and tradeoff between earnings management method is not only interesting but also can provide insights for investors to be more scrutiny about corporate behavior.

Second, this research is also motivated by the growing significance of efficiency and integrity of financial information. Both stock price crash and earnings management are closely related to efficiency of accounting informa tion and allocation of resources. They are important topics either in academic research or in practice. Integrity of financial reporting plays crucial role in financial market as they may have impact on decision making of relevant stakeholders, which has been emphasized by SEC of the US (Graham et al., 2005). Stock price crash indicates inefficiency of accounting information disclosure and misallocation of resources. Accrual-based earnings management may not provide true picture of a company’s performance and thus may change decision making by investors. It then may make information disclosure become inefficient. Stock price crash and accrual-based earnings management could have pervasive impact on investors in practice. Therefore, it is important to conduct research in stock price crash and accrual-based earnings management. This paper intends to provide more study in these important research fields.

(8)

8

Third, this research is partly motivated by prior literature. Prior literature focuses on studying reasons for the rise of stock price crash and mechanism to reduce stock price crash risk. Prior studies have found that stock price crash can be related to information asymmetry. It has been documented that stock price crash can arise due to managers hoard bad news and accelerate disclosure of good news (K im et al., 2011b). Moreover, Hutton et al. (2009) find that earnings management provides opportunities for managers to hoard bad news, which increases the likelihood of stock price crash. But it is not clear about whether stock price crash would have impact on the tradeoff between earnings management methods. This paper intends to extend further on previous research. Whether investors and companies will become more scrutiny after stock price crash? Whether companies will trade off accrual based earnings management with real earnings manageme nt after stock price crash? I expect that after stock price crash, managers are more likely to engage in real earnings management, which is harder to detect, in order to regain shareholders’ confidence, if agency problems still exist. So this paper will make contribution to fill the void in literature.

Stock price crash can be observed either in the whole market or in firm level. This paper intends to research stock price crash at firm level. At firm level, stock price crash can be explained by agency theory. Starting from Jin and Myers (2006), researchers have found the stock price crash can have link to information asymmetry between shareholders and managers who are coupled with self- interest. According to Kim et al. (2014), when facing with bad performance, managers have the incentive to overstate financial performance due to the existence of agency problems and this can be achieved by strategically withholding bad news and accelerating disclosure of good news. When doing this, managers hope poor current performance can be camouflaged by pleasant performance in the future. Such factors as formal compensation contracts and career concerns may serve as motivations for such asymmetric disclosure between bad news and good news (Watts, 2008; Ball, 2009). If managers have the ability to withhold bad news by using methods like earnings management, bad news could be stockpiled in a certain period. With the operation of

(9)

9

the company, however, these pieces of bad news will get accumulated within the company and there is a threshold for the company to absorb and hide these pieces of negative news. O nce the company cannot hide these pieces of bad news or the costs to hold them are very high, these pieces of bad news will relea se in the market all of a sudden. Then stock price crash would happen.

I first examine whether stock price crash is associated with prior level of accrual-based earnings management. I find that stock price crash is positively associated with prior level of accrual-based earnings management, while I find little evidence to support that stock price crash is associated with prior level of real earnings management. The results are consistent with prior literature which finds that stock price crash is associated to corporate financial information opacity and accrual-based earnings management could lead financial reporting of companies become less transparent (Hutton et al., 2009).

I then examine whether stock price crash could have impact on earnings management behavior through two separate regression models. I find that the level of real earnings management increase after stock price crash, while accrual-based earnings management decline after stock price crash. Besides, the level of real earnings management and the level of accrual-based earnings management is negatively associated, implying companies switch between these two earnings management methods. Additionally, of three general real earnings management activities including managing earnings through cash flow from operating, production cost and discretionary expenses, I find discretionary expenses such as R&D investments, advertising expenses and administrative expenses are positively associated with prior stock price crash. Stock price crash may lead to increase of discretionary expenses. The results are in consistent with prior survey conducted by Graham et al. (2005) who find 80 percent of surveyed CFOs are likely to decrease research and development (R&D) activities, advertising expenses, and maintenance expenditures when they are engaging in earnings management activities to increase earnings. However, in terms of magnitude, magnitude of the increase in the level of real earnings management is not very significant, so we do not have to worry about it

(10)

10

a lot. By contrast, we should pay more attention to the decline of accrual-based earnings management, which has larger magnitude.

This paper makes contribution from several perspectives. First of all, this paper contributes to both stock price crash risk literature and earnings management literature. I contribute to stock price crash literature by examining the relationship between stock price crash and impact on switching of earnings management behavior. Although stock price crash is an important and interesting phenomenon in stock market, research on this topic is just handful. Moreover, prior literature related to stock price crash focus on investigating factors that lead to stock price crash and mechanisms that could relieve stock price crash risks. Rare literature studies consequences of stock price crash. Not even to say research on the impact on switching between real earnings management and accrual-based earnings management. This paper is the first to study on the consequence of stock price crash on switching of earnings management methods. Prior literature has documented that stock price crash has relationship with corporate financial information opacity. Particularly, accrual-based earnings management may lead to stock price crash. I extend the research by examining consequences of stock price crash on earnings management behavior. Apart from stock price crash literature, I also make contribution to earnings management literature. Prior literature has documented that companies may switch between real earnings management and accrual-based earnings management due to SOX, audit quality, seasoned stock issuing (Cohen et al., 2010; Chai et al., 2011; Cohen and Zarowin, 2010), I add new factor that would lead to switch between real earnings management and accrual-based earnings management. Therefore, I make contribution to fill the void in both stock price crash literature and earnings management literature.

Secondly, this paper provides some insights for investors. As has mentioned, stock price crash is an important phenomenon in stock market and could lead to invasive results for investors (K im et al., 2011b). So it is important for investors to identify this kind of risk. Stock price crash implies there is bubble in the value of the firm and stock of the firm has been overvalued, which should cause scrutiny of

(11)

11

investors. Moreover, this paper finds that stock price crash is associated with prior accrual-based earnings management behavior. So investors should pay attention to earnings management behavior of the company and increase their ability to identify accrual-based earnings management behavior conducted by companies. In addition, this paper also finds that after stock price firms are more likely to use real earnings management, which may harm normal operation activities and value of the company. Although there is no significant relationship between prior level of real earnings management and stock price crash, real earnings management may harm value of firm and thus investors’ wealth in the long term. So this paper promotes investors to keep an eye on real earnings management activities after stock price crash. In sum, investors should increase their ability to identify stock price crash risk and pay attention to associated earnings management activities.

