Taming the dragon: stock exchange comment letters and earnings volatility

Abstract

This study explores the impact of comment letters on earnings volatility motivated by the broad scope of stock exchange comment letters in China and regulators’ concerns regarding earnings-related issues. Using a manually collected sample of Chinese comment letters, we find that receiving a comment letter is associated with increased earnings volatility. Our baseline results survive a host of exercises to mitigate endogeneity, including difference-in-differences, propensity score matching, entropy balancing, and instrumental variables. Channel analysis reveals that comment letters are associated with increased earnings volatility through dampened supply chain stability (economic channel) and reduced use of earnings management to smooth earnings (accounting channel). Our findings contribute to the literature on comment letters in non-US economies and provide important and timely implications for regulators.

Keywords:

• comment letter
• earnings volatility
• supply chain
• earnings management

JEL:

• G18
• K22
• M41
• M48
• M42

1. Introduction

Comment letters represent an important regulatory oversight of corporate disclosures (Cassell et al. 2013). Prior studies primarily focus on the Securities and Exchange Commission (SEC) comment letters in the US and identify various economic consequences of receiving a comment letter, including increased cost of debt (Cunningham et al. 2017), negative stock returns (Dechow et al. 2016), enhanced corporate disclosure (Bozanic et al. 2017), and a switch to real activities manipulation from accrual-based earnings management (Cunningham et al. 2020). Research on comment letters issued by other jurisdictions is scarce, although they may have features distinct from those of the SEC comment letters.

Accompanied by dramatically increased number of comment letters issued by Chinese stock exchanges since 2014,¹ recent studies have focused on Chinese settings and explored the determinants of receiving a comment letter from the perspectives of information asymmetry (Chen et al. 2021), political ranking (Feng and Wei 2024), political connections (Chen et al. 2020), and the characteristics of senior executives (Quan et al. 2021), as well as the consequences of comment letters, such as stock price reaction (Duan et al. 2019), stock price synchronicity (Cao et al. 2021, 2023), analyst forecasts (Hu et al. 2022a), audit opinion (Hu et al. 2022b), and cash holding (Yao and Hong 2023). However, most prior research focuses on a specific type of comment letters, whereas Chinese stock exchange comment letters cover various aspects of financial reporting and firms’ day-to-day operations. Therefore, we extend this stream of research by examining the impact of stock exchange comment letters on firms’ earnings volatility.²

We focus on corporate earnings volatility as it is an overarching concern, especially in China. As a developing economy undergoing transformation and adjustment, and an emerging market with imperfect legal, regulatory, and investor protection systems, China has exhibited great macroeconomic volatility, policy uncertainty, and information asymmetry (Cheng and Cheung 2021, Lam and Du 2004, Wen et al. 2021). Consequently, Chinese firms experience greater earnings volatility than their US counterparts (Huang and Wirjanto 2012, Luo et al. 2017). Furthermore, when juxtaposed with developed countries, China exhibits relatively lower levels of institutional ownership; regulations for transparent and timely information disclosures are less stringent, resulting in a complex and opaque information environment (Gu et al. 2019).

As an aggregate of all economic transactions, earnings information plays a primary and critical role in Chinese markets, allowing various stakeholders to understand and analyze corporate operations (Chen et al., 2001). Earnings volatility thus considerably matters. Volatile earnings may discourage investors’ acquisition of a firm’s earnings information, as they would consider such information less predictive of future prospects and thus less useful. Additionally, volatile earnings may adversely affect investors’ judgment of the firm’s actual operations and performance (Bryan and Mason 2020). Therefore, investors tend to consider greater earnings volatility to indicate higher investment risk and demand a higher risk premium (Francis et al. 2004). Auditors and securities analysts also closely monitor earnings information. Evidence suggests that volatile earnings attract auditor attention and are associated with higher audit fees (Bryan et al. 2018, Rahman and Wu 2021) and auditor resignations (Bryan and Mason 2020). Greater earnings volatility also makes it harder to predict future earnings and increases errors in analyst forecasts (Dichev and Tang 2009).

Given the considerable impact of earnings volatility on the cost of capital, market value, and firm performance (Allyannis and Rountree 2005), managers tend to prefer smooth earnings to volatile earnings and even sacrifice economic value to reduce earnings volatility (Graham et al. 2005). Additionally, China is profoundly influenced by Confucian culture characterised by conservatism and uncertainty avoidance. Chinese investors exhibit greater risk aversion (Huang et al. 2022, Li et al. 2013), making their investment decisions more susceptible to earnings volatility. In short, earnings volatility in China warrants further investigation.

We find it particularly pertinent to investigate the repercussions of Chinese stock exchange comment letters on earnings volatility. The annual review process of the Chinese stock exchanges emphasises earnings-related matters. In instances where companies encounter issues, these exchanges issue comment letters and mandate supplementary disclosures, clarifications, risk alerts, revisions, and restatements to address these concerns. Chinese comment letters may also be extended to encompass various aspects of a firm’s day-to-day operational management. Consequently, it is logical to anticipate that companies receiving comment letters could experience fluctuations in earnings because such volatility is already embedded in financial reporting anomalies and operational disruptions. The bedrock of earnings inherently reflects a firm’s daily operational activities.

However, it remains uncertain ex ante whether and how receiving a comment letter influences forward earnings volatility. On the one hand, comment letters could fullfill a governance role by compelling struggling firms to enhance their disclosures, explanations, and adjustments, thereby potentially mitigating earnings volatility. On the other hand, these letters might contribute to escalated earnings volatility through economic and accounting channels, which we discuss below.

Dichev and Tang (2009) suggest that both economic and accounting factors influence earnings volatility. Economic factors refer to the economic shocks and changes in a firm’s business environment. These factors arise from outside the firm and may increase earnings volatility as they affect a firm’s business patterns. Accounting factors refer to managers’ accounting choices that influence reported earnings. In line with this framework, we propose two distinct channels through which comment letters increase earnings volatility.

The first channel is an external economic mechanism through which comment letters affect a firm’s supply chain. The contents of comment letters are not only visible to problematic firms but also accessible to external stakeholders. Besides sending letters directly to firms, Chinese stock exchanges also publish detailed comment letters on their official websites – unconditionally accessible and highly regarded sites used by stock exchanges to disclose official information. Moreover, to ensure the availability and timeliness of disclosure, the China Securities Regulatory Commission (CSRC) authorises several mainstream media such as the China Securities Journal, Shanghai Securities News, Securities Times, and Securities Daily to disseminate comment letter information.

Additionally, mainstream online media compete with reporting and disseminating comment letter-related news.³ Consequently, comment letters are easily accessible to the public and provide a readily available channel for acquiring corporate information, thereby playing an important role in Chinese capital markets. As important stakeholders in the supply chain, suppliers and customers are keen to ensure supply chain stability. Comment letters uncover a firm’s disclosure deficiencies, potential violations, and problematic practices, and send a negative signal to outsiders regarding the possibility of scrutiny by regulators in the future (Duan et al. 2019). This adversely affects suppliers’ and customers’ perceptions of the firm’s credibility and uncertainty, and weakens its bargaining power in the supply chain and affects the terms and conditions in their procurement and sales contracts. Therefore, we expect greater earnings volatility for firms subjected to comment letters.

The second channel is an internal accounting mechanism through which comment letters may refrain managers from using earnings management to smoothen earnings. Ensuring the authenticity of corporate disclosure is an important target of stock exchange comment letter regulation.⁴ Consequently, stock exchanges are particularly concerned about earnings-related issues in their annual review processes⁵ and issue comment letters to companies that are observed or suspected to have earning-related opportunistic managerial behaviours. Notably in our sample, approximately 69.12% of the comment letters (80% of the financial reporting comment letters) are related to earnings-related issues.

Companies with severe earnings-manipulating behaviours and fraudulent practices are subject to continuing investigations and penalties from the stock exchanges and CSRC. Even if the issues identified in a comment letter are not directly related to earnings, they may indicate to outsiders that the firm has disclosure deficiencies and is under heightened scrutiny by regulators. Using a sample of SEC comment letters, Cunningham et al. (2020) found that, facing tightened monitoring after receiving a comment letter, managers tend to reduce accrual-based earnings management and increase real-activity earnings management to avoid undue scrutiny.

Since Chinese stock exchange comment letters cover firms’ daily operations, firms receiving a comment letter may have little incentive to increase their level of real earnings management. Earnings management is widely used by managers to smoothen earnings (Trueman and Titman 1988). Graham et al. (2005) survey 401 financial executives and show that 97% of the surveyed CFOs preferred smooth earnings over volatile earnings, as smooth earnings help reduce risk premiums and enhance earnings predictability. In the same survey, 78% of respondents admitted that they would sacrifice value for smooth earnings, which is consistent with ample empirical evidence that managers tend to use earnings management to smoothen earnings (Barton 2001, Pincus and Rajgopal 2002). Taken together, reduced use of earnings management could weaken a firm’s ability to smoothen earnings (Kirschenheiter and Melumad 2002, Leuz et al. 2003, Roychowdhury 2006), predicting increased earnings volatility.

To test our predictions, we manually collect 4660 comment letters from the websites of the Shanghai and Shenzhen Stock Exchanges from 2014 to 2018. Our baseline analysis shows a positive association between the receipt (frequency) of comment letters and volatility of annual and quarterly earnings. To mitigate identification concerns, we adopt a difference-in-differences (DID) design based on a propensity score-matched (PSM) sample and obtain consistent results. The DID analysis not only confirms our parallel trend assumption (i.e. there are no systematic differences between the treated and control groups) but also shows that the impact of comment letters persists for the next three years. Further, we use leadership change in stock exchanges (CSRC) and a firm’s geographic distance to regulators as instruments for comment letters and perform instrumental variable (IV) analysis. Again, the inference remains unchanged.

In our channel analysis, we find that receiving comment letters increases earnings volatility through economic and accounting mechanisms. In the economic channel, a firm’s bargaining power in its supply chain weakens after receiving a letter, manifested by the decreased level of trade credit, increased volatility of trade credit, and greater supplier and customer concentration. Moreover, operating revenue and expenses become more volatile for firms receiving comment letters. For the accounting channel, we show that earnings management is a mediator that links comment letters to earnings volatility. That is, firms receiving comment letters reduce both accrual-based and real-activity earnings management, which limits their ability to smoothen earnings and increases earnings volatility.

Our study contributes to the literature in two notable ways. First, our investigation extends previous research on comment letters. The existing literature primarily focuses on SEC comment letters in the US (Bozanic et al. 2017, Brown et al. 2018, Cunningham et al. 2017, Cunningham et al. 2020), whereas a handful of studies use a non-US setting. Unlike prior research, our study explores the different features of the Chinese stock exchange and SEC comment letters, focusing on how comment letters affect earnings volatility. Our setting is unique, as Chinese stock exchange comment letters review a much broader range of topics than their US counterparts (Drienko and Sault 2011). We show that, unlike SEC comment letters, most frequently issued comment letters in China do not focus on financial reporting issues, but on firms’ daily operational concerns (i.e. concern letters).

Moreover, in contrast to evidence from the US, we fail to find a switch-up from accrual-based earnings management to real earnings management for comment letter firms. Rather, we show that letter-receiving firms tend to reduce both types of earnings management, consistent with the notion that Chinese firms are not only wary of receiving financial reporting comment letters, but are also alert to concern letters that probe into a firm’s real activities. Overall, the unique characteristics of Chinese stock exchange comment letters extend the scope of this study.

Second, this study contributes to the literature on the determinants of earnings volatility. Previous research shows that earnings volatility affects various stakeholders (Bryan et al. 2018, Bryan and Mason 2020, Dichev and Tang 2009, Graham et al. 2005). Dichev and Tang (2009) suggest that economic and accounting factors may drive earnings volatility. Aligning with this insight, our study demonstrates that comment letters affect earnings volatility through changes in bargaining dynamics in the supply chain (economic channel) and the ability to use earnings management to smoothen earnings (accounting channels). Overall, our findings demonstrate that stock exchange comment letters are important predictors of forward earnings volatility.

Our study has important implications for regulators. We show that comment letters have product market and accounting consequences. Although regulators may dampen fluctuations in corporate earnings by identifying anomalies and irregularities in financial reporting and day-to-day operations of a firm; comment letters to mitigate these issues may increase firms’ earnings volatility. Considering that Chinese companies experience higher earnings volatility than their US counterparts (Huang and Wirjanto 2012), our results contribute to regulators’ improved assessment of the impact of comment letters on earnings-related issues.

