Big N Auditors and Earnings Response Coefficients A Comparison Study between the US and China *

Transcription

1 DOI /s 年 6 月第 6 卷第 2 期 中国会计与财务研究 C h i n a A c c o u n t i n g a n d F i n a n c e R e v i e w Volume 6, Number 2 June 204 Big N Audors and Earnings Response Coefficients A Comparison Study between the US and China * Jun Du and Gaoguang Zhou Received 5 th of December 203 Accepted 28 th of March 204 The Author(s) 204. This article is published wh open access by The Hong Kong Polytechnic Universy Abstract In the past 20 years, there have been considerable developments and changes in the aud industry in the US and China. Building on the seminal work of Teoh and Wong (993), this essay reviss the tests of the prediction that earnings reports auded by larger aud firms have higher earnings response coefficients (ERCs). Specifically, we investigate whether the posive association between ERCs and perceived aud qualy holds in eher the US or China using long-term historical data. Empirical analysis based on US data from 984 to 202 shows that the ERCs of Big N clients are generally higher than those of non-big N clients, but such differences have dissipated since 2002, the year the Sarbanes-Oxley Act (SOX) came into effect. However, we find weak evidence in China during 995 to 202 that Chinese investors value the credibily of large international audors. The results also reveal that the relationship between Big N and ERCs is stronger in the US than in China. Overall, our results show that audor reputation is valued by investors in more developed markets, indicating that both instutional and industry structure affect the relationship between Big N and ERCs. I. Introduction Audors provide at least two major valuable functions to increase the credibily of financial reporting. The first role of audors is to verify financial reports prepared by firms and disclose the breaches in their clients accounting reports (DeAngelo, 98; Watts and Zimmerman, 98). Through the auding process, audors increase the credabily of financial reports and help reduce information asymmetry (Watts and Zimmerman, 986). The second role is the insurance role. By providing implic insurance protection against investor loss (Watts and Zimmerman, 986), audors are accountable for expressing their opinions. As a result, the assurance service provided by audors gives investors confidence in financial reports. * The authors would like to thank Prof C. S. Agnes Cheng (the reviewer) and Prof Joanna Ho (the discussant) for their helpful comments and suggestions. Jun Du, Assistant Professor, School of Accounting and Finance, The Hong Kong Polytechnic Universy; jun.du@polyu.edu.hk. Gaoguang Zhou, Department of Accounting, Hong Kong Shue Yan Universy; ggzhou@hksyu.edu. Special Issue: Fundamental Analyses of Accounting Information Comparative Studies between the Chinese and US Capal Markets

2 84 Du and Zhou The effectiveness of audors in increasing the credibily of financial reporting varies wh the size of the auding firm. DeAngelo (98) argues that Big N audors, concerned about their valuable reputation capal, are more likely to provide higher aud qualy. As earnings qualy is related to the stock price reaction to earnings, earnings numbers that are certified by high qualy audors (Big N audors) are likely to be more informative. Using a sample from 973 to 988, Teoh and Wong (993) show that the ERCs of Big 8 clients are significantly higher than those of non-big 8 clients. Do large audors play the same role in the China market as their counterparts in the US? Due to the distinct instutional background in China, whether large audors increase financial reporting credibily has become of interest to researchers and regulators. Several studies have shed some light on the joint test of investors response to earnings surprise and perceived aud qualy in the Chinese stock markets. For example, Gul et al. (2003) find that the top 0 audors in China during the period 996 to 997, including one international and nine domestic audors, are perceived to be of higher qualy, which leads to higher earnings response coefficients (ERCs) for their clients earnings surprise. Lin et al. (2009) also find similar evidence that firms auded by the top 0 audors have higher ERCs for unexpected earnings in the period. While prior studies have documented evidence that investors response to earnings information is posively associated wh perceived aud qualy, there has been a growing interest in investigating this issue in a more recent sample period given that the structure and regulations of the aud market have changed in recent years in both developed and emerging markets. In the US context, both the aud market and the regulation environment changed significantly after 2002 (including 2002); these changes are likely to reduce the aud qualy gap between Big N and non-big N audors. First, the size and the client portfolios of second-tier audors grew rapidly after 2002 (GAO, 2006; Rama and Read, 2006; Cassell et al., 203). Wh their greater firm size, these audors have a stronger incentive to protect their reputation capal (DeAngelo, 98). Second, the Public Company Accounting Oversight Board (PCAOB) inspects public company aud firms annually. The inspection results show that the Big N audors and some second-tier audors met the creria during the period from 2004 to 20 (Cassell et al., 203), suggesting that second-tier audors are likely to provide comparable aud qualy to Big N audors. These two factors are likely to contribute toward reducing the financial reporting credibily gap between Big N and non-big N audors. Recent studies (e.g. Boone, Khurana, and Raman, 200; Chang, Cheng, and Reichelt, 200; Cassell et al., 203) provide evidence to show that the qualy (or perceived qualy) difference between Big N and second-tier audors becomes indistinguishable after the Sarbanes-Oxley Act (hereinafter, SOX) came into effect. The development of the aud industry in China was prompted by the re-emergence of the Chinese stock markets in the late 990s. However, for a long time, Chinese audors have been cricised for lacking independence because most of them were sponsored by government agencies until 999, when the Chinese Government launched a disaffiliation programme to cut the official business ties. This reform has helped to enhance aud qualy in respect to the number of modified opinions (DeFond et al., 999). Nonetheless, a wave of aud failures in the early 2000s triggered a public outcry over audor independence in China. To rebuild the market s confidence in the aud industry, the Chinese regulator has imposed a higher ligation and sanction risk on aud firms since 200. Chen et al. (200) find that audors have been more likely to issue modified opinions to important clients since 200, when legal and regulatory instutions