Finally, this paper also provides several important implications for regulators as well as accounting policy makers. As has mentioned, stock price crash implies there is inflation in prior stock price and there is misallocation of resources. Moreover, stock price crash is not helpful to maintain stability of the stock market and stock price crash may destroy investors’ confidence. So regulators need to take action to maintain stability of stock market and regain investors’ confidence. Regarding to the relationship between earnings management activities and stock price crash, this paper finds that stock price crash is associated with prior accrual-based earnings management and stock price crash may lead to substitute accrual-based earnings management by real earnings management. This paper promotes regulators to understand earnings management behavior conducted by companies and relevant risks of stock price crash. For accounting policy makers, this paper could also provide implications. Accrual-based earnings management may not provide useful and efficient information for investors as financial reporting under earnings management may not reflect true performance and financial status of the company. Moreover, it undermines integrity of accounting information. Regulators could make possible policies to eliminate stock price crash risks and associate earnings management activities.

(12)

12

The remainder of this paper proceeds as follows. Section 2 provides a review of relevant literature and discuss of hypothesis development for the study. Section 3 covers research method and design for hypothesis tests, followed by a discussing of major findings in section 4. Finally, in section 5, there are discussions of these major findings and conclusions.

2. Literature review and hypotheses

2.1 Agency theory and information asymmetry disclosure

2.1.1 Agency theory

Due to separation of ownership and control, the operation of a company can be seen through an explicit or implicit contract between principals (owners) and agent (managers). Under this contract relationship, according to Jensen and Meckling (1976), the principals engage the agent to provide some services on their behalf and authorize some decision making rights to the agent in daily operation. Under the assumption that both parties are utility maximization, managers are self- interested and they do not always behave in the best interest of owners. Managers have the incentives to take behaviors to maximize their own utilities which may not be optimal for shareholders value (Watts and Zimmerman, 1986). Possible self- interest behaviors may include appropriating firm’s resources, investing in perks, consuming perks and manipulating financial performance (Day, 2008).

2.1.2 Information asymmetry and information disclosure

As managers take responsibility of daily operation of the company, they are more likely to gain private information and specific knowledge about the company (Macintosh and Quattrone, 2010). Owners, on the other side, may not gain such kind of specific information due to they do not engage daily operation of the company directly. Under this circumstance, managers can gain access to some important information and these pieces of information deem to be private information, but

(13)

13

owners have no access. As a result, there is information asymmetry between managers and shareholders. Based on agency theory, managers have incentives to opportunistically exploit this information asymmetry to serve for interests when putting effort. Particularly, in terms of information disclosure, managers have the tendency to hoard private information, accelerate disclosure of good news and delay disclosure of bad news, motivated by such factors such as career concerns, compensation and reputation (Ball, 2009; Khan and Watts, 2009). Empirical studies show that managers have the tendency to delay disclosure on bad news driven by asymmetric payoffs between good news and bad news (Kothari et al., 2009).

2.2 Earnings manage ment and tradeoff between accrual-based earnings management and real earnings management

2.2.1 Earnings management

As has been mentioned, driven by self- interest, managers tend to behave in their own interests rather than on the interests of shareholders. They may get engaged in such behaviors as consuming perks and invest in improper projects. According to Merchant and Van der Stede (2011), managers often engage in gamesmanship like earnings management to make their performance seem good and to hide their improper actions. Accrual-based earnings management and real earnings management are two commonly used earnings management methods (Cohen et al., 2007). Accrual-based earnings management refers to that within the allowance of accounting standards, managers choose different accounting estimates to treat with accrual-based items to affect their reported performance, aiming to achieve several goals like boosting earnings, earnings smoothing and reducing earnings (Dechow and Skinner, 2000). Generally, accrual-based earnings management has no impact on the value of the company. By contrast, the other method, real earnings management which manages earnings through operating activities such as through sales and operating expense activities, could have real impact on the company (Graham et al., 2005). Actually, real earnings management is more costly to the company as it can affect business and operation of the company, so it can be harmful to the company (Graham

(14)

14

et al., 2005). Moreover, real earnings management is harder to detect by investors and auditors (Cohen et al., 2007).

Previous studies have documented that several factors may serve as motivations for companies to engage in earnings management. One possible driver for companies to engage in earnings management is the desire to maintain proper valuation so as to lower financial cost either debt financing or equity financing (Dechow et al., 1996). Furthermore, respond to economic rewards of high stock price is a significant incentive for managers to manage reported earnings (Bergstresser and Philippon, 2006; Bergstresser et al., 2006). Cornett et al. (2008) find that with significant economic rewards, incentive-based compensation could encourage earnings management. According to Cohen et al. (2007), managers with equity-based compensation are more sensitive to short-term stock price, and thereby they are more likely to engage in earnings management. Besides, to avoid small negative reported earnings is also an important incentive as researchers have found that small negative reported earnings can lead to large stock price declines (K inney et al., 2002). In addition, beating analysts’ forecast can also be a motivation for earnings management (Cheng and Warfield, 2005).

2.2.3 Tradeoff between accrual-based earnings management and real earnings management

The major difference between accrual-based earnings management and real earnings management is that real earnings management is more costly to the firm and harder to detect. Prior literature find evidence that firms may use different earnings management methods under certain circumstance and they may use them as substitutes (Cohen and Zarowin, 2010; Zang, 2012). Zang (2012) find that managers’ tradeoff between these two methods is mainly decided by the consideration of relative cost of these methods. Moreover, he notes that the extent of accrual-based earnings management would be adjusted based on the level of real activities manipulation realized. Cohen et al. (2007) find that after the passage of SO X, accrual-based earnings management declined significantly, while the extent of real earnings

(15)

15

management increased largely, as firms are more likely to become wary in response to the security of regulators and investors after SOX. Cohen and Zarowin (2010) show that the level of real earnings management declined after company had seasoned equity offerings. In addition, previous studies also show enhanced audit quality (Chi et al., 2011) and external monitor (Wongsunwai, 2013) could affect companies’ switch between real earnings management and accrual-based earnings management.

2.3 Stock price crash and information asymmetry

Starting from Jin and Myers (2006), researchers have linked information asymmetry between managers and owners with stock price crash. Driven by self- interest, managers have incentives to accelerate disclosure of good news and delay disclosure of bad news, aiming to make their reported performance look pleasant and hoping current poor performance could be camouflaged by pleasant performance in the future (K im et al., 2014). Previous empirical study has provided evidence that managers have the tendency to delay disclosure of bad news to investors largely due to asymmetric payoffs between good news and bad news (Kothari et al., 2009). When the accumulation of bad news excess a threshold, however, it is impossible for the company to hide bad news any more either because it is costly to absorb these pieces of bad news or managers are not able to hoard bad news any more (Kim et al., 2014). As a result, the large amount of bad news will altogether be released in the market all of a sudden, which leads to stock price crash (Hutton et al. 2009). Therefore, the asymmetric disclosure of information by managers can induce stock price crash (Kim et al., 2014; Hamm et al., 2014).