2. Institutional background

In China, the China Securities Regulatory Commission (CSRC)⁶ and Shanghai and Shenzhen Stock Exchanges are jointly responsible for comment letter supervision. However, in practice, the two stock exchanges, by virtue of their position in the front line of supervision and unique information advantages, primarily manage the issuance of comment letters.⁷ The Division of Corporate Management of the stock exchanges conducts the review process. For firms with disclosure deficiencies and potential violations, the Division issues comment letters and requires them to provide supplemental information or make amendments. As with the SEC’s comment letters, there can be multiple rounds after the issuance of a comment letter until concerns are fully addressed. For firms that refuse to respond or modify as required and for those who violate laws and regulations, the stock exchange report to the CSRC, which may conduct a series of follow-up investigations and even take punitive regulatory measures, such as warnings and fines, to ensure the effectiveness of comment letter supervision. Chen et al. (2020) document several determinants of receiving stock exchange comment letters in China, including political connections, corporate governance structures, financial restatement, and firm age, consistent with Cassell et al. (2013) and Heese et al. (2017), who delve into corresponding themes using the US data.

3. Literature review and hypothesis development

3.1. Comment letters

Prior research has explored the determinants and consequences of comment letters. For instance, Cassell et al. (2013) find that low profitability, high complexity, small auditors, and weak corporate governance are positively associated with receiving the SEC’s comment letters. Kubick et al. (2016) show that the SEC tends to issue comment letters to firms with aggressive tax avoidance. Chen et al. (2021) find that companies in China with severe information asymmetry problems are more likely to receive merger and acquisition comment letters. Studies based in the US also document that comment letters positively relate to a firm’s political connections (Correia 2014, Heese et al. 2017); Chen et al. (2020) investigate whether this relationship holds for Chinese publicly listed firms and find similar results. By contrast, Feng and Wei (2024) show that state-owned enterprises (SOEs) exhibit a lower likelihood of receiving comment letters than non-SOE firms. Furthermore, their findings indicate an even lower probability of firms controlled by the central government receiving comments. Moreover, Quan et al. (2021) show that board secretaries with financial expertise appear to be negatively associated with the likelihood and frequency of annual report comment letters in China.

For the consequences of comment letters, prior studies have shown that comment letters may change the information environment of a company. For example, Bozanic et al. (2017) find that firms increase their corporate disclosures after receiving a comment letter. Similarly, firms tend to amend filings in response to issues identified in comment letters (Johnston and Petacchi 2017). Wang (2016) finds that firms receiving a letter experience decreased analysts’ forecast errors as such firms revise their segment disclosures. Brown et al. (2018) examine the spillover effect of comment letters and find that firms tend to modify their disclosures when their industry leaders or peers receive comment letters from the SEC.

Additionally, on the US evidence, Cunningham et al. (2020) find that, after receiving a comment letter, firms reduce accrual-based earnings management and increase real activities manipulation due to tightened regulatory scrutiny from the SEC. Kubick et al. (2016) show that firms reduce tax avoidance after receiving tax-related comment letters. Dechow et al. (2016) find that insiders make greater stock sales before comment letters become publicly available. Auditors reassess litigation risks and charge higher audit fees when their clients receive comment letters (Gietzmann and Pettinicchio 2014). Institutional investors adjust their equity holding downward in firms that receive comment letters (Gietzmann and Isidro 2017). Banks increase interest rates when borrowers receive a comment letter (Cunningham et al., 2017).

Based on Chinese evidence, Duan et al. (2019), using a sample of comment letters on annual reports issued by the Shanghai Stock Exchange, find that stock price reactions to comment letter announcements are significantly negative, and comment letter firms are more likely to amend their annual reports than non-comment letter firms. Hu et al. (2022a, b) highlight that the availability of comment letters and firms’ responses to the public leads to a positive (negative) correlation between comment letters and analysts’ forecast accuracy (optimism), underscoring auditors’ tendency to issue modified or conservative audit opinions to client companies that receive comment letters. Furthermore, Cao et al. (2021) and Cao et al. (2023) find that stock exchange comment letters are associated with reduced stock price synchronicity, and Yao and Hong (2023) find that firms receiving cash-related comment letters tend to reduce their excess cash holdings.

Although the literature has documented various economic consequences of comment letters, few studies have examined how comment letters influence earnings volatility. We intend to fill this void in the literature as earnings volatility is economically meaningful and represents serious concerns of various stakeholders, especially in China (Allyannis and Rountree 2005, Dichev and Tang 2009, Huang and Wirjanto 2012, Jayaraman 2008, Luo et al. 2017).

3.2. Earnings volatility

The critical importance of earnings volatility is featured by Graham et al. (2005), who document that an overwhelming majority of surveyed financial executives prefer smooth earnings to volatile earnings, and that 78% of the surveyed CFOs express that they would even sacrifice economic value to avoid volatile earnings. Senior executives prefer smooth earnings because earnings volatility reflects firm risk and affects the decision-making of stakeholders, such as capital providers, analysts, and credit rating agencies. Dichev and Tang (2009) find that analysts’ forecast errors increase with earnings volatility, indicating that even sophisticated investors encounter challenges when attempting to forecast prospects amid such volatility. Earnings volatility also increases auditors’ perceived risk of client firms. Bryan et al. (2018) find that auditor fees are higher for firms with more volatile earnings. Bryan and Mason (2020) show a positive association between earnings volatility and auditor resignations. Huang and Wirjanto (2012) reveal that the P/E ratio is negatively associated with earnings volatility in China because earnings volatility appears to cancel out the positive effect of growth opportunities on P/E ratios.

Extant research suggests that the determinants of earnings volatility are derived from economic and accounting factors (Dichev and Tang 2009). First, earnings volatility arises from instability in the operating environment. For example, economic shocks and changes in the product market can ivnfluence a firm’s business and make its earnings more volatile. Second, reported earnings reflect managerial discretion in the application of accounting principles. One example is matching revenue and expenses, which refers to the use of accounting accruals to influence reported earnings. In addition, managers can use earnings management through real activity manipulation. Both the earnings management strategies can modify earnings patterns and affect earnings volatility. We expect that receiving a comment letter may increase earnings volatility through both economic and accounting channels.

3.3. Hypothesis development

Comment letters are issued to probe a company’s problematic practices and are published in detail on stock exchange websites and further disseminated by financial media. Receiving a comment letter is widely considered bad news and thus adversely affects stakeholders’ perception of comment letter firms (Gietzmann and Isidro 2013, Gietzmann and Pettinicchio 2014), which is accompanied by negative market reaction (Bozanic et al. 2017, Duan et al. 2019). As the key stakeholders in supply chains and product markets, suppliers and customers are wary of a firm’s negative publicity, from which they can infer the instability of the firm’s operations and, in turn, an unstable supply chain. Graham et al. (2005) suggest that business stability is an important consideration for suppliers and customers when deciding on trade volumes, terms, and conditions. Therefore, receiving a comment letter may cast doubtv on the firm’s credibility and stability, weaken its bargaining power in the supply chain, and adversely affect the outcomes of economic transactions (e.g. decreased accounts payable to suppliers and lower advances from customers). A change in supply chain dynamics is an economic factor that adversely affects a comment letter firm’s bargaining power in transactions. Furthermore, a comment letter might alert and potentially disillusion its customer base due to the negative publicity. Therefore, the wholesale and retail markets of a letter-receiving company can experience adverse impacts. While companies can attempt to counteract the adverse repercussions of comment letters by engaging in negotiations with their suppliers and customers, such efforts come at a cost and can be time-consuming to effectively address concerns. Consequently, the cumulative effect of comment letters may introduce disruptions in supply chains and product markets, influencing a firm’s operational dynamics. Consequently, this dynamic contributes to heightened earnings volatility.

Comment letters may also make earnings more volatile through accounting channels by restricting earnings management. Earnings-related issues are central to the stock exchange annual review process. For firms observed or suspected of having earning-related opportunistic managerial behaviours, stock exchanges may issue comment letters requiring additional disclosures, explanations, revisions, or financial restatements in an attempt to curtail earnings manipulation. These firms may be subjected to constant scrutiny, multiple rounds of inquiry, and subsequent investigations until earnings-related issues are fully addressed. In addition to regulatory oversight, problematic accounting practices targeted by comment letters may also attract public attention. Heightened regulatory scrutiny and public awareness increase the penalty and outrage costs of earnings management, consequently constraining such practices (Cunningham et al. 2020). Notably, although the issues identified in a comment letter are not directly related to earnings-related problems, they signal that the firm may have a disclosure deficiency and potential violation, which may increase the likelihood of tightened oversight by regulators and the public regarding earning-related managerial opportunistic behaviours, thereby restricting earnings management.

Prior literature shows that managers regard earnings volatility as high-risk and tend to use earnings management to smoothen earnings (Barton 2001, Graham et al. 2005, Pincus and Rajgopal 2002, Trueman and Titman 1988). When a firm achieves low (high) earnings during a certain period, managers are inclined to make upward (downward) adjustments to smoothen their earnings. Burgstahler and Dichev (1997) provide evidence that senior executives use earnings management to avoid earnings declines and losses. The accrual-based and real activity-based earnings management can be used to smoothen earnings. Leuz et al. (2003) argue that insiders reduce the variability of reported earnings by managing accruals, whereas Roychowdhury (2006) suggests that managers may also engage in earnings management through real activity manipulation to avoid reporting losses or missing certain earnings benchmarks.

Focusing on SEC comment letters, Cunningham et al. (2020) document a ‘switch-up’ earnings management pattern; that is, comment letter firms tend to use more real earnings management as a substitute for accrual earnings management. This is because managers’ discretion to use accounting accruals is a major focus of SEC’s comment letters, whereas earnings management through real activities is less likely to be commented upon in the SEC’s review process. However, comment letters issued by Chinese stock exchanges target both accounting practices (the use of accruals) and real activities (operational activities), with the former targeted by financial reporting comment letters and the latter by concern letters. Therefore, we expect Chinese comment letter firms to reduce both types of earnings management. Consequently, comment letters limit firms’ ability to use earnings management to smoothen earnings, thereby increasing earnings volatility.

These arguments suggest that comment letters affect a firm’s earnings volatility through economic and accounting channels. As both channels predict a positive association between comment letters and earnings volatility, we hypothesise the following:

Hypothesis: Receiving a comment letter has a positive relation with forward earnings volatility.

4. Research design

4.1. Data and sample

We manually collect 4660 comment letters during the period of 2014–2018 from the websites of the Shanghai and Shenzhen Stock Exchanges. Financial data are obtained from the China Securities Market Accounting Research (CSMAR) database and RESSET databases. Panel A of Table 1 shows the distribution of comment letters by type and year. Among the eight types of comment letters issued by stock exchanges, concern letters have the highest frequency, whereas supervision letters are the least frequent. During the sample period, the number of comment letters increased annually, indicating that comment letters have become an important regulatory oversight employed by Chinese stock exchanges. Panel B of Table 1 reports the distribution of the comment letters by industry and year.⁸ Firms in the manufacturing industry receive the most letters, whereas those in the health and social work industry account for the least. Thus, we control for industry and year fixed effects in our regression analyses.

Table 1. Sample distribution.

Panel A: Distribution of comment letters by type & year
Year	Financial reporting	M&A and Restructuring				Concern letter	Supervision letter	Other types	Total number of comment letters
		M&A review opinion	License reorganisation	Non-license reorganisation	Major assets reorganisation
2014	0	2	0	3	0	0	0	0	5
2015	127	133	177	60	0	84	0	12	593
2016	208	141	288	59	0	402	2	40	1140
2017	291	0	170	70	107	322	1	224	1185
2018	448	0	136	53	123	723	0	254	1737
Total	1074	276	771	245	230	1531	3	530	4660
Panel B: Distribution of comment letters by industry & year
Year	A	B	C	D	E	F	G	H	I	J	K	L	M	N	O	Q	R	S	Total
2014	1		3			1													5
2015	79	7	325	8	9	27	11	12	31	1	35	6	3	1			14	23	593
2016	165	20	654	20	17	38	16	16	68	5	59	11	3	1	2	1	17	27	1140
2017	170	16	657	26	9	47	23	10	45	4	54	7	7	2	2	1	75	31	1185
2018	287	17	960	22	12	49	33	10	86	6	83	13	13	4	4	2	92	44	1737
Total	702	60	2,599	76	47	162	83	48	230	16	231	37	26	8	8	4	198	125	4660

Panel A shows the distribution of comment letters by type and year. Panel B reports the distribution of comment letters by industry and year. Industries are classified according to the Guidance on the Industry Category of Listed Companies (Revised in 2012) issued by CSRC, where A = Agriculture, forestry, farming and fishery, B = Mining, C = Manufacturing, D = Electricity, thermal, gas and water, E = Construction, F = Wholesale and retail trade, G = Transport, storage and post, H = Hotels and catering service, I = Information transmission, software and information technology, J = Financial industry, K = Estate, L = Leasing and operating Services, M = Scientific research and technical service industry, N = Water conservancy, environment and public facilities management, O = Resident services, repairs and other services, Q = Health and social work, R = Culture, sports and entertainment industry, S = Comprehensive.