3 Big N Audors and Earnings Response Coefficients 85 became harsher. However, whether audor independence is impaired by polics remains an important concern (e.g. Wang, Wong, and Xia, 2008; Chan, Lin, and Wong, 200). In this study, we follow Teoh and Wong (993) and extend the sample period to 202. Instutional details are omted in the empirical tests, but the major historical trends are incorporated into the research design. Consistent wh Teoh and Wong (993), we find a posive relationship between Big N and ERCs in the US during the period 983 to 202. However, the results are not consistently significant. We then spilt our sample into two periods wh 2002 as the cut-off year as the market and regulation environment changed significantly after the SOX came into effect. We find that the indicator of Big N audors is significantly posively associated wh ERC in the pre-sox era, suggesting that Big N audors provided a better aud service to their clients during the period 983 to 202. The results based on the post-sox sample show that Big N clients financial reports do not exhib significantly higher ERC compared wh non-big N clients financial reports. Our results are consistent wh previous findings (Boone et al., 200; Chang et al., 200; Cassell et al., 203) and suggest that industry structure and regulations are important factors that affect investors perception of aud qualy. Due to the data limation of analyst forecasts, we use real earnings change as a proxy for earnings surprise in the China sample. Using a sample, we fail to find clear evidence that firms auded by international Big 5 audors have higher ERCs than their counterparts auded by local audors. In addion, we find that the market reaction to different perceived aud qualy is stronger in the post-200 period when the Chinese regulator imposed more stringent regulations in response to great expectations for the improvement of aud qualy (Chen et al., 200). However, is worth noting that our results are sensive when considering the issue of audor choice. The overall results from the China market are much weaker than those from the US market; this difference may be explained by data or research design limations as well as by different instutional development levels. Our study provides empirical results on the association between ERCs and perceived aud qualy in the two largest economies from a comparative perspective. The statistical summary is helpful to understanding aud industry structure in both the US and China markets and how aud industry and instutional reforms affect investors perception of aud qualy. Our results are generally consistent wh recent studies (Boone et al., 200; Chang et al., 200; Cassell et al., 203) and support the notion that aud market competion and regulation environment are two important forces that affect perceived aud qualy. The comparison results are also likely to provide addional evidence to support the argument that instutional factors are one of the important factors that affect perceived aud qualy internationally (Khurana and Raman, 2004; Francis and Wang, 2008). The rest of the paper is organised as follows: Section II describes the data and the samples; Section III presents the research methodology; Section IV discusses the empirical results; and Section V concludes the article. II. Research Design 2. Model specification Teoh and Wong s (993) paper was the first to investigate the effect of Big N audors on ERCs. They argue that because investors cannot assess firms earnings qualy directly, they are likely to assess by inferring from the qualy (skills) of the audor. To

4 86 Du and Zhou examine the effect of Big N audors on ERCs, they extend previous studies (Kormendi and Lipe, 987; Easton and Zmijewski, 989, among others) by adding the Big N audor identy into the ERC model, in which larger coefficients of earnings suggest greater market response and hence better earnings qualy. They also consider other