Based on this idea, recent studies have documented several factors that are related to hoarding bad news behavior of managers and asymmetric disclosure behavior of managers, which then are associated to stock price crash. K im et al. (2011a) find that equity-based compensation could induce hoarding bad news behavior of managers. Moreover, Iannotta and Pennacchi (2011) find stocks with more analysts’ coverage have greater tendency to experience stock price crash and they also provide evidence to support that stock price crash is mainly caused by

(16)

16

release of former hidden bad news. Furthermore, Callen and Fang (2011) find that lack of auditor monitoring could also encourages hoarding bad news. In addition, corporate tax avoidance and career concerns may also serve as incentives for managers to engage in withholding bad news behavior (K im et al., 2011b; Hermalin and Weisbach, 2012).

2.4 Stock price crash and earnings management

As has mentioned, the happen of stock price crash can be explained hoarding bad news mechanisms and asymmetrical disclosure behavior of managers. Furthermore, previous studies have shown hoarding bad news has largely been confined to transparency of financial information reporting system (Hamm et al., 2014). Hutton et al. (2009) find that poor earing quality in financial report could provide opportunities for managers to hoard bad news, which then increase the likelihood of stock price crash. They use discretionary accruals as measurement of earnings management and they find evidence to support information management is one cause of stock price. More specific, companies would like to use earnings management method to shelter bad news. Once piles of bad news are gathered up to a certain point, and then they are likely to be released at once, leading to stock price crash. Hence, information transparency could reduce the likelihood of stock price crash in the future (Jin and Myers, 2006; Hutton et al., 2009; Hamm et al., 2014). Moreover, K im et al. (2014) indicates accounting conservatism is another way to contribute to accounting information quality and reduce stock price crash risk.

2.5 Hypothesis development

Due to agency problems and information symmetry, managers have incentives to accelerate disclosure of good news and delay the disclosure of bad news. Moreover, previous studies have provided substantive evidence to support that stock price crash can be caused by suddenly release of former large amount of hidden bad news (Hutton et al., 2009; Iannotta and Pennacchi, 2011; K im et al., 2011a). According to

(17)

17

Hutton et al. (2009), earnings management makes financial information reporting of company become less opaque and then increases the probability that former large amount of hidden bad news would be released in the stock market at once, so earnings management would increase stock price crash risk of the company. Based on agency theory and previous studies about information quality and stock price crash, the first hypothesis is developed as follows, which is a confirmation on the findings of previous studies:

H1: Ceteris paribus, companies with higher level of accrual-based earnings management are more likely to experience stock price crash.

Cohen and Zarowin (2010) and Zang (2012) indicate that managers may use accrual-based earnings management and real earnings management as substitutes. After stock price crash, confidence of investors on the stock may be undermined and they may become more scrutiny and wary on particular stock. They may notice that managers may hoard bad news in the past and they would pay special attention to accrual-based earnings management activities. Under such circumstance, the cost of accrual-based earnings management would increase. On the other side, there are reversals of abnormal accruals under accrual-based earnings management, while there are no such reversals under real earnings management. Considering relative costs of real earnings management and accrual-based earnings management and possible impact of these two methods, firms may tend to use real earnings management when they want to hide bad news. Under the assumption that agency problem still exists, managers are more likely to engage in real earnings management when they want to hide bad news. Therefore, there is a tradeoff between these two earnings management after stock price crash. This leads to the second hypothesis:

H2a: Ceteris paribus, after stock price crash, the level of real earnings management would increase.

H2b: Ceteris paribus, after stock price crash, the level of accrual-based earnings management would decrease.

(18)

18

3. Research method

3.1 Sample and data collection

This paper chooses Chinese listed companies as research sample. It will collect data from CSMAR (China Securities Market & Accounting Research) database, which is one of the largest databases for Chines listed company in China. Initial research time period is designed as from 2005 to 2013. Currently, there are about 2000 companies listed in Chinese market according to CSMAR database, so it will generate about enough sample to conduct the quantitative research. After dropping financial industry data and missing data to run Modified Jones Model and NCSKEW, there are around 12413 samples, which is sufficient for the study. There are several reasons to choose Chinese listed companies as research sample. The most important reason is to exclude the impact of SOX on earnings management and stock price crash. Hutton et al. (2009) find the relationship between earnings management and stock price crash dissipated after the passage of SOX, which may due either to agency problems got relieved or to managers were less able to hide bad news in new regulatory environment. Hamm et al. (2014) also find this weaker association. Moreover, Cohen et al. (2007) provide empirical evidence that firms switched accrual-based earnings management with real earnings management after SOX. Hence, using sample from the US companies may not be a proper choice to test hypotheses of this paper. China released similar regulation comparable to SOX called the Basic Norms for Enterprise Internal Control in 2008 (Leng and Zhao, 2013). However, due to different regulation environment and enforcement efficiency, the impact of Chines SOX the Basic Norms for Enterprise Internal Control on listed companies is not as high as that in the US. Accordingly, by using Chinese sample the impact of SOX and similar regulation on my research may be reduced. The second reason is that Chinese stock market is a large market, so there are a lot of data available. All data needed to conduct this study are available in selected database. Third, Chinese stock market is similar with the US stock market in terms of

(19)

19

accrual-based earnings management and real earnings management as prior literature have documented similar behavior in Chinese market (Szczesny et al., 2008; Chen and Yuan, 2004). Therefore, it is reasonable to choose Chinese companies as sample to conduct the study.

It is worth to mention that there was significant structural reform in Chinese stock market in 2005 which is called spilt share structure reform. Year 2005 is an important cutoff in Chinese stock market. Due to historical reason, Chinese stock market has the characteristic of spilt share structure which means that stocks of a company are split into tradable and non-tradable shares. Tradable shares can be traded in the stock market, while non-tradable shares cannot be traded in the stock market. Almost two-thirds of stocks of a company are non-tradable shares and most of non-tradable shares are state owned. By contrast, tradable shares, less than one-third of total shares, are public holding. Under this situation, shareholders of tradable and non-tradable shares have different rights and listed companies and large shareholders may not care about stock price, which may affect test of hypothesis of this paper. After 2005 non-tradable shares reform, all shares were tradable and the difference between tradable shares and non-tradable shares was eliminated (Firth et al., 2010). Moreover, state reduced its holdings in listed companies and almost all traded shares are public holding. Consequently, corporate governance quality improved (Jingu, 2007). It seems that structure of Chinese stock market was more comparable to US market after 2005. Based on these considerations, I select the sample of Chinese listed companies from year 2005.