Considering the unique accounting and regulatory requirements of the finance and insurance industries, we exclude them from our sample. Firm-years with missing information to construct our regressors are also excluded. Our final sample consists of 10,049 firm-year observations from 2014 to 2018.

4.2. Key variables

4.2.1. Comment letters

Following previous studies (Bozanic et al. 2017, Cassell et al. 2013), we use three proxies to capture the receipt and frequency of comment letters. Our first proxy (CL1) is an indicator variable equal to one if a firm receives a comment letter in a year and zero otherwise. The second proxy (CL2) uses the natural logarithm of one plus the number of comment letters received by a firm in a year. In addition, we use the natural logarithm of one plus the highest frequency with which a firm is inquired about the same concern as the third proxy (CL3), which captures the severity of the issues in comment letters.⁹

4.2.2. Earnings volatility

We construct two primary measures of earnings volatility, following previous studies (Edmonds et al. 2015, Ferris et al. 2017). First, we use a three-year rolling standard deviation of the adjusted return on total assets (EV1). Specifically, we measure a firm’s return on assets as the ratio of profit before interest and tax to total assets. We then calculate the return on assets adjusted for annual-industry average (AROA). Based on Eq. (1), we calculate the standard deviation of AROA over the window from year t to t + 2 and use it as the first proxy for earnings volatility. (1) $\mathrm{EV}{1}_{i\left[t,t+2\right]}=\sqrt{{\displaystyle \frac{1}{N-1}{{\sum }_{t=k}^N\left(\mathrm{ARO}{A}_{\mathrm{it}}-{\displaystyle \frac{1}{N}{\sum }_{t=1}^N\mathrm{ARO}{A}_{\mathrm{it}}}\right)}^2}}(N=3,t=k,k+1,k+2)$(1) Our second measure is the standard deviation of a company’s quarterly industry-adjusted return on assets during the period from year t to t + 2 using quarterly data (EV2) (Ferris et al. 2017). Specifically, we calculate the return on assets adjusted for the industry average for each quarter based on a firm’s quarterly statements. Then, we calculate the standard deviation of the quarterly industry-adjusted ROA using a window from year t to t + 2 (12 quarters in total).¹⁰ Notably, both measures of earnings volatility are computed during periods t to t + 2, that is, three years after receiving a comment letter. This is because we expect that for firms receiving comment letters, it takes time for their suppliers and customers to modify the terms and conditions of transactions in the supply chain and product markets. Similarly, comment letter firms take time to alter their earnings management patterns. As such, although the comment letters in our sample date from 2014 to 2018, we calculate the ROA from 2014 to 2020.

4.3. Baseline model

Following the existing literature (Cunningham et al. 2020, Gietzmann and Pettinicchio 2014), we use Eq. (2) to test the impact of comment letters on earnings volatility.

(2) $\begin{array}{rl}E{V}_{i[t,t+2]}& =a_0+a_{1}C{L}_{\mathrm{it}}+a_2\mathrm{SIZ}{E}_{\mathrm{it}}+a_3\mathrm{AG}{E}_{\mathrm{it}}+a_4\mathrm{GRO}{W}_{\mathrm{it}}+a_5\mathrm{BIG}{4}_{\mathrm{it}}+a_6\mathrm{CAS}{H}_{\mathrm{it}}\\ & +a_7\mathrm{IN}{D}_{\mathrm{it}}+a_8\mathrm{OW}{N}_{\mathrm{it}}+a_9\mathrm{DU}{A}_{\mathrm{it}}+a_{10}\mathrm{PE}{N}_{\mathrm{it}}+\mathrm{Industr}{y}_i+\mathrm{Yea}{r}_t+{\nu }_{\mathrm{it}}\end{array}$(2) where, suppressing subscripts, the dependent variable EV is the three-year forward earnings volatility measured by EV1 and EV2. Our test variable is CL, the proxy for comment letters, using CL1, CL2 and CL3; a₁ is the coefficient of the test variable, and the hypothesis predicts a significantly positive a₁.

We control for an array of variables following prior studies (Faccio et al. 2016, Jayaraman 2008, John et al. 2008). Firm size (SIZE) is the natural logarithm of the total assets at the beginning of the period. Firm age (AGE) is measured by the number of years since the establishment of a firm. GROW is the proxy for growth opportunity measured by the operating income growth rate. BIG4 is an indicator variable equal to one if a firm’s auditor is one of the big four audit firms, and zero otherwise. Cash holding (CASH) is measured by the ratio of cash and cash equivalents to total assets at the beginning of the period. Board independence (IND) is the percentage of independent directors on a board of directors (Lu and Wang 2018). Ownership concentration (OWN) is the total shareholding of the top ten shareholders. We also control for CEO duality (DUA), which equals one if the CEO is also the chairperson and zero otherwise (Gietzmann et al. 2016, Goergen et al. 2020). PEN is an indicator variable equal to one if the company receives a penalty from the CSRC in a year and zero otherwise (Chen et al. 2005). We also include industry (Industry) and year (Year) fixed effects to identify coefficients from the within-industry and within-year variations in earnings volatility.¹¹ All continuous variables are winsorized at the top and bottom 1% to control for the impact of extreme values.

5. Main results

5.1. Summary statistics

Table 2 Panel A reports the summary statistics for our full sample. The mean values of our two measures for earnings volatility (EV1 and EV2) are 0.054 and 0.032, respectively. The mean EV1 is greater than the mean ROA volatility (0.019) reported by Ferris et al. (2017), suggesting that Chinese firms experience greater earnings volatility. The mean CL1 is 0.206, indicating that 20.6% of the sample firm-years have received a comment letter, which is greater than the percentage (11%) reported by Chen et al. (2020) who use an earlier sample, suggesting that Chinese stock exchanges have issued an increasing number of comment letters in more recent years. The growth opportunity for our sample firms (GROW) as captured by the average operation income growth rate is 20.9%. The mean values of SIZE and AGE are 22.47 and 22.51, respectively. The big four audit firms (BIG4) audit 6.3% of our sample. The mean and median values of cash holding (CASH) are 0.150 and 0.121, respectively. The proportion of independent directors (IND) in our sample firms is 37.4%. The average total shareholding of the top ten shareholders (OWN) is 54.97% and 22.6% of our firm-years have CEO duality (DUA). 16.4% of our sample receive a regulatory penalty (PEN). Overall, our descriptive statistics are comparable to those reported in prior literature using data from China (Chen et al. 2020, Liu et al. 2015).

Table 2. Summary statistics.

Panel A: Summary statistics for variables in the main analysis
Variable	N	Mean	SD	Min	Q1	Median	Q3	Max
EV1	10,049	0.054	0.090	0.002	0.015	0.029	0.049	0.666
EV2	10,049	0.032	0.040	0.004	0.012	0.019	0.035	0.259
CL1	10,049	0.206	0.404	0.000	0.000	0.000	0.000	1.000
CL2	10,049	0.176	0.363	0.000	0.000	0.000	0.000	1.386
CL3	10,049	0.153	0.306	0.000	0.000	0.000	0.000	1.099
SIZE	10,049	22.470	1.315	19.470	21.600	22.340	23.260	26.220
AGE	10,049	22.510	5.166	12.000	19.000	22.000	27.000	35.000
GROW	10,049	0.209	0.671	−0.658	−0.048	0.086	0.259	4.792
BIG4	10,049	0.063	0.242	0.000	0.000	0.000	0.000	1.000
CASH	10,049	0.150	0.121	0.009	0.066	0.115	0.197	0.633
IND	10,049	0.374	0.053	0.333	0.333	0.357	0.429	0.571
OWN	10,049	54.970	15.060	21.900	43.870	54.980	65.610	89.840
DUA	10,049	0.226	0.418	0.000	0.000	0.000	0.000	1.000
PEN	10,049	0.164	0.371	0.000	0.000	0.000	0.000	1.000
Panel B: Summary statistics for comment letter firms (CL1 = 1) versus non-comment letter firms (CL1 = 0)
Variable	CL1 = 0		CL1 = 1		Mean Diff
	N	Mean	N	Mean
EV1	7980	0.046	2069	0.084	−0.039***
EV2	7980	0.028	2069	0.048	−0.019***
CL2	7980	0.000	2069	0.853	−0.853***
CL3	7980	0.000	2069	0.742	−0.742***
SIZE	7980	22.515	2069	22.319	0.195***
AGE	7980	22.430	2069	22.810	−0.380**
GROW	7980	0.188	2069	0.287	−0.099***
BIG4	7980	0.070	2069	0.034	0.036***
CASH	7980	0.150	2069	0.149	0.002
IND	7980	0.373	2069	0.379	−0.006***
OWN	7980	55.350	2069	53.519	1.831***
DUA	7980	0.214	2069	0.273	−0.059***
PEN	7980	0.141	2069	0.255	−0.114***

This table reports the summary statistics of our sample. Panel A presents the summary statistics for the full sample, while Panel B presents mean differences between comment letter firms and non-comment letter firms. Our sample consists of Chinese listed firms from 2014 until 2018. All the continuous variables are winsorized at the top and bottom 1% level. See Appendix A for variable definitions.

Panel B in Table 2 presents the mean differences in the main variables between comment letter firms and non-comment letter firms. For non-comment letter firms (CL1 = 0), the mean EV1 (EV2) is 0.046 (0.028), while the mean for comment letter firms (CL1 = 1) is 0.084 (0.048). The differences are statistically significant at the 1% level, indicating that firms that receive comment letters have greater earnings volatility than non-receiving firms.

Table 3 reports the correlation matrix. It shows that the correlations between the proxies for comment letters (CL1, CL2 and CL3) and earnings volatility measures (EV1 and EV2) are significantly positive (p < 0.01). We also estimate variance inflation factors (VIFs) for all the regressors in Eq. (2) and find that VIFs are well below the cutoff threshold of 10 to trigger multicollinearity concerns (Kennedy 1992).

Table 3. Correlation matrix.

Variable	EV1	EV2	CL1	CL2	CL3	SIZE	AGE	GROW	BIG4	CASH	IND	OWN	DUA	PEN
EV1	1	0.705***	0.179***	0.187***	0.183***	0.182***	0.013	−0.067***	−0.075***	−0.075***	0.025**	−0.096***	0.038***	0.105***
EV2	0.796***	1	0.164***	0.174***	0.169***	−0.225***	0.003	−0.072***	−0.071***	−0.016	0.025**	−0.067***	0.057***	0.119***
CL1	0.173***	0.197***	1	0.994***	0.997***	−0.051***	0.028***	0.026***	−0.061***	−0.018*	0.048***	−0.049***	0.057***	0.124***
CL2	0.205***	0.232***	0.950***	1	0.996***	−0.054***	0.0303***	0.023**	−0.061***	−0.021**	0.051***	−0.051***	0.059***	0.131***
CL3	0.188***	0.212***	0.981***	0.968***	1	−0.052***	0.029***	0.022**	−0.061***	−0.018**	0.049***	−0.050***	0.057***	0.128***
SIZE	−0.130***	−0.192***	−0.060***	−0.068***	−0.064***	1	0.077***	0.111***	0.278***	−0.118***	−0.011	0.319***	−0.126***	−0.089***
AGE	−0.009	0.011	0.029***	0.036***	0.034***	0.049***	1	−0.078***	0.019*	−0.027***	−0.043***	−0.106***	−0.098***	0.038***
GROW	−0.013	0.016	0.059***	0.053***	0.053***	0.043***	0.009	1	0.006	0.084***	−0.006	0.101***	0.035***	−0.057***
BIG4	−0.050***	−0.059***	−0.061***	−0.059***	−0.061***	0.332***	0.014	−0.013	1	−0.033***	0.021**	0.199***	−0.060***	−0.068***
CASH	−0.059***	−0.027***	−0.005	−0.014	−0.006	−0.122***	−0.020**	0.178***	−0.032***	1	0.019*	0.037***	0.035***	−0.052***
IND	0.036***	0.040***	0.046***	0.052***	0.048***	0.001	−0.057***	0.005	0.030***	0.016	1	0.004	0.120***	−0.002
OWN	−0.069***	−0.069***	−0.049***	−0.054***	−0.053***	0.355***	−0.113***	0.109***	0.214***	0.055***	−0.006	1	−0.049***	−0.075***
DUA	0.039***	0.053***	0.057***	0.060***	0.057***	−0.121***	−0.101***	0.022***	−0.060***	0.032***	0.124***	−0.055***	1	0.035***
PEN	0.115***	0.141***	0.124***	0.141***	0.134***	−0.096***	0.040***	−0.001	−0.068***	−0.033***	0.002	−0.076***	0.035***	1

This table reports the correlation matrix for the variables in our main analysis. Spearman coefficients are in the upper triangle, and Pearson coefficients are in the lower triangle. All the continuous variables are winsorized at the top and bottom 1% level. See Appendix A for variable definitions. ***, **, * statistically significant at the 1%, 5%, 10% levels, respectively.