3.2 Measures of variables

3.2.1 Stock price crash

As in K im et al. (2014), this paper uses the likelihood of extreme negative firm-specific weekly returns (NCSKEW) to measure stock price crash. More specific, crash, or extreme negative week, regards to the week in which firm-specific returns 3.2 standard deviations is below the mean firm-specific weekly returns of the whole fiscal year. If there are no substantial abnormal returns, the distribution of returns

(20)

20

would be normal distribution. Under stock price crash, however, the distribution of returns would present fat tail characteristic. The chosen 3.2 is consistent with the frequency of 0.1% in the normal distribution. NCSKEW is obtained as follows:

First of all, using firm-specific weekly returns of a given stock in a given year to run a market model regression to obtain R-Square and residual returns:

𝑟_𝑖,𝑡 = 𝛼_𝑖 + 𝛽_1𝑖𝑟_{𝑚 ,𝑡−2}+ 𝛽_2𝑖𝑟_{𝑚 ,𝑡−1}+ 𝛽_3𝑖𝑟_{𝑚 ,𝑡} + 𝛽_4𝑖𝑟_𝑚,𝑡+1+ 𝛽_5𝑖𝑟_𝑚,𝑡+2+ 𝜀_𝑖,𝑡 (1) Where ri,t refers to the return of stock i in week t and rm,t refers to the return of

stock I in week t after considering CRSP value-weighted market index. In consistent with Dimson (1979), this paper adds the lead and lag terms in equation (1) so as to adjust for the impact of nonsynchronous trading. Firm-specific weekly return of stock i in week t is:

𝑊_𝑖,𝑡 = ln⁡（1 + 𝜀_𝑖,𝑡） (2)

Where 𝜀_𝑖,𝑡 is residual from the regression (1).

Based on Wi,t, the measure of stock price crash, that is, the negative conditional

return skewness (NCSKEW) of firm j in year t is calculated as follows: 𝑁𝐶𝑆𝐾𝐸𝑊_𝑖,𝑡 = −[𝑛(𝑛 − 1)32∑ 𝑊_𝑖,𝑡3]/[(𝑛 − 1)(𝑛 − 2)(∑ 𝑊_𝑖,𝑡2)

3 2] (3)

Where n is the number of trading weeks in the fiscal year t. The greater the NCSKEW, the larger stock price crash risk.

3.2.3 Accrual-based earnings management

This paper uses discretional accruals to measure the level of accrual-based earnings management. To obtain discretional accruals, it applies modified Jones Model (Jones, 1991). First of all, according to equation (4), run a regression to estimate coefficient⁡𝛼̂₁,⁡𝛼̂₂,⁡𝛼̂₃, using each two-digit SIC code of China's Securities Regulatory Commission (CSRC). 𝑇𝐴𝑖,𝑡 𝐴𝑠𝑠𝑒𝑡𝑖,𝑡−1 = 𝛼1 𝐴𝑠𝑠𝑒𝑡𝑖,𝑡−1+ 𝛼2 Δ𝑅𝐸𝑉𝑖 ,𝑡 𝐴𝑠𝑠𝑒𝑡𝑖,𝑡−1+ 𝛼3 𝑃𝑃𝐸𝑖 ,𝑡 𝐴𝑠𝑠𝑒𝑡𝑖,𝑡−1+ 𝜀𝑖,𝑡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡(4) In equation (4):

𝑇𝐴_𝑖,𝑡 = 𝐸𝐵𝐸𝐼_𝑖,𝑡 − 𝐶𝐹𝑂_𝑖,𝑡, which refers to total accruals of firm i in year t. Where EBEI is the earnings before extraordinary items and discontinued operatio ns and CFO

(21)

21

is cash flow from operations.

𝐴𝑠𝑠𝑒𝑡_{𝑖,𝑡−1⁡⁡}is total assets of firm i in year t-1.

Δ𝑅𝐸𝑉_𝑖,𝑡 is changes in revenue from year t-1 to year t.

𝑃𝑃𝐸_𝑖,𝑡⁡is gross value of property, plant and equipment of firm i in year t.

Then using the coefficient estimates from equation (4) to estimate firm-specific normal accruals in equation (5). Discretionary accrual is defined as the differences between total accruals and adjusted normal accruals, which is calculated according to equation (6): 𝑇𝐴𝑖,𝑡 𝐴𝑠𝑠𝑒𝑡_{𝑖,𝑡−1} = 𝛼1 𝐴𝑠𝑠𝑒𝑡_{𝑖,𝑡−1}+ 𝛼2 Δ𝑅𝐸𝑉𝑖 ,𝑡 𝐴𝑠𝑠𝑒𝑡_{𝑖,𝑡−1}+ 𝛼3 𝑃𝑃𝐸𝑖 ,𝑡 𝐴𝑠𝑠𝑒𝑡_{𝑖,𝑡−1}+ 𝜀𝑖,𝑡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡(5) Disacc_𝑖,𝑡 = 𝑇𝐴𝑖,𝑡 𝐴𝑠𝑠𝑒𝑡_{𝑖,𝑡−1}− ( 𝛼 ̂₁ 𝐴𝑠𝑠𝑒𝑡_{𝑖 ,𝑡−1}+ 𝛼̂2 Δ𝑅𝐸𝑉_{𝑖 ,𝑡}−Δ𝑅𝐸𝐶_{𝑖 ,𝑡} 𝐴𝑠𝑠𝑒𝑡_{𝑖,𝑡−1} + 𝛼̂3 𝑃𝑃𝐸_𝑖,𝑡 𝐴𝑠𝑠𝑒𝑡_{𝑖,𝑡−1}) (6)

Where Δ𝑅𝐸𝐶𝑖,𝑡 is the change in accounts receivables from year t-1 to year t.

Then I can obtain⁡DA_𝑖,𝑡, that is, discretionary accruals of firm i in year t. Here, I use the absolute value because our hypotheses do not predict any specific direction for earnings management. The greater the Disacc, the larger accrual-based earnings management.

3.2.4 Real earnings management

Following Roychowdhury (2006) and Cohen et al. (2007), this paper uses abnormal cash flow from operations (CFO), discretionary expenses and production costs to measure the level of real earnings management. According to Roychowdhury (2006), companies may engage in real operating activities to manipulate earnings and these activities may include reduce costs of goods sold by overproducing inventory, cutting discretionary expenditures such as R&D investments expenses, advertising expenses and selling, general, and administrative (SG&A) expenses. Measures of manipulation in these real operating activities could indicate level of real earnings management. Other studies such as Cohen et al. (2010) offer evidence to support that these measures could capture the level of real earnings management.

First of all, using the model developed by Dechow et al. (1998) to generate normal levels of CFO, discretionary expenses and production expenses. The normal

(22)

22

CFO is expressed as a linear function of sales and changes in sales as applied in Cohen et al. (2007). Run a cross-functional regression based on industry and year to estimate CFO, which is shown below:

𝐶𝐹𝑂_𝑖,𝑡 𝐴𝑠𝑠𝑒𝑡𝑠_{𝑖,𝑡−1}= 𝑘₁ 𝐴𝑠𝑠𝑒𝑡𝑠_{𝑖,𝑡−1}+ 𝑘2 𝑆𝑎𝑙𝑒𝑠_𝑖,𝑡 𝐴𝑠𝑠𝑒𝑡𝑠_{𝑖,𝑡−1}+ 𝑘3 ∆𝑆𝑎𝑙𝑒𝑠_𝑖,𝑡 𝐴𝑠𝑠𝑒𝑡𝑠_{𝑖,𝑡−1}+ 𝜀𝑖,𝑡⁡⁡⁡⁡⁡(7)

Abnormal level of CFO is calculated as the difference between actual CFO and normal CFO calculated using estimated coefficient in equation (7).

As in Cohen et al. (2007), defining production cost as the sum of cost of goods sold and change in inventory levels in a given fiscal year. Production cost is modeled as a linear function of sales, described in equation (8):

𝐶𝑂𝐺𝑆_𝑖,𝑡 𝐴𝑠𝑠𝑒𝑡𝑠_{𝑖,𝑡−1} = 𝑘_1,𝑡 𝐴𝑠𝑠𝑒𝑡𝑠_{𝑖,𝑡−1}+ 𝑘2 𝑆𝑎𝑙𝑒𝑠_𝑖,𝑡 𝐴𝑠𝑠𝑒𝑡𝑠_{𝑖,𝑡−1}+ 𝜀𝑖,𝑡⁡⁡⁡⁡⁡(8)

And then, model growth in inventory as follows: ∆𝐼𝑁𝑉_𝑖,𝑡 𝐴𝑠𝑠𝑒𝑡𝑠_{𝑖,𝑡−1} = 𝑘_1,𝑡 𝐴𝑠𝑠𝑒𝑡𝑠_{𝑖,𝑡−1}+ 𝑘2 ∆𝑆𝑎𝑙𝑒𝑠_𝑖,𝑡 𝐴𝑠𝑠𝑒𝑡𝑠_{𝑖,𝑡−1}+ 𝑘3 ∆𝑆𝑎𝑙𝑒𝑠_{𝑖,𝑡−1} 𝐴𝑠𝑠𝑒𝑡𝑠_{𝑖,𝑡−1}+ 𝜀𝑖,𝑡⁡⁡⁡⁡⁡(9)

The normal level of production cost is estimated as follows by using equation (8) and (9). 𝑃𝑟𝑜𝑖,𝑡 𝐴𝑠𝑠𝑒𝑡𝑠_{𝑖,𝑡−1}= 𝑘1,𝑡 𝐴𝑠𝑠𝑒𝑡𝑠_{𝑖,𝑡−1}+ 𝑘2 𝑆𝑎𝑙𝑒𝑠𝑖,𝑡 𝐴𝑠𝑠𝑒𝑡𝑠_{𝑖,𝑡−1}+ 𝑘3 ∆𝑆𝑎𝑙𝑒𝑠𝑖,𝑡 𝐴𝑠𝑠𝑒𝑡𝑠_{𝑖,𝑡−1}+ 𝑘4 ∆𝑆𝑎𝑙𝑒𝑠𝑖,𝑡−1 𝐴𝑠𝑠𝑒𝑡𝑠_{𝑖,𝑡−1}+ 𝜀𝑖,𝑡⁡⁡⁡⁡⁡(10)

As for the normal level of discretionary expenses, they are modeled as below: 𝐷𝐸_𝑖,𝑡 𝐴𝑠𝑠𝑒𝑡𝑠_{𝑖,𝑡−1} = 𝑘_1,𝑡 𝐴𝑠𝑠𝑒𝑡𝑠_{𝑖,𝑡−1}+𝑘2 𝑆𝑎𝑙𝑒𝑠_{𝑖,𝑡−1} 𝐴𝑠𝑠𝑒𝑡𝑠_{𝑖,𝑡−1}+𝜀𝑖,𝑡⁡⁡⁡⁡⁡(11)

It is worth to mention that model discretionary expenses as a function of lagged sales in order to tackle with the mechanical problem that if companies manage sales upward to increase reporting earnings in a given year would lead to significant low residuals from running a regression using sales in year t in equation (11).

In above equations:

𝐶𝐹𝑂_𝑖,𝑡 represents cash flow from operations of firm i in period t.

𝑃𝑟𝑜𝑖,𝑡 represents production costs of firm i in period t, which is defined as the

sum of cost of goods sold and the changes in inventory level.

𝐷𝐸_𝑖,𝑡 represents discretionary expenses of firm i in period t, which is defined as the sum of advertising, R&D and SG& A expenses.

(23)

23

costs (ab_Prod) and discretionary expenses (ab_DisEx) are calculated as the differences between actual level and estimated normal level generated from equation (7), (10) and (11). Use RM_PROXY to indicate sum of abnormal CFO, production cost and discretionary expenses.

3.2.5 Control variables

As in former research (Cohen et al., 2007; K im et al., 2014), this paper uses following variables as control variables: firm size (SIZE), financial leverage (LEV), operating performance (ROA), Revenue (REV).

3.3 Empirical design

3.3.1 Design for Hypothesis 1

I examine the relationship of accrual-based earnings management and stock price crash developed in hypothesis 1 through following regression:

NCSKEW_𝑡 = 𝛼₀+ 𝛼₁× 𝐷𝑖𝑠𝑎𝑐𝑐_𝑡−1+ 𝛼₂× 𝑅𝑀_𝑃𝑅𝑂𝑋𝑌_𝑡−1+⁡ 𝛼₃× 𝑆𝑖𝑧𝑒_𝑡 + 𝛼₄× 𝑅𝑂𝐴_𝑡+ 𝛼₅× 𝐿𝑒𝑣_𝑡 + 𝛼₆× 𝑅𝐸𝑉_𝑡 + 𝜀

Hypothesis 1 examines whether accrual-based earnings management would lead to stock price crash, which is to test former results of prior literature. To test this hypothesis, I use NCSKEW_𝑡as dependent variable and use last period discretionary accruals Disacct-1 as explanatory variable. Moreover, since real earnings management

and accrual-based earnings management coexist, I control the impact of real earnings management by putting variable of RM_PROXY in the model and running regression simultaneously. To support Hypothesis 1, NCSKEW_𝑡 is predicted to be significantly associated with 𝐷𝑖𝑠𝑎𝑐𝑐_𝑡−1. That is, 𝛼₁ is predicted to be positive and should be at significant level in statistics.