5.2. Baseline results

We present the baseline results of Eq. (2) in Table 4. Consistent with our hypothesis, the coefficients on comment letters (CL) are significantly positive (p < 0.01) across all six columns, suggesting that firms that receive comment letters (CL1) have greater forward earnings volatility than firms that do not. Earnings volatility also increases with the number of comment letters (CL2) and number of inquiries for the same issue (CL3). The coefficients of CL in Columns (1), (2), and (3) are 0.0321, 0.0435, and 0.0466, respectively, which are economically significant given that the mean (standard deviation) of EV1 is 0.054 (0.090).¹² Similarly, the coefficients on CL in Columns (4), (5), and (6) are 0.0155, 0.0209, and 0.0224, respectively, suggesting economic significance as the mean (standard deviation) of EV2 is 0.032 (0.040). Note that we control for industry and year fixed effects in the baseline regressions. For robustness, we control for firm and stock exchange fixed effects and find consistent results (not tabulated).

Table 4. Baseline regression results of comment letters (CL) on earnings volatility (EV)

Variable	(1)	(2)	(3)	(4)	(5)	(6)
	EV = EV1			EV = EV2
	CL = CL1	CL = CL2	CL = CL3	CL = CL1	CL = CL2	CL = CL3
CL	0.0321***	0.0435***	0.0466***	0.0155***	0.0209***	0.0224***
	(8.9933)	(9.1882)	(9.1652)	(9.4694)	(9.8228)	(9.7201)
SIZE	−0.0101***	−0.0099***	−0.0100***	−0.0059***	−0.0058***	−0.0058***
	(−8.7076)	(−8.5349)	(−8.6356)	(−9.8056)	(−9.6529)	(−9.7484)
AGE	0.0001	0.0000	0.0001	0.0002	0.0002	0.0002
	(0.2978)	(0.1735)	(0.2247)	(1.1878)	(1.0772)	(1.1234)
GROW	−0.0009	−0.0010	−0.0009	0.0011	0.0010	0.0011
	(−0.5043)	(−0.5732)	(−0.4965)	(1.3089)	(1.2520)	(1.3255)
BIG4	0.0011	0.0011	0.0012	0.0031*	0.0031*	0.0031*
	(0.4297)	(0.4343)	(0.4732)	(1.8272)	(1.8303)	(1.8564)
CASH	−0.0531***	−0.0515***	−0.0531***	−0.0172***	−0.0164***	−0.0172***
	(−6.3594)	(−6.2738)	(−6.3867)	(−3.8193)	(−3.6924)	(−3.8300)
IND	0.0354*	0.0319	0.0339*	0.0222**	0.0205*	0.0215**
	(1.7815)	(1.6272)	(1.7162)	(2.0179)	(1.8845)	(1.9624)
OWN	−0.0001	−0.0001	−0.0001	0.0001	0.0001	0.0001
	(−0.1719)	(−0.1604)	(−0.1549)	(0.8413)	(0.8592)	(0.8600)
DUA	0.0023	0.0019	0.0022	0.0017	0.0016	0.0017
	(0.7976)	(0.6809)	(0.7585)	(1.1901)	(1.0834)	(1.1566)
PEN	0.0206***	0.0191***	0.0199***	0.0111***	0.0104***	0.0107***
	(6.2117)	(5.8185)	(6.0258)	(7.1323)	(6.7235)	(6.9462)
Constant	0.2708***	0.2663***	0.2693***	0.1542***	0.1521***	0.1535***
	(9.7662)	(9.6670)	(9.7250)	(10.6483)	(10.5632)	(10.6167)
Industry FE	Included	Included	Included	Included	Included	Included
Year FE	Included	Included	Included	Included	Included	Included
N	10,049	10,049	10,049	10,049	10,049	10,049
R^2-adj	0.1043	0.1130	0.1080	0.0995	0.1096	0.1036

This table reports OLS estimates for our baseline model of Eq. (2). The dependent variable, earnings volatility (EV), has two variants: EV1 and EV2. The independent variable comment letter (CL) has three variants CL1, CL2, and CL3. See Appendix A for variable definitions. ***, **, * statistically significant at the 1%, 5%, 10% levels, respectively. The t-statistics presented in parentheses are based on standard errors clustered by firm and robust standard errors.

Regarding our control variables, SIZE and CASH are significantly negative (p < 0.01), indicating that large, cash-rich firms have lower earnings volatility, which is consistent with previous research (Edmonds et al. 2015, Li et al. 2013). The coefficients of PEN are significantly positive (p < 0.01), indicating that penalty firms have greater forward earnings volatility than non-penalty firms.

5.3. Robustness tests

5.3.1. Alternative measures of earnings volatility

In addition to EV1 and EV2, we use five alternative measurements for earnings volatility (EV3, EV4, EV5, EV6, and EV7). Our first alternative measure is based on Eq. (3). Specifically, we construct EV3 as the three-year (from t to t + 2) moving average range (i.e. the difference between the maximum and minimum) of the annual-industry adjusted return on assets. (3) $\mathrm{EV}{3}_{i\left[t,t+2\right]}={\displaystyle \frac{1}{N}\left[{max}_{n\in \left[0,2\right]}\mathrm{ARO}{A}_{\mathrm{it}\pm n}-{min}_{n\in \left[0,2\right]}\mathrm{ARO}{A}_{\mathrm{it}\pm n}\right]}$(3) The second alternative measure is based on Eq. (4). Specifically, we employ the three-year (from t to t + 2) average of the annual standard deviation of the quarterly industry-adjusted return on assets. (4) $\mathrm{EV}{4}_{i\left[t,t+2\right]}={\sum }_t^n\mathrm{sd}.\mathrm{ROA}\mathrm{\_quarter}/N$(4) where N = 3, t = k, k + 1, k + 2, sd.ROA_quarter is the annual standard deviation of the quarterly industry-adjusted return on assets.

For the third alternative measurement, we use the standard deviation of the quarterly return on assets in year t to measure annual earnings volatility (EV5) and extend our test period by including firm-years in 2019 so that we have a one-year lead (lag) on our outcome variable (test variable) to control for potential reverse causality.

Finally, we use stock return volatility in the capital market as a proxy for corporate earnings volatility. Specifically, we calculate the standard deviation of daily (EV6) and monthly stock returns (EV7) for each year. Parallel to our main proxies for earnings volatility (EV1 and EV2), we calculate the average stock returns over the window from year t to t + 2 for EV6 and EV7.

Note, so far, all our measures are positively related to earnings volatility. For robustness, we use earnings smoothness, which conversely measures earnings volatility. Following prior research (Bhattacharya et al. 2003), we calculate earnings smoothness (ES) based on Eq. (5). (5) $E{S}_{\mathrm{it}}={\displaystyle \frac{\mathrm{SD}\left(\mathrm{CF}{O}_{\mathrm{it}-3}/A_{\mathrm{it}-4},\mathrm{CF}{O}_{\mathrm{it}-2}/A_{\mathrm{it}-3},\mathrm{CF}{O}_{\mathrm{it}-1}/A_{\mathrm{it}-2},\mathrm{CF}{O}_{\mathrm{it}}/A_{\mathrm{it}-1}\right)}{\mathrm{SD}\left(N{I}_{\mathrm{it}-3}/A_{\mathrm{it}-4},N{I}_{\mathrm{it}-2}/A_{\mathrm{it}-3},N{I}_{\mathrm{it}-1}/A_{\mathrm{it}-2},N{I}_{\mathrm{it}}/A_{\mathrm{it}-1}\right)}}$(5) where, CFO is net cash flow from operating activities, A is total assets at the end of the period, and NI is net profit. SD stands for standard deviation.

The results are presented in Table A3 of the Online Appendix. Panel A reports the results using EV3, EV4 and EV5 as the proxies for earnings volatility. The coefficients on comment letters (CL) are significantly positive (p < 0.01) in all columns. Panel B reports the results using EV6, EV7, and ES as the dependent variables. The coefficients of CL in columns (1) to (6) remain significantly positive (p < 0.01), whereas in columns (7) to (9), where the dependent variable captures earnings smoothness (ES), the coefficients of CL are significantly negative (p < 0.01 or p < 0.05), suggesting that receiving a comment letter predicts increased earnings volatility (dampened earnings smoothness). Overall, our baseline results are robust to alternative earnings volatility measures.

5.3.2. Controlling for CEO power and previous receipt of a comment letter

We further examine whether our results are robust to control for the influence of two potential confounding factors. The first factor is CEO power. CEOs who delegate authority to other executives with different opinions tend to discuss and reach a consensus within the top management team (Abernethy et al. 2022). In doing so, aggressive decision-making can be reduced, thereby leading to less volatile earnings. By contrast, powerful CEOs are more likely to make aggressive decisions and thus increase earnings volatility (Adams et al. 2005, Cheng 2008). Further, powerful CEOs may engage in opportunistic managerial behaviours, such as tunnelling and earnings manipulation, thus increasing the likelihood of law violations and receiving comment letters (Dechow et al. 1995, Dechow et al. 2016).

Following Adams et al. (2005), we use CEO duality to measure CEO power in our baseline model. For robustness, we include the following additional proxies for CEO power: (1) CEO pay slice (PaySlice), measured as the percentage of CEO compensation in the total compensation of the top three executives (Bebchuk et al. 2011); (2) CEO pay gap (PayGap), measured as the difference between the CEO’s total compensation and the median total compensation of other executives (Armstrong et al. 2012, Kini and Williams 2012); (3) CEO pay mix (PayMix), measured by the ratio of long-term variable pay (stock options, restricted stock and appreciation rights) to total compensation (Armstrong et al. 2012); (4) CEO tenure (Tenure) (Abernethy et al. 2015). We replace DUA in Eq. (2) with these alternative measures and report the results in Columns (1) – (8) in Table A4 of the Online Appendix. After controlling for CEO power, the coefficients of CL in all the columns remain significantly positive (p < 0.01 or p < 0.05).¹³ Overall, our findings are robust when CEO power is controlled for.

Considering the possible persistence of firms receiving comment letters, we further control for whether a firm has previously received a comment letter by including an indicator PCL equal to one if the firm has received a comment letter in previous years, and zero otherwise in our baseline model of Eq. (2). Columns (9) and (10) of Table A4 report the results. After controlling for PCL, the coefficient of CL1 remain significantly positive (p < 0.01). The coefficients of PCL are also significantly positive (p < 0.01), suggesting a habitus for receiving comment letters.

5.4. Identification strategy

In our baseline analysis, although we control for a broad set of confounding factors that may affect both receiving comment letters and earnings volatility, our estimates could be biased because of endogeneity issues arising from correlated omitted variables. For better identification, we perform the following analyses. First, we follow previous studies (Beck et al. 2010, Bozanic et al. 2017) and employ a PSM-DID design. We use all the control variables in Eq. (2) and one-year lagged earnings volatility (Lag_EV5) as covariates to perform a year-by-year one-to-one PSM of the treatment (CL1 = 1) and control groups (CL1 = 0). We then fit a two-way fixed-effect DID model using the PSM sample. The regression results indicate that a firm’s earnings volatility increases after receiving comment letters, which is consistent with our main results. We further test the parallel trend assumption and examine the dynamic effect of comment letters using a dynamic DID design. Our results indicate that there is no violation of the parallel trend assumption and that the impact of comment letters on earnings volatility persists in the subsequent three years. For robustness, we also examine the effect of comment letters on one-year earnings volatility constructed by quarterly ROA (EV5) using the dynamic DID test and reach a consistent conclusion. Taken together, our inferences are robust to the PSM-DID design.