3.3.2 Design for Hypothesis 2

I examine the relationship of stock price crash and switching between real earnings management and accrual-based earnings management developed in Hypothesis 2 through following two separate regressions:

(24)

24

𝑅𝑀_𝑃𝑅𝑂𝑋𝑌_𝑡 = 𝛽₀+ 𝛽₁× 𝑁𝐶𝑆𝐾𝐸𝑊_𝑡−1+ 𝛽₂× 𝐷𝑖𝑠𝑎𝑐𝑐_𝑡 + ⁡𝛽₃× 𝑆𝑖𝑧𝑒_𝑡+ 𝛽₄× 𝑅𝑂𝐴_𝑡 + 𝛽₅× 𝐿𝑒𝑣_𝑡 + 𝛽₆ × 𝑅𝐸𝑉_𝑡 + 𝜀

𝐷𝑖𝑠𝑎𝑐𝑐_𝑡 = 𝛾₀ + 𝛾₁ × 𝑁𝐶𝑆𝐾𝐸𝑊_{𝑡 −1}+ 𝛾₂ × 𝑅𝑀_𝑃𝑅𝑂𝑋𝑌_𝑡 + ⁡𝛾₃× 𝑆𝑖𝑧𝑒_𝑡 + 𝛾₄ × 𝑅𝑂𝐴_𝑡 + 𝛾₅ × 𝐿𝑒𝑣_𝑡 + 𝛾₆ × 𝑅𝐸𝑉_𝑡 + 𝜀

Hypothesis 2 examines whether stock price cash would have impact on switching behavior of between earnings management activities. To test this hypothesis, I use 𝑅𝑀_𝑃𝑅𝑂𝑋𝑌_𝑡 and 𝐷𝑖𝑠𝑎𝑐𝑐_𝑡as dependent variables respectively and use last period stock price crash 𝑁𝐶𝑆𝐾𝐸𝑊_𝑡−1as explanatory variable. To support hypothesis 2, 𝑅𝑀_𝑃𝑅𝑂𝑋𝑌_𝑡 is predicted to be positively associated with 𝑁𝐶𝑆𝐾𝐸𝑊_{𝑡 −1}, while 𝐷𝑖𝑠𝑎𝑐𝑐_𝑡 is predicted to be negatively associated with 𝑁𝐶𝑆𝐾𝐸𝑊_{𝑡 −1}. Hence, if these relationships come true, 𝛽₁ is supposed to be positive and 𝛾₁ is supposed to be negative. Moreover, since real earnings management and accrual-based earnings management coexist, to test whether firms tradeoff between these two methods, I also put 𝐷𝑖𝑠𝑎𝑐𝑐_𝑡 and 𝑅𝑀_𝑃𝑅𝑂𝑋𝑌_𝑡 in the regression model. To support Hypothesis 2, 𝑅𝑀_𝑃𝑅𝑂𝑋𝑌_𝑡 and 𝐷𝑖𝑠𝑎𝑐𝑐_𝑡is predicted to present negative relationship. So 𝛽₂ and 𝛾₂ are predicted to be negative and should be at significant level in statistics. In addition, I also test the relationship between other proxies of real earnings management abnormal cash flow from operating (ab_CFO), abnormal production costs (ab_Prod) and abnormal discretionary expenses (ab_DisEx) and stock price crash.

4.0 Results

4.1 Descriptive Statistics

Table 1 presents summary statistics of major variables of full sample. There are total 12413 observations in my sample from year 2005 to 2013. Sample firms have average leverage ratio of -4.4007 and ROA of 0.0318.

(25)

25

Table 1: Descriptive Statistics of major variables Panel A – Descriptive statics of major variables

Variable Observations Mean Std. Dev. Min Max

NCSKEW 12413 -0.2321 0.6233 -1.9978 1.5978

Disacc 12413 -0.0383 0.1216 -0.4921 0.3415

abDisacc 12413 0.0926 0.1004 0.0007 0.6090

Size 12413 -6.27e+07 6.17e+08 -2.95e+09 2.53e+09

Lev 12413 -4.4007 63.2822 -333.9297 293.2737

ROA 12413 0.0318 0.0703 -0.2602 0.2533

Rev 12413 4.92e+09 1.41e+10 8469.06 4.33e+11

RM_PROXY 12413 0.0285 0.1819 -0.4571 0.8108

ab_CFO 12413 0.0517 0.1211 -0.2994 0.5243

ab_Prod 12413 -0.5411 0.2252 -0.7850 0.7536

ab_DisEx 12413 0.0308 0.0701 -0.0577 0.3282

Panel B – Correlation of Variables

NCSKEW Disacc ab

Disacc SIZE Lev ROA Rev

RM_ PRO XY ab_ CFO ab_

Prod ab_ DisEx

NCSKEW 1 Disacc -0.003 1 abDisacc 0.000 -0.185 1 SIZE 0.006 0.166 -0.017 1 Lev -0.007 0.028 0.025 0.076 1 ROA -0.003 0.021 -0.026 0.114 0.004 1 Rev -0.038 -0.105 0.107 -0.176 0.018 0.119 1 RM_PROXY -0.036 0.007 0.182 -0.166 -0.002 -0.151 -0.065 1 ab_CFO 0.001 -0.964 0.186 -0.145 -0.027 0.146 0.102 0.000 1 ab_Prod -0.010 0.734 0.005 0.057 0.012 -0.123 -0.087 0.336 -0.718 1 ab_DisEx 0.049 0.003 0.106 -0.033 0.020 0.123 0.015 0.070 0.032 -0.084 1

(26)

26

Panel C– Number of variables Fiscal year N % 2006 1412 11.38% 2007 1490 12.00% 2008 1528 12.31% 2009 1585 12.77% 2010 1564 12.60% 2011 1594 12.84% 2012 1621 13.06% 2013 1619 13.04% Total 12413 100%

As for one of major variables NCSKEW, average measure of stock price crash NCSKEW is -0.2321, implying that average extent of stock price crash is not very high. As a matter of fact, this implies that firms in my sample experience extreme jump in their returns on average. In contrast with average NCSKEW of -0.20 in the US sample (K im and Zhang, 2012a; 2012), average level of stock price crash in Chines market and US market is similar. In the large sample of US firms from 1964-2007 used by K im and Zhang (2012a; 2012b), mean of NCSKEW is -0.20. Firm in my sample experienced highest stock price crash has NCESKEW of 1.5978 and firm in the sample experienced highest extreme jump has NCESKEW of -1.9978.

Another main variable of interest is absolute value of discretionary accruals (abDisacc). I use absolute value of discretionary accruals because I have not predicted any specific direction for earnings management in the hypothesis. Besides, absolute value of discretionary includes accruals reversals following earnings management. The average for abDisacc is 0.0926 with a standard deviation of 0.1004, indicating average level of accrual-based earnings management is high. For sample firms, average absolute value of discretionary accruals as percentage of total assets is 9.26%. This may seem as high level. It is noticed that maximum abDisacc is 0.6090 and minimum abDisacc is 0.0007, indicating there are discretionary accruals for all sample firms. Since I do not tell direction of earnings management, this implies that all sample firms are engaging in accrual-based earnings management more or less.

(27)

27

The average for Disacc is -0.0383 with a standard deviation of 0.1216.

In terms of proxies of real earnings management, the average for RM_PROXY (sum of abnormal cash flow from operating, abnormal production costs and discretionary expenses) is 0.0285 with a standard deviation of 0.1819. Maximum RM_PRO XY is 0.8108 and minimum RM_PROXY is -0.4571. In terms of average number, RM_PRO XY is smaller than abDisacc in magnitude, which indicates that average level of real earnings management is lower than average level of accrual-based earnings management. This observation is in consistent with the s urvey of Graham et al.’s (2005) and supports that real earnings management costs more than accrual-based earnings management. The average for ab_CFO is 0.0517, ab_Prod is -0.5411 and ab_DisEx is 0.0308, respectively.

4.2 Results of Hypothesis Tests

4.2.1 Accrual-based earnings management and stock price crash

Table 2: Regression results of accrual-based earnings management and stock price crash

Test relationship between stock price crash and prior accrual-based earnings management based on following model:

NCSKEW_𝑡 = 𝛼₀+ 𝛼₁× 𝐷𝑖𝑠𝑎𝑐𝑐_𝑡−1+ 𝛼₂× 𝑅𝑀_𝑃𝑅𝑂𝑋𝑌_𝑡−1+⁡ 𝛼₃× 𝑆𝑖𝑧𝑒_𝑡 + 𝛼₄× 𝑅𝑂𝐴_𝑡+ 𝛼₅× 𝐿𝑒𝑣_𝑡 + 𝛼₆× 𝑅𝐸𝑉_𝑡 + 𝜀

(28)

28 Dependent variable 𝐍𝐂𝐒𝐊𝐄𝐖_𝒕 Coefficient 𝑫𝒊𝒔𝒂𝒄𝒄_𝒕−𝟏 0.1229** (0.043) 𝑹𝑴_𝑷𝑹𝑶𝑿𝒀𝒕 −𝟏 -0.0307 (0.358) SIZE 1.17e-11 (0.218) Lev -.000072 (0.447) ROA .0716 (0.411) Rev -1.29e-12*** (0.002) Intercept -0.2668 (0.000) Adj R2 0.12%

Notes: Standard errors clustered by firm, robust p-values reported in parentheses. *, **, *** Significant at 10%, 5%, 1% levels, respectively, based on two-tailed tests.

I start by examining whether stock price crash is associated with prior year ’s accrual-based earnings management, which has been studied by prior papers. Different from prior study, this paper takes both accrual-based accounting earnings management and real earnings management into consideration and run simultaneous regression. Table 2 presents results of regression model. Consistent with preliminary

(29)

29

analysis, it shows positive and significant relationship between discretionary accruals and stock price crash, while the relationship between prior year ’s real earnings management and stock price is not significant. The coefficient for lag absolute Disacc is 0.1229, positively significant. It indicates that the greater extent of discretionary accruals, the greater extent of stock price crash. This result is consistent with prior study which shows that stock price crash is positively associated with financial information opacity (Hutton et al., 2009; Kim and Zhang, 2012).

Furthermore, the results suggest the relationship between real earnings management and stock price crash is not significant. It may largely due to relative cost of these two earnings management methods. Real earnings management does more harm to firm than accrual-based earnings management as it has impact on operation activities of the firm. When managers have incentive to hide bad news, they may tend to choose relative lower cost method. The lager extent of accumulation of bad news, stock price crash is more likely to happen. Managers rarely choose manage earnings through real operation activities, so the relationship between prior real earnings management and stock price crash is not significant.

In terms of control variables, NCSKEW is positively and significantly associated with revenue. This implies that firms with higher revenue are more likely to experience stock price crash.

4.2.2 Stock price crash and real earnings management

Table 3: Regression results of real earnings management and prior stock price crash

Test relationship between real earnings management and prior stock price crash based on following model:

𝑅𝑀_𝑃𝑅𝑂𝑋𝑌_𝑡 = 𝛽₀+ 𝛽₁× 𝑁𝐶𝑆𝐾𝐸𝑊_𝑡−1+ 𝛽₂× 𝐷𝑖𝑠𝑎𝑐𝑐_𝑡 + ⁡𝛽₃× 𝑆𝑖𝑧𝑒_𝑡+ 𝛽₄× 𝑅𝑂𝐴_𝑡 + 𝛽₅× 𝐿𝑒𝑣_𝑡 + 𝛽₆ × 𝑅𝐸𝑉_𝑡 + 𝜀⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡(𝑎)

𝑎𝑏_𝐶𝐹𝑂_𝑡 = 𝛽₀+ 𝛽₁× 𝑁𝐶𝑆𝐾𝐸𝑊_𝑡−1+ 𝛽₂× 𝐷𝑖𝑠𝑎𝑐𝑐_𝑡 +⁡ 𝛽₃× 𝑆𝑖𝑧𝑒_𝑡+ 𝛽₄ × 𝑅𝑂𝐴_𝑡 + 𝛽₅ × 𝐿𝑒𝑣_𝑡+ 𝛽₆× 𝑅𝐸𝑉_𝑡 + 𝜀⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡(𝑏)

(30)

30 𝑎𝑏_𝑃𝑟𝑜𝑑_𝑡 = 𝛽₀+ 𝛽₁× 𝑁𝐶𝑆𝐾𝐸𝑊_𝑡−1 + 𝛽₂× 𝐷𝑖𝑠𝑎𝑐𝑐_𝑡 +⁡ 𝛽₃× 𝑆𝑖𝑧𝑒_𝑡 + 𝛽₄× 𝑅𝑂𝐴_𝑡 + 𝛽₅× 𝐿𝑒𝑣_𝑡 + 𝛽₆× 𝑅𝐸𝑉_𝑡 + 𝜀⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡(𝑐) 𝑎𝑏_𝐷𝑖𝑠𝐸𝑥_𝑡 = 𝛽₀ + 𝛽₁× 𝑁𝐶𝑆𝐾𝐸𝑊_{𝑡 −1}+ 𝛽₂× 𝐷𝑖𝑠𝑎𝑐𝑐_𝑡 + ⁡𝛽₃× 𝑆𝑖𝑧𝑒_𝑡 + 𝛽₄× 𝑅𝑂𝐴_𝑡 + 𝛽₅× 𝐿𝑒𝑣_𝑡+ 𝛽₆× 𝑅𝐸𝑉_𝑡 + 𝜀⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡(𝑑) Model (a) (b) (c) (d) Dependent variables 𝑹𝑴_𝑷𝑹𝑶𝑿𝒀_𝒕 𝒂𝒃_𝑪𝑭𝑶_𝒕 𝒂𝒃_𝑷𝒓𝒐𝒅_𝒕 𝒂𝒃_𝑫𝒊𝒔𝑬𝒙_𝒕