Second, a concern arising from the above PSM procedure is that matching leads to the loss of a considerable portion of the sample. To address this issue and align our treatment and control firms across higher moments of the covariates, following prior studies (Chahine et al. 2020, Francoeur et al. 2023), we include all control variables in Eq. (2) and Lag_EV5 as covariates and perform an entropy balancing analysis. Our results indicate successful balancing and suggest that our inferences are robust to entropy balancing.

Furthermore, we employ an instrumental variable approach to mitigate endogeneity because of the correlated omitted variables in unobservable dimensions. Our first instrument is constructed based on the leadership change in stock exchanges and CSRC (LeaderChange), and our second instrument is based on the average geographic distance between a firm and regulators (Distance). We believe that both instruments meet the relevance and exclusion criteria. The results of the first-stage regression, under-identification, weak identification, and over-identification tests collectively indicate that our instrumental variables are valid. The second-stage regression results are consistent with our prior findings. Overall, our inferences are robust when the IV approach is used.¹⁴ Further details are provided in Online Appendix.

5.5. Channel tests

5.5.1. Economic channels

Comment letters may adversely affect suppliers’ and customers’ perceptions of a comment letter firm’s business stability, which will negatively affect the firm’s bargaining power in its supply chain and product markets. Consequently, suppliers and customers may change the terms and conditions of trade, thereby increasing earnings volatility. For this economic channel, we explore the mediating effect based on three factors: (1) trade credit volatility (TCV), measured as the three-year rolling standard deviation of trade credit during the period of t to t + 2, where trade credit (TC) is the sum of accounts payable, notes payable, and advances received scaled by the total asset (Mateut 2014); (2) the volatility of operating revenue and operating expenses in a year (RCV1), measured as the sum of the standard deviation of quarterly operating revenue and standard deviation of quarterly operating expenses. (3) Volatility of operating revenue and expenses across years (RCV2). We measure this as the sum of the standard deviation of annual operating revenue from t to t + 2 and the standard deviation of annual operating expenses during the same period.

Note that the measurements for supply chain stability are based on financial statement items, which are susceptible to managerial manipulation, and may relate to the accounting mechanism. To better distinguish between economic and accounting mechanisms, we measure supply chain stability using supplier and customer concentration, which are presumably more difficult for managers to manipulate. Following prior studies (Ho et al. 2023, Zhang et al. 2020), we contend that a higher supplier concentration indicates that firms have fewer suppliers to provide raw materials or intermediate products for production and thus have greater reliance and switching costs on major suppliers, which will increase suppliers’ bargaining power, making it easier for them to extract economic rent by controlling product quality and pricing. Consequently, greater supplier concentration is associated with a firm’s increased exposure to supply disruptions, which weakens supply chain stability. Similarly, higher customer concentration suggests greater reliance on a few customers, and lower bargaining power in negotiations with them (Dong et al. 2021, Huang et al. 2016).

For measurement, we include supplier concentration (SUC), measured as the ratio of annual purchases from the top five suppliers to total purchases (Zhang et al. 2020); customer concentration (CUC), measured as the ratio of annual sales of the top five customers to total sales (Dong et al. 2021); and supply chain concentration (SCC), calculated as the average of the top five suppliers’ purchase ratios and top five customers’ sales ratio.¹⁵ As with our main measurements for earnings volatility, we use a three-year rolling window to calibrate the following: (1) average supplier concentration (ASUC), measured as the three-year moving average of supplier concentration (SUC) during the period t to t + 2; (2) average customer concentration (ACUC), measured as the three-year moving average of customer concentration (CUC) during the period t to t + 2; (3) average supply chain concentration (ASCC), measured as the three-year moving average of supply chain concentration (ASCC) during the period t to t + 2; (4) supplier concentration volatility (VSUC), calculated as the three-year rolling standard deviation of supplier concentration (SUC) during the period t to t + 2; (5) customer concentration volatility (VCUC), calculated as the three-year rolling standard deviation of customer concentration (CUC) during the period t to t + 2; and (6) supply chain concentration volatility (VSCC), calculated as the three-year rolling standard deviation of supply chain concentration (SCC) during the period t to t + 2.

To test the mediating effect, we employ the approach of Ferris et al. (2017). In the baseline analysis, we show that comment letters are positively associated with earnings volatility. Next, we examine the association between comment letters (CL) and channel variables based on Eq. (6). Finally, we augment our baseline model by adding channel variables (CV) and examine the relationship between comment letters, mediators, and earnings volatility using Eq. (7). (6) $\mathrm{CV}(\mathrm{ChannelVariables}{)}_{\mathrm{it}}={\gamma }_0+{\gamma }_{1}C{L}_{\mathrm{it}}+\mathrm{Control}{s}_{\mathrm{it}}+\mathrm{Industr}{y}_i+\mathrm{Yea}{r}_t+e_{\mathrm{it}}$(6) (7) $E{V}_{i[t,t+2]}={\eta }_0+{\eta }_{1}C{L}_{\mathrm{it}}+{\eta }_{2}C{V}_{\mathrm{it}}+\mathrm{Control}{s}_{\mathrm{it}}+\mathrm{Industr}{y}_i+\mathrm{Yea}{r}_t+n_{\mathrm{it}}$(7) The results are summarised in Table 5. Panel A reports the results of the comment letters (CL1) on trade credit (TC), trade credit volatility (TCV), and the volatility of operating revenue and expenses (RCV1 and RCV2, respectively). In column (1), the coefficient of CL1 is significantly negative (p < 0.05), suggesting that comment letter firms experience a decline in trade credit, which is consistent with the notion that comment letter firms experience less favourable terms and conditions in transactions with suppliers and customers. In columns (2) – (4), the coefficients of CL1 are significantly positive (p < 0.01 or p < 0.05), indicating that receiving a comment letter is associated with increased TCV and increased RCV1 and RCV2, consistent with our predictions.

Table 5. Channel test: Economic channels.

Panel A: The impact of comment letters on supply chain and product markets
Variable	(1)	(2)	(3)	(4)
	TC	TCV	RCV1	RCV2
CL1	−0.0092**	0.0034***	0.0171**	0.0403***
	(−2.3364)	(3.3187)	(2.0594)	(4.9445)
Constant	−0.2547***	0.0507***	0.3827***	0.6290***
	(−5.1186)	(5.1405)	(3.9916)	(6.7409)
Control variables	Included	Included	Included	Included
Industry FE	Included	Included	Included	Included
Year FE	Included	Included	Included	Included
N	7301	6805	7972	10,041
R^2-adj	0.1797	0.0450	0.0992	0.0825
Panel B: The impact of CL1, TVC, RVC1, and RVC2 on earnings volatility (EV1 and EV2)
Variable	(1)	(2)	(3)	(4)	(5)	(6)
	EV1	EV2	EV1	EV2	EV1	EV2
	CV = TVC		CV = RVC1		CV = RVC2
CL1	0.0206***	0.0103***	0.0260***	0.0122***	0.0293***	0.0142***
	(5.4213)	(6.1155)	(7.3576)	(7.5524)	(8.6473)	(9.0390)
CV	0.4618***	0.2239***	0.0485***	0.0201***	0.0705***	0.0314***
	(6.1001)	(6.6802)	(4.6165)	(4.5305)	(6.6302)	(7.1387)
Constant	0.2005***	0.1275***	0.2383***	0.1429***	0.2254***	0.1347***
	(7.1132)	(8.5721)	(8.8418)	(9.9246)	(8.4711)	(9.6075)
Control variables	Included	Included	Included	Included	Included	Included
Industry FE	Included	Included	Included	Included	Included	Included
Year FE	Included	Included	Included	Included	Included	Included
N	6805	6805	7972	7972	10,041	10,041
R^2-adj	0.1255	0.0958	0.1236	0.1007	0.1370	0.1324
Panel C: The impact of comment letters on supply chain concentration
Variable	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
	Annual value			The average during the period of t to t + 2			The volatility during the period of t to t + 2
	SUC	CUC	SCC	ASUC	ACUC	ASCC	VSUC	VCUC	VSCC
CL1	1.4621**	1.8105***	1.8182***	1.5153***	2.1885***	1.9791***	1.5330***	1.6613***	1.3072***
	(2.3560)	(2.7708)	(3.5949)	(2.5843)	(3.3997)	(4.0495)	(7.5278)	(7.6506)	(7.7133)
Constant	130.7695***	112.2723***	119.9409***	127.9376***	111.5407***	119.8580***	25.5063***	27.1981***	20.9859***
	(15.6516)	(12.3116)	(17.2760)	(15.5043)	(12.1471)	(17.1965)	(11.5598)	(11.8906)	(11.3750)
Control variables	Included	Included	Included	Included	Included	Included	Included	Included	Included
Industry FE	Included	Included	Included	Included	Included	Included	Included	Included	Included
Year FE	Included	Included	Included	Included	Included	Included	Included	Included	Included
N	8921	9191	9319	8445	8754	8876	9080	9178	9277
R^2-adj	0.1477	0.1484	0.1870	0.1597	0.1571	0.2075	0.0818	0.0884	0.0950
Panel D: The impact of CL1, CC, ACC, and VCC on earnings volatility (EV1 and EV2)
Variable	(1)	(2)	(3)	(4)	(5)	(6)
	EV1			EV2
	CV = SCC	CV = ASCC	CV = VSCC	CV = SCC	CV = ASCC	CV = VSCC
CL1	0.0325***	0.0314***	0.0304***	0.0157***	0.0152***	0.0145***
	(8.8777)	(8.8621)	(8.4550)	(9.3376)	(9.2722)	(8.7876)
CV	0.0005***	0.0007***	0.0024***	0.0002***	0.0003***	0.0013***
	(5.6117)	(6.6751)	(7.8879)	(5.7086)	(6.3741)	(9.6111)
Constant	0.2097***	0.1821***	0.2156***	0.1213***	0.1107***	0.1227***
	(7.1183)	(6.3167)	(7.4676)	(7.9996)	(7.2465)	(8.0993)
Control variables	Included	Included	Included	Included	Included	Included
Industry FE	Included	Included	Included	Included	Included	Included
Year FE	Included	Included	Included	Included	Included	Included
N	9319	8876	9277	9319	8876	9277
R^2-adj	0.1135	0.1136	0.1249	0.1091	0.1139	0.1284

This table reports the results of economic channels. Panel A presents the results of comment letters (CL1) on trade credit (TC), trade credit volatility (TCV), the volatility of operating revenue (RCV1), and the volatility of operating expenses (RCV2). Panel B reports the results of comment letters (CL1) and channel variables (TVC, RCV1, and RCV2) on earnings volatility (EV1 and EV2). Panel C presents the results of comment letters (CL1) on the proxies for supplier concentration (SUC, ASUC, and VSUC), customer concentration (CUC, ACUC, and VCUC) and supply chain concentration (SCC, ASCC, and VSCC). Panel D reports the results of comment letters (CL1) and channel variables (SCC, ASCC, and VSCC) on earnings volatility (EV1 and EV2). See Appendix A for variable definitions. ***, **, * statistically significant at the 1%, 5%, 10% levels, respectively. The t-statistics presented in parentheses are based on standard errors clustered by firm and robust standard errors.

Panel B of Table 5 reports the results of comment letters (CL1) and the mediators (TCV, RCV1 and RCV2) of earnings volatility (EV). In all six columns, the coefficients of TCV, RCV1, and RCV2 are significantly positive (p < 0.01). Moreover, the coefficients of CL1 are smaller than those reported in Table 4 in the absence of the channel variables. Our results suggest that comment letters increase earnings volatility through trade credit volatility and the volatility of operating revenue and expenses, thereby supporting the economic mechanism.

Panel C of Table 5 reports the results of comment letters (CL1) on the yearly and three-year supply chain concentrations (SUC, CUC, SCC, ASUC, ACUC, ASCC, VSUC, VCUC, and VSCC). The coefficients of CL1 are significantly positive (p < 0.01or p < 0.05) across all columns, indicating that receiving a comment letter is not only associated with higher yearly (SUC, CUC, and SCC) and three-year supplier, customer, and supply chain concentrations (ASUC, ACUC, and ASCC) but also relates to greater volatility of supplier, customer, and supply chain concentrations across years (VSUC, VCUC, and VSCC).