Coefficient Coefficient Coefficient Coefficient

𝑵𝑪𝑺𝑲𝑬𝑾_{𝒕 −𝟏} 0.0043* -0.0051* 0.0126 0.0053***

(0.083) (0.099) (0.206) (0.000)

𝑫𝒊𝒔𝒂𝒄𝒄_𝒕 -0.0315*** 0.8348*** -0.7563*** -0.0031

(0.001) (0.000) (0.000) (0.118)

SIZE -4.79e-11*** -1.05e-10*** -8.14e-11*** -7.19e-12***

(0.000) (0.000) (0.000) (0.000)

Lev 0.000036 -0.00008*** 0.00007 0.00003**

(0.153) (0.006) (0.459) (0.011)

ROA -0.3193*** 0.8067*** -1.2717*** -0.1645***

(0.000) (0.000) (0.000) (0.000)

Rev -9.53e-13 *** -3.58e-13** -1.18e-12** -7.94e-14

(0.000) (0.011) (0.011) (0.162)

Intercept - 0.0444 -0.0583 0.0825 0.0283

(0.000) (0.000) (0.000) (0.000)

Adj R2 5.17% 71.97% 17.46% 1.95%

Notes: Standard errors clustered by firm, robust p-values reported in parentheses. *, **, *** Significant at 10%, 5%, 1% levels, respectively, based on two-tailed tests.

(31)

31

Hypothesis 2 tests whether stock price crash would have effect on switching of earnings management methods. More specific, hypothesis 2a tests whether the extent of real earnings management would increase after stock price crash. Table 3 depicts results of regression model of real earnings management and stock price crash. The coefficient on prior stock price crash is significantly positive (0.0043), which indicates that the greater extent of prior year’s stock price crash, the greater extent of real earnings management. If the negative conditional return skewness (NCSKEW) increased by 1, level of real earnings management would increase by 0.43% of total assets. This result suggests that real-earnings management extent is positively associated with prior year ’s stock price crash. By contrast, the coefficient on accrual-based earnings management variable discretionary accruals is significantly negative (-0.0315), which indicates that real earnings management has negative relationship with accrual-based earnings management. When firms are engaging in large extent of real earnings management activities, they are less likely engaging in large accrual-based earnings management activities. The negative coefficient implies that firms substitute accrual-based earnings management by real earnings management after stock price crash, which is consistent with Graham et al. (2005) Zang (2006) and Cohen et al. (2008).

These results imply that the extent of real earnings management increase and firms are more likely to engage in real earnings management activities after stock price crash. In other words, high risk of stock price crash may lead to increase of real earnings management. Firms use relatively more real earnings management after stock price crash may largely due to the increase level of scrutiny of investors and increase cost of accrual-based earnings management. These findings support hypothesis 2a.

The coefficient on firm SIZE is negative and significant, which indicates the extent of real earnings management has negative association with firm size. This may due to relative higher cost of real earnings management activities. As real earnings management activities are associated with normal operation activities and do more harm to firm value compared with accrual-based earnings management, they cost

(32)

32

more to large companies. Considering complexity of operation activities, it is harder to conduct real earnings management activities for large companies. Moreover, it costs more for large companies.

The coefficient on firm performance ROA is negative and significant, which indicates firms with higher RO A are less likely use real earnings management. It suggests that firms with good performance are less likely to engage in real earnings management activities which may have impact on normal operation activities. This result is institutive as firms with good performance are less willing to engage in activities that do harm to operation activities and affect its firm performance.

The coefficient on firm revenue income REV is negative and significant, which indicates firms with larger income are less likely use real earnings management. The result is also institutive as firms with large income are also less willing to be engaging in activities that will harm their normal operation activities.

Colum 3 to 5 in Table 3 depicts the relationship of abnormal CFO, production cost and discretionary expenses and stock price crash. The results suggest although the sum of real earnings management increase after stock price crash, the impact of stock price crash on individual item is different.

The coefficient of NCSKEW on ab_CFO is negative and significant, which indicates that abnormal cash flow from operating activities is negatively associated with prior stock price crash. Moreover, abnormal CFO is significantly positive associated with current year’s abnormal accruals. The coefficient of NCSKEW on ab_Prod is positive but insignificant in statistics. Despite the weak statistics, the positive coefficient suggests that stock price crash may lead to increase of abnormal product costs. The results also imply abnormal production costs are negatively associated with current year’s discretionary accruals. The coefficient of NCSKEW on ab_DieEx is positive and significant, which indicates that abnormal discretionary expenses such as R&D expenses, advertising expenses and administrative expenses are positively associated with prior stock price crash. The strong relationship suggests stock price crash would lead to increase of abnormal discretionary expenses. As for coefficient on Disacc, it is negative but insignificant. Negative coefficient implies

(33)

33

firms are switching accrual-based earnings management by managing discretionary expenses.

Putting all the results together, I find evidence support that stock price crash could increase the extent of real earnings management. I also find evidence support firms trade off real earnings management and accrual-based earnings management after stock price crash as real earnings management is negatively associated with accrual-based earnings management. However, the results also indicate that stock price crash may have different impact on different real earnings management activities. Of all the three activities, only discretionary expenses are significantly positive associated with prior stock price crash. Abnormal production costs are positively associated with prior stock price crash, while abnormal cash flow from operation is significantly negative associated with prior stock price crash. O nly abnormal production costs are significantly negative associated with current year ’s discretionary accruals. Discretionary expenses are negative associated with current year ’s discretionary accruals, but the relationship is insignificant in statistics. Despite weak statistics, there is tradeoff between accrual earnings management and managing earnings through production costs and discretionary expenses. Whereas, it is noted that the relationship between abnormal cash flow from operation and discretionary accruals is positive and significant. This implies that companies may not substitute accrual-based earnings management and real earnings management through managing cash flow from operation. It is also noticed that the relationship between abnormal cash flow from operation and discretionary accruals is positive but insignificant. This implies that stock price crash may not have strong impact on managing cash flow from operation.

The results imply that after stock price crash, most companies may tend to manage discretionary expenses such as R&D expenses, advertising expenses and administrative expenses. Stock price crash has significant impact on these discretionary expenses. These results are in consistent with findings of the survey conducted by Graham et al. (2005). In their findings, there are 80 percent of CFOs in the survey stating that they would decrease research and development (R&D)

Stock price crash and switching of earnings management : empirical evidence from Chinese listed companies