Panel D of Table 5 reports the results of comment letters (CL1) and mediators (VSUC, VCUC, and VSCC) of earnings volatility (EV).¹⁶ The coefficients of CL1 across all columns are significantly positive (p < 0.01). The coefficients of supply chain concentration (CV) in all columns are also significantly positive (p < 0.01). Our results suggest that comment letters are associated with higher earnings volatility through an increased level and volatility of supply chain concentration, lending credence to economic mechanisms.

5.5.2. Accounting channels

Next, we explore the accounting mechanism by investigating whether comment letters affect earnings volatility via earnings management (EM). Following previous studies (Cunningham et al. 2020, Zang 2012), we use discretionary accruals to proxy for accrual-based earnings management (AEM) based on the modified Jones model as follows: (8) $\mathrm{Accrual}{s}_t/A_{t-1=}{\alpha }_{1+}{\alpha }_2\left(1/A_{t-1}{)}_{+}{\alpha }_3\right((\mathrm{\Delta }{S}_{\mathrm{it}}-\mathrm{\Delta RE}{C}_{\mathrm{it}})/A_{t-1}{)}_{+}{\alpha }_4\left(\mathrm{PP}{E}_{\mathrm{it}}/A_{t-1}\right)$(8) where, Accruals_t is the earnings before extraordinary items and discontinued operations, minus operating cash flow in t; A_t-1 is the total assets at the end of t-1; ΔS_it is the change in net sales from year t-1 to t; ΔREC_it is the change in net receivables from year t-1 to t; PPE_it is the gross property, plant, and equipment. Our test variable AEM is the estimated residual from Eq. (8).

Following previous studies (Cunningham et al. 2020, Roychowdhury 2006), we construct real-activities-based earnings management (REM) as the aggregate of two separate measures: abnormal production and abnormal discretionary expenditures based on Eq. (9). (9) $\mathrm{RE}{M}_{\mathrm{it}}=\mathrm{APRO}{D}_{\mathrm{it}}\text{–}\mathrm{AEX}{P}_{\mathrm{it}}$(9) where, APROD is the abnormal production cost. A larger value indicates a higher degree of production manipulation. AEXP is abnormal discretional expenditure; a smaller value of AEXP represents a higher degree of expenditure manipulation.

We also construct a measure of total earnings management (EM) as the sum of AEM and REM (Cohen and Zarowin 2010, Cunningham et al. 2020). Next, we follow Ferris et al.’s (2017) approach to test the mediating effect of earnings management. In the main analysis, we demonstrated that comment letters are positively associated with earnings volatility. In this section, we re-estimate the relationship between comment letter (CL1) and earnings volatility (EV) using Eq. (10), using lagged CL1 as our test variable to allow sufficient time for comment letter firms to adjust their earnings management. Next, following Cunningham et al. (2020), we construct Eqs. (11) – (13) to examine the impact of comment letter (CL1) on earnings management (include AEM, REM, and EM). Then we test the impacts of CL1 and EM on earnings volatility (EV) using Eq. (14). (10) $\begin{array}{rl}E{V}_{i[t,t+2]}& =a_0+a_{1}C{L}_{\mathrm{it}-1}+a_2\mathrm{SIZ}{E}_{\mathrm{it}}+a_3\mathrm{AG}{E}_{\mathrm{it}}+a_4\mathrm{GRO}{W}_{\mathrm{it}}+a_5\mathrm{BIG}{4}_{\mathrm{it}}+a_6\mathrm{CAS}{H}_{\mathrm{it}}\\ & +a_7\mathrm{IN}{D}_{\mathrm{it}}+a_8\mathrm{OW}{N}_{\mathrm{it}}+a_9\mathrm{DU}{A}_{\mathrm{it}}+a_{10}\mathrm{PE}{N}_{\mathrm{it}}+\mathrm{FirmFE}+{\nu }_{\mathrm{it}}\end{array}$(10) (11) $\begin{array}{rl}\mathrm{RE}{M}_{\mathrm{it}}& ={\text{β}}_0+{\text{β}}_{1}C{L}_{\mathrm{it}-1}+{\text{β}}_2\mathrm{MarketShar}{e}_{\mathrm{it}}+{\text{β}}_3\mathrm{Zscor}{e}_{\mathrm{it}}+{\text{β}}_4\mathrm{INS}{T}_{\mathrm{it}}+{\text{β}}_5\mathrm{ET}{R}_{\mathrm{it}}+{\text{β}}_6\mathrm{BIG}{4}_{\mathrm{it}}\\ & +{\text{β}}_7\mathrm{AuditTenur}{e}_{\mathrm{it}-1}+{\text{β}}_8\mathrm{NO}{A}_{\mathrm{it}}+{\text{β}}_9\mathrm{Cycl}{e}_{\mathrm{it}}+{\text{β}}_{10}\mathrm{RO}{A}_{\mathrm{it}}+{\text{β}}_{11}\mathrm{AdjLnAsset}{s}_{\mathrm{it}}\\ & +{\text{β}}_{12}\mathrm{Mt}{B}_{\mathrm{it}-1}+{\text{β}}_{13}\mathrm{StockIssuanc}{e}_{\mathrm{it}+1}+{\text{β}}_{14}\mathrm{Ear}{n}_{\mathrm{it}}+\mathrm{FirmFE}+{\nu }_{\mathrm{it}}\end{array}$(11) (12) $\begin{array}{rl}\mathrm{AE}{M}_{\mathrm{it}}& ={\text{β}}_0+{\text{β}}_{1}C{L}_{\mathrm{it}-1}+{\text{β}}_2\mathrm{MarketShar}{e}_{\mathrm{it}}+{\text{β}}_3\mathrm{Zscor}{e}_{\mathrm{it}}+{\text{β}}_4\mathrm{INS}{T}_{\mathrm{it}}+{\text{β}}_5\mathrm{ET}{R}_{\mathrm{it}}+{\text{β}}_6\mathrm{BIG}{4}_{\mathrm{it}}+\\ & +{\text{β}}_7\mathrm{AuditTenur}{e}_{\mathrm{it}-1}+{\text{β}}_8\mathrm{NO}{A}_{\mathrm{it}}+{\text{β}}_9\mathrm{Cycl}{e}_{\mathrm{it}}+{\text{β}}_{10}\mathrm{RO}{A}_{\mathrm{it}}+{\text{β}}_{11}\mathrm{AdjLnAsset}{s}_{\mathrm{it}}\\ & +{\text{β}}_{12}\mathrm{Mt}{B}_{\mathrm{it}1}+{\text{β}}_{13}\mathrm{StockIssuanc}{e}_{\mathrm{it}+1}+{\text{β}}_{14}\mathrm{Pred}\mathrm{\_RE}{M}_{\mathrm{it}}+{\text{β}}_{15}\mathrm{Unpred}\mathrm{\_RE}{M}_{\mathrm{it}}\\ & +\mathrm{FirmFE}+{\nu }_{\mathrm{it}}\end{array}$(12) (13) $\begin{array}{rl}E{M}_{\mathrm{it}}& ={\text{β}}_0+{\text{β}}_{1}C{L}_{\mathrm{it}-1}+{\text{β}}_2\mathrm{MarketShar}{e}_{\mathrm{it}}+{\text{β}}_3\mathrm{Zscor}{e}_{\mathrm{it}}+{\text{β}}_4\mathrm{INS}{T}_{\mathrm{it}}+{\text{β}}_5\mathrm{ET}{R}_{\mathrm{it}}+{\text{β}}_6\mathrm{BIG}{4}_{\mathrm{it}}+\\ & +{\text{β}}_7\mathrm{AuditTenur}{e}_{\mathrm{it}-1}+{\text{β}}_8\mathrm{NO}{A}_{\mathrm{it}}+{\text{β}}_9\mathrm{Cycl}{e}_{\mathrm{it}}+{\text{β}}_{10}\mathrm{RO}{A}_{\mathrm{it}}+{\text{β}}_{11}\mathrm{AdjLnAsset}{s}_{\mathrm{it}}\\ & +{\text{β}}_{12}\mathrm{Mt}{B}_{\mathrm{it}-1}+{\text{β}}_{13}\mathrm{StockIssuanc}{e}_{\mathrm{it}+1}+{\text{β}}_{14}\mathrm{Ear}{n}_{\mathrm{it}}+\mathrm{FirmFE}+{\nu }_{\mathrm{it}}\end{array}$(13) (14) $\begin{array}{rl}E{V}_{i[t,t+2]}& ={\lambda }_0+{\lambda }_{1}C{L}_{\mathrm{it}-1}+{\lambda }_{2}C{V}_{\mathrm{it}}+{\lambda }_3\mathrm{SIZ}{E}_{\mathrm{it}}+{\lambda }_4\mathrm{AG}{E}_{\mathrm{it}}+{\lambda }_5\mathrm{GRO}{W}_{\mathrm{it}}+{\lambda }_6\mathrm{BIG}{4}_{\mathrm{it}}\\ & +{\lambda }_7\mathrm{CAS}{H}_{\mathrm{it}}+{\lambda }_8\mathrm{IN}{D}_{\mathrm{it}}+{\lambda }_9\mathrm{OW}{N}_{\mathrm{it}}+{\lambda }_{10}\mathrm{DU}{A}_{\mathrm{it}}+{\lambda }_{11}\mathrm{PE}{N}_{\mathrm{it}}+\mathrm{FirmFE}+{\nu }_{\mathrm{it}}\end{array}$(14)

We add control variables following Zang (2012) and Cunningham et al. (2020), including the firms’ market share (MarketShare), firms’ financial health (Zscore), level of institutional ownership (INST), firms’ effective tax rate (ETR), auditor scrutiny (BIG4 and AuditTenure), the extent to which earnings have been manipulated (NOA), extent to which earnings can be manipulated (Cycle), firm performance leading up to the fourth quarter (ROA), industry-adjusted firm size (AdjLnAssets), growth and investment opportunities (MtB), equity financing (StockIssuance) and pre-managed earnings (Earn). Zang (2012, p. 676) contends that ‘real activities manipulation must occur during the fiscal year and is realised by the fiscal year-end, after which managers still have the chance to adjust the level of accrual-based earnings management.’ Accordingly, in AEM model of Eq. (12), we replace Earn with the predicted and unpredicted levels of REM, which are estimated using Eq. (11). See Appendix A for the definitions of the variables. We estimate all equations using ordinary least squares model with robust standard errors clustered by firm. Following Cunningham et al. (2020), we include firm fixed effects to control for unobservable, time-invariant firm characteristics. This approach facilitates a more precise comparison between our findings and those of Cunningham et al. (2020).

The results are reported in Panel A of Table 6. Columns (1) – (3) show the impact of comment letters on earnings management. While the coefficient of CL1 in column (1) is not statistically significant, the coefficients in columns (2) and (3) are significantly negative (p < 0.05 or p < 0.1), suggesting that comment letter firms have lower accrual-based earnings management (AEM) and total earnings management (EM). Columns (4) and (5) report the impact of comment letters on earnings volatility. The coefficients of CL1 are significantly positive (p < 0.01), consistent with our main results. Columns (6) and (7) show the impact of comment letters (CL1) and mediators (EM) on earnings volatility. The coefficients of EM are significantly negative (p < 0.01 or p < 0.05), indicating that lower earnings management is associated with greater earnings volatility. The coefficients of CL1 remain significantly positive (p < 0.01), but their magnitude is smaller than those reported in columns (4) and (5). Overall, our channel tests suggest that EM mediates the relationship between comment letters and earnings volatility, consistent with the notion that comment letters increase earnings volatility through reduced earnings management.

Table 6. Channel test: Accounting channels.

Panel A: The impact of comment letters on earnings management (REM, AEM, and EM) and earnings volatility
Variable	(1)	(2)	(3)	(4)	(5)	(6)	(7)
	REM	AEM	EM	EV1	EV2	EV1	EV2
						CV = EM
CL1	−0.0142	−0.0162**	−0.0253*	0.0212***	0.0094***	0.0190***	0.0087***
	(−1.5394)	(−2.2749)	(−1.9478)	(7.5178)	(7.6376)	(7.0355)	(7.1803)
CV						−0.0156**	−0.0090***
						(−2.4380)	(−3.0976)
Constant	−0.0418	−0.0204	−0.0359	−0.4238***	−0.1351***	−0.3906***	−0.1292**
	(−0.9300)	(−0.7055)	(−0.7750)	(−4.2216)	(−2.7913)	(−3.8049)	(−2.5152)
Control variables	Included	Included	Included	Included	Included	Included	Included
Firm FE	Included	Included	Included	Included	Included	Included	Included
N	5625	4785	5603	7927	7927	7575	7575
R^2-adj	0.0149	0.0936	0.0293	0.0537	0.0430	0.0521	0.0467
Panel B: The impact of different types of comment letters on earnings management (REM, AEM, and EM)
Variable	(1)	(2)	(3)	(4)	(5)	(6)
	REM	AEM	REM	AEM	REM	AEM
FCL	−0.0351**	−0.0270*
	(−2.1461)	(−1.8151)
CCL			−0.0165*	−0.0110
			(−1.6696)	(−1.0807)
MCL					0.0287	−0.0115
					(0.7278)	(−1.2569)
Constant	−0.0263	−0.0395	−0.0283	−0.0161	−0.0356	−0.0087
	(−0.7622)	(−1.0668)	(−0.8230)	(−0.5501)	(−0.6098)	(−0.3202)
Control variables	Included	Included	Included	Included	Included	Included
Firm FE	Included	Included	Included	Included	Included	Included
N	4980	4083	4988	4121	4973	3986
R^2-adj	0.0271	0.0943	0.0355	0.0875	0.0249	0.0538

This table reports the results of accounting channels. Panel A presents the impact of comment letters (CL1) on earnings management (REM, AEM, and EM) and earnings volatility (EV1and EV2), while controlling for firm fixed effects (Cunningham et al. 2020). Panel B reports the impacts of different types of comment letters (FCL, CCL, and MCL) on earnings management. See Appendix A for variable definitions. ***, **, * statistically significant at the 1%, 5%, 10% levels, respectively. The t-statistics presented in parentheses are based on standard errors clustered by firm and robust standard errors.

Note that we find AEM to be lower following receiving a comment letter, which is consistent with Cunningham et al. (2020). However, unlike prior research, we fail to find a ‘switch-up’ from accrual-based earnings management to real-activity earnings management. This is probably because Chinese stock exchange comment letters have a broader coverage that provide supervision not only for financial reporting, but also for day-to-day operations, thus limiting a firm’s ability to replace AEM with REM.

Given the broad range of Chinese comment letters, we are interested in understanding whether the three main types of comment letters – financial reporting, concern, and M&A and restructuring comment letters – have varying impacts on AEM and REM. Financial reporting comment letters oversee annual, semiannual, and quarterly financial reports, whereas concern letters cover a broad range of issues related to firms’ day-to-day management of operational activities. M&A and restructuring comment letters oversee M&A and major asset reorganisation. Accordingly, we construct three indicator variables, FCL, CCL, and MCL, equal to one if firms receive a financial reporting comment letter, concern letter, and an M&A and restructuring comment letter, respectively, and zero otherwise.

Panel B of Table 6 presents the results. The coefficients of FCL in columns (1) and (2) are significantly negative (p < 0.05 or p < 0.10), indicating that receiving financial reporting comment letters not only reduces AEM but also spills over to REM. The coefficient of CCL in column (3) is significantly negative (p < 0.01) but not significant in column (4), suggesting that concern letters help reduce REM but have little impact on AEM. Columns (5) and (6) show that the coefficients of MCL are not statistically significant; thus, M&A and restructuring comment letters have no impact on AEM or REM. Collectively, our results suggest that the impacts of Chinese stock exchange comment letters differ from those of SEC comment letters in the US. Specifically, financial reporting comment letters and concern letters have a significantly negative association with REM, which can be explained by the distinct characteristics of China’s comment letters – a broader scope of reviews. Financial reporting comment letters not only focus on the completeness and faithfulness of financial reporting, but also pay close attention to related party transactions, which is a common method used by Chinese firms for production and expenditure manipulation (REM). Furthermore, concern letters cover a broad range of issues relating to firms’ daily operational activities and thus often require firms to disclose more information on production and expenditures. These unique features of Chinese comment letters appear to effectively curtail real-activity earnings management.

5.6. Textual analysis: comment letter characteristics

In this section, we explore the impacts of the various characteristics of comment letters, as different characteristics may have distinct implications for economic outcomes (Ryans 2021). Based on specific dictionaries consisting of lists of keywords, we use the textual module of Python to extract these keywords from the full text of a comment letter, and record their frequency as the basis for constructing our test variables. Our first set is structure-related characteristics, including the number of dialogues during a year (CL_Dialogue), where one dialogue includes a comment letter and a corresponding response letter; number of questions (CL_Question), measured as the natural logarithm of the number of questions raised in comment letters during a year; accounting professionalism (CL_Pro), measured as the natural logarithm of the number of accounting terms used in comment letters received during a year;¹⁷ and readability (CL_Readability), measured by the number of infrequently used Chinese characters in comment letters received during a year,¹⁸ thus higher values mean more difficult-to-read comment letters.

Next, we examine the topic-related characteristics of comment letters, including R&D and innovation relevance (CL_Innovation), which is measured as an indicator variable equal to one if the proportion of R&D and innovation keywords¹⁹ in comment letters received in a year is greater than the sample average, and zero otherwise; governance relevance (CL_Governance), an indicator equal to one if the proportion of corporate governance keywords²⁰ in comment letters received in a year is greater than the sample average, and zero otherwise; and inquiries for further disclosures, explanations, and corrections in a comment letter (CL_Disclosure), an indicator equal to one if the proportion of disclosure, explanation, and correction keywords²¹ in comment letters received in a year is greater than the sample average, and zero otherwise.

Panel A in Table 7 presents the results of the structure-related characteristics. The coefficients of CL_Dialogue and CL_Question are significantly positive (p < 0.01), whereas the coefficients of CL_Pro are not statistically significant. The coefficient of CL_Readability is significantly positive (p < 0.10). These results suggest that earnings volatility increases with the number of rounds of dialogues and number of questions, as these factors may reflect the severity of the issues revealed in a comment letter. Earnings also become more volatile after a firm receives a less readable comment letter, which is consistent with Kuang et al. (2020), suggesting that a more difficult-to-read regulatory enforcement report concerning a firm’s misreporting is interpreted by outsiders as greater uncertainty.

Table 7. Textual analysis: Comment letter characteristics.

Panel A: Structure-related characteristics
Variable	(1)	(2)	(3)	(4)	(5)	(6)	(7)	（8）
	EV = EV1				EV = EV2
CL_Dialogue	0.0640***				0.0321***
	(4.5585)				(5.2662)
CL_Question		0.0167***				0.0084***
		(4.1197)				(4.7847)
CL_Pro			0.0017				0.0009
			(0.7074)				(0.7988)
CL_Readability				0.0003*				0.0001*
				(1.7761)				(1.6605)
Constant	0.4257***	0.4403***	0.4765***	0.4883***	0.2285***	0.2352***	0.2544***	0.2595***
	(4.2862)	(4.5912)	(4.9962)	(5.2817)	(5.0577)	(5.3505)	(5.8019)	(6.0955)
Control variables	Included	Included	Included	Included	Included	Included	Included	Included
Industry FE	Included	Included	Included	Included	Included	Included	Included	Included
Year FE	Included	Included	Included	Included	Included	Included	Included	Included
N	1947	2069	2026	2068	1947	2069	2026	2068
R^2-adj	0.1126	0.1037	0.0934	0.0936	0.1129	0.1017	0.0884	0.0891

Panel B: Topic-related characteristics
Variable	(1)	(2)	(3)	(4)	(5)	(6)
	EV = EV1			EV = EV2
CL_Innovation	−0.0076			−0.0032
	(−1.3083)			(−1.1662)
CL_Governance		−0.0091*			−0.0044*
		(−1.7932)			(−1.8241)
CL_Disclosure			−0.0102*			−0.0052**
			(−1.8653)			(−2.0063)
Constant	0.4925***	0.4967***	0.4984***	0.2612***	0.2634***	0.2644***
	(5.2847)	(5.3360)	(5.2923)	(6.0897)	(6.1554)	(6.1303)
Control variables	Included	Included	Included	Included	Included	Included
Industry FE	Included	Included	Included	Included	Included	Included
Year FE	Included	Included	Included	Included	Included	Included
N	2069	2069	2069	2069	2069	2069
R^2-adj	0.0930	0.0935	0.0938	0.0886	0.0894	0.0899

This table reports the results of comment letter characteristics. Panel A presents the results of the structure-related characteristics, while Panel B presents the results of the topic-related characteristics. See Appendix A for variable definitions. ***, **, * statistically significant at the 1%, 5%, 10% levels, respectively. The t-statistics presented in parentheses are based on standard errors clustered by firm and robust standard errors.

Panel B of Table 7 presents the results for topic-related characteristics. We find that the coefficients of CL_Governance and CL_Disclsoure are significantly negative (p < 0.05 or p < 0.10), suggesting that comment letters that propel firms to enhance their corporate governance and corporate disclosure are associated with decreased forward earnings volatility. This is probably because such comment letter firms might strive to strengthen their governance structures and improve disclosure quality following such letters, which helps reduce information asymmetry (Bozanic et al. 2017), contributing to decreased earnings volatility.

5.7. Additional analyses

We perform several additional analyses. First, we explore whether the impact of comment letters varies with a firm’s ownership structure. The results show that the effect of comment letters is attenuated for state-owned enterprises (SOEs), suggesting that state ownership may help firms buffer against the adverse influences of comment letters. Second, we find that the impact of comment letters is less potent for firms with more media coverage and internet searches, indicating that an alternative information channel attenuates the effect of comment letters. Furthermore, we investigate whether the receipt of a comment letter affects a firm’s financial constraints through increased earnings volatility; we show that, indeed, earnings volatility is a mediator through which comment letters exacerbate financial constraints. See the Online Appendix for further detail.

6. Conclusion

Distinct from SEC comment letters, Chinese stock exchange comment letters not only provide supervision on financial reporting but also oversee a broad range of operational activities, and thus should have distinct implications for earnings-related issues. However, there is little research examining the impacts of Chinese comment letters on these issues. Our study fills this void by examining whether and how stock exchange comment letters affect earnings volatility. In line with Dichev and Tang’s (2009) framework, we propose that receiving a comment letter increases earnings volatility through both economic and accounting channels. Consistent with our predictions, we find that decrease in supply chain stability and reduced use of earnings management to smooth earnings are two mechanisms through which comment letters predict increased earnings volatility.

The interpretation of our results is contingent upon two significant caveats. First, despite efforts to mitigate endogeneity concerns, establishing a causal relationship using observational data remains a formidable empirical challenge. Second, while we have delved into several economic and accounting channels, these should not be regarded as the exclusive pathways through which comment letters influence earnings volatility. Future research may endeavour to further investigate exogenous variations in the receipt of such letters and their multifaceted effects.

Nevertheless, despite these limitations, our findings carry significant and timely regulatory implications. One of the primary aims of issuing comment letters is to address earnings-related concerns by detecting anomalies and irregularities in a firm’s financial reporting and daily operations, with the ultimate goal of reducing earnings volatility. However, an unintended consequence emerges: the receipt of a comment letter appears to contribute to heightened earnings volatility through both economic and accounting channels. As such, our findings hold valuable insights for regulators, prompting a reevaluation of the impacts of comment letters on earnings-related matters.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Supplemental data

Supplemental data for this article can be accessed online at https://doi.org/10.1080/00014788.2024.2361607.

Correction Statement

This article has been corrected with minor changes. These changes do not impact the academic content of the article.

Additional information

Funding

This work was supported by National Social Science Fund of China [grant number 22BGL090].

Notes

1 According to information on the websites of the Shanghai and Shenzhen Stock Exchanges (the principal regulatory enforcement agency in China’s capital markets), the number of comment letters issued in China increased from 5 in 2014 (the start of information disclosure reform) to more than 1700 in 2018.

2 Our setting differs from SEC comment letters in several notable ways. First, the SEC reviews corporate filings in the US and issues comment letters to detect corporate disclosure problems and law violations (Cassell et al. 2013, Cunningham et al. 2020). In China, the Shanghai and Shenzhen Stock Exchanges fullfill this task. Second, the information advantage of stock exchanges, arising from their roles in managing the platform for corporate disclosure and stock listing, enables them to review more aspects of corporate behaviours in a timelier manner. Third, more importantly, the stock exchange comment letters in China appear to be more diverse in types of comment letters compared to their counterparts in the US. The most frequently issued comment letters in China are concern letters that cover various issues concerning a firm’s day-to-day management of operational activities (see examples of concern letters in Table A1 of the Online Appendix), followed by financial reporting comment letters based on reviews of firms’ quarterly, semi-annual, and annual filings. Overall, these unique characteristics of Chinese stock exchange comment letters render us an appealing setting.

3 For example, Sina Finance reported on 25 November 2022, ‘Guolian’s share price fell by the daily limit and received a comment letter from the Shengzhen Stock Exchange’; Tencent News reported that ‘Due to the high debt acquisition of Kunming Drug, Chinese Resources Sanjiu received a comment letter’ on 1 December 2022; Finance.china.com.cn reprinted the news ‘Focusing on the authenticity of corporate performance, the stock exchange comment letters cast doubt on annual reports’ from the China Securities Journal on 13 April 2022.

4 The stock exchange aims to ensure the authenticity of information disclosure by issuing comment letters. Therefore, the earning-related issues such as earnings manipulation that seriously impairs the quality of financial reporting, will be explicitly requested to be corrected and even subject to subsequent investigation and penalties.

5 Specifically, the general earnings-related issues that are of concern to the stock exchange include the significant changes in operating income and cash flows, reasonableness of operating income changes, accuracy and reasonableness of provision for impairment, accounting for non-operating income, and whether earnings manipulation is engaged.

6 The CSRC is authorized by laws, regulation, and the State Council (the highest administrative authority in China, akin to a cabinet or executive branch in other countries) to supervise and administer the national securities markets. The Shanghai and Shenzhen Stock Exchanges are under the direct control of the CSRC. In practice, the stock exchange is in the front line of information disclosure supervision and frequently issues comment letters, while the CSRC is more concerned about the essence of securities transactions and mainly adopts the method of on-site inspection to implement supervision. It is worth noticing that the stock exchange is not authorized to impose administrative punishment. For firms with severe violations or refusing to respond and revise, the stock exchanges need to report to the CSRC, which is authorized to carry our further administrative investigation and punishment.

7 The legal basis of the stock exchanges review includes ‘Securities Law’, ‘the Standards on Information Disclosure Content and Format of Companies that publicly issue Securities’, ‘Rules on Compilation and Reporting of Information Disclosure of Companies that Publicly issue Securities’, ‘Guidelines on Industrial Information Disclosure of Listed Companies’, and ‘Rules on Stock Listing’.

8 Moreover, we provide a breakdown of the sample distribution by year and industry for both the treatment group (comment letter firms) and the control groups (non-comment letter firms) in Table A2 of the Online Appendix.

9 For example, suppose a company received three comment letters in a year; one of them concerns issue A, while the other two letters concern issue B. In this case, CL1 equals 1, CL2 (unlogged) equals 3, and CL3 (unlogged) equals 2.

10 In a robustness test, we also construct a measure of earnings volatility using quarterly earnings volatility within a year. See Section 5.3.1. for further detail.

11 In Section 5.4.1. we also control for firm fixed effect and find consistent results.

12 For example, earnings volatility (EV1) for comment letter firms is 0.0321 greater than that for non-comment letter firms; the difference is equivalent to a 59% (36%) of the mean (one standard deviation) of EV1.

13 However, this analysis is subject to large data attrition due to the merge with executive compensation data.

14 We also test for the robustness of our results to potential selection on unobservables using the method developed by Oster (2019). We employ the bootstrap method (with 1000 reps) to derive the 95% confidence interval (CI) of the bias-adjusted treatment effects β*. The results show that the CI of β* excludes zero, satisfying the robustness criteria recommended by Oster (2019). Similar to Call et al. (2018), we follow the recommendation by Oster (2019) and set the hypothetical R-squared from the fully specified regression model (R_max) as 1.3 times the R-squared from the original regression model. The calculated δ for our test variables, all greater than 3, exceeds the threshold of 1. Combined, these results suggest that our findings are less likely to be solely driven by unobservable selection.

15 For robustness, we also calculate Herfindahl–Hirschman Index (HHI), which equals the sum of squared market share of each top five suppliers and/or top five customers. Our untabulated results based on HHI generate qualitatively similar inferences.

16 For brevity, we do not tabulate the results of comment letters (CL1) and other mediators (SUC, CUC, SCC, ASUC, ACUC, and ASCC) on earnings volatility (EV), which are largely consistent.

17 Accounting terms are common accounting keywords, including assets, liabilities, equity, revenue, profit, loss, cash, cost, accruals, payable, receivable, deferred, impairment, and their variants.

18 Infrequently used Chinese characters will increase the costs of information-processing for readers (Li, 2008). We obtain 1000 frequently used Chinese characters from the ‘List of Commonly Used Characters in Modern Chinese’ and then use Python to subtract the frequently used characters from all characters to count the number of infrequently used characters in a comment letter.

19 We use the text module of Python to extract R&D and innovation keywords from a comment letter. These keywords include research, development, innovation, patent, high-tech, intellectual property, experiment, and their variants.

20 We extract keywords relating to corporate governance from a comment letter, including corporate governance, internal control, related party transaction, shareholder, CEO, executive compensation, executive shareholding, independent director, board of supervisors, ownership, internal supervision, and their variants.

21 Such keywords include supplementary explanation, further explanation, supplementary disclosure, further disclosure, modification, error, revision, correction, and their variants.

Appendix A

Variable definitions

Table

Variable	Definition
Dependent variables
EV1	The three-year (t to t + 2) rolling standard deviation of annual industry-adjusted return on assets
EV2	The three-year (t to t + 2) rolling standard deviation of quarterly industry-adjusted return on assets
EV3	The three-year (t to t + 2) moving average range (the difference between maximum and minimum values) of annual industry-adjusted return on assets
EV4	The three-year (t to t + 2) average of annual standard deviation of quarterly industry-adjusted return on assets
EV5	The standard deviation of quarterly return on assets in year t
EV6	The three-year average of annual standard deviation of daily stock returns
EV7	The three-year average of annual standard deviation of monthly stock returns
ES	Earnings smoothness, measured as the ratio of the variability of net cash flow from operating activities to the variability of the firm’s net profit
Independent variables
CL1	Equal to one if the firm receives a comment letter in year t, and zero otherwise
CL2	The natural logarithm of one plus the number of comment letters received by a firm in year t
CL3	The natural logarithm of one plus the number of times a firm is inquired for the same concern in year t
Control variables
SIZE	Firm size, the natural logarithm of the total assets at the beginning of the period
AGE	The number of years since the establishment of a firm
GROW	The annual operating income growth rate
BIG4	Equal to one if the audit firm is one of the big four audit firms, and zero otherwise
CASH	Cash holding, the ratio of cash and cash equivalents to the total assets at the beginning of the period
IND	Board independence, the percentage of independent directors on the board of directors
OWN	Equity concentration, the total percentage shareholding of top 10 shareholders
DUA	CEO duality, equal to one if the CEO is also the chairperson, and zero otherwise
PEN	Equal to one if the company has received penalties from Securities Regulatory Commission for the year, and zero otherwise
Channel variables and additional control variables
TC	Trade credit, the sum of accounts payable, notes payable, and advances received scaled by total asset
TCV	Trade credit volatility, the three-year rolling standard deviation of trade credit from t until t + 2
RCV1	The sum of the standard deviation of quarterly operating revenue and the standard deviation of quarterly operating expenses in any given year
RCV2	The sum of the standard deviation of annual operating revenue from t until t + 2 and the standard deviation of annual operating expenses during the same period.
SUC	Supplier concentration, measured as the ratio of annual purchase from top five suppliers to total purchase
CUC	Customer concentration, measured as the ratio of annual sales to top five customers to total sales
SCC	Supply chain concentration, calculated as the average of top five suppliers’ purchase ratio and top five customers’ sales ratio
ASUC	The three-year moving average of supplier concentration (SUC) during the period of t to t + 2
ACUC	The three-year moving average of customer concentration (CUC) during the period of t to t + 2
ASCC	The three-year moving average of supply chain concentration (SCC) during the period of t to t + 2
VSUC	The three-year rolling standard deviation of supplier concentration (SUC) during the period of t to t + 2
VCUC	The three-year rolling standard deviation of customer concentration (CUC) during the period of t to t + 2
VSCC	The three-year rolling standard deviation of supply chain concentration (SCC) during the period of t to t + 2
REM	Earnings management through real activities manipulation, the sum of production manipulation and expenditure manipulation
AEM	Accrual-based earnings management, equal to the estimated residual of modified Jones model
EM	Total earnings management, the sum of REM and AEM
MarketShare	A firm’s sales divided by total sales of its industry at the beginning of year t
Zscore	1.2 × (Working Capital/Assets) + 1.4 × (Retained Earnings/Assets) +0.6 × (EBIT/Assets) + 0.999 × (Market value of equity/book value of total liabilities) + 3.3 × (Operating income/total assets), provided by CSMAR database
INST	Percentage institutional ownership
ETR	Effective tax rate, provided by CSMAR database
AuditTenure	An indicator variable equal to one if the number of years the auditor has audited the client is above the sample median, and zero otherwise
NOA	An indicator variable equal to one if the net operating assets divided by lagged sales is above the median of the corresponding industry-year, and zero otherwise
Cycle	The days receivable plus the days inventory less the days payable
ROA	Return on assets, computed using net income for the rolling four quarters ending with the third quarter of year t, divided by the average of beginning total assets in each of the four quarters
AdjLnAssets	Industry-adjusted log total assets
MtB	Market-to-book ratio
StockIssuance	An indicator variable equal to one if the firm issues equity in the year, and zero otherwise
Earn	The earnings before extraordinary items divided by lagged total assets minus discretionary accruals and production costs, plus discretionary expenditures
Pred_REM	The fitted value from estimating Eq. (12)
Unpred_REM	The residual from estimating Eq. (12).
PCL	An indicator variable equal to one if the firm has previously received the comment letters, and zero otherwise
Instrumental variables
LeaderChange	An indicator variable equal to one if there is a change in the leadership of the stock exchanges or CSRC in the current year, and zero otherwise
Distance	The natural logarithm of the total geographic distance between a firm and regulators (i.e. stock exchanges and CSRC)
CEO power variables
PaySlice	CEO pay slice, measured as the percentage of CEO compensation in the total compensation of the top three executives
PayGap	CEO pay gap, measured as the difference between the CEO’s total compensation and the median total compensation of other executives
PayMix	CEO pay mix, measured by the ratio of long-term variable pay (stock options, restricted stock, and appreciation rights) to total compensation
Tenure	CEO tenure, measured by the number of years a CEO is in office
Comment letter types
FCL	An indicator variable equal to one if firms receive a financial reporting comment letter in year t, and zero if firms receive other types of comment letters.
MCL	An indicator variable equal to one if firms receive an M&A and restructuring comment letter in year t, and zero if firms receive other types of comment letters.
CCL	An indicator variable equal to one if firms receive a concern letter in year t, and zero if firms receive other types of comment letters
Structure-related characteristics of comment letters
CL_Dialogue	The number of dialogues, where one dialogue includes the issuing of one comment letter and one response letter
CL_Question	The natural logarithm of the number of questions raised in comment letters received during a year
CL_Pro	Accounting professionalism, measured by the natural logarithm of the number of accounting terms used in comment letters received during a year
CL_Readability	Readability, measured by the number of infrequently used Chinese characters in comment letters during a year
Topic-related characteristics of comment letters
CL_Innovation	An indicator variable equal to one if the proportion of R&D and innovation keywords in comment letters received in a year is greater than the sample average, and zero otherwise
CL_Governance	An indicator variable equal to one if the proportion of corporate governance keywords in comment letters received in a year is greater than the sample average, and zero otherwise
CL_Disclosure	An indicator variable equal to one if the proportion of disclosure, explanation and error correction keywords in comment letters received in a year is greater than the sample average, and zero otherwise
Other variables
SOE	An indicator variable equal to one if a firm is a state-owned enterprise, and zero otherwise
IC1	The natural logarithm of one plus the number of online news articles of a firm during a year
IC2	The natural logarithm of the annual average of Baidu search index of a firm during a year
CF	The net cash flow from operations scaled by total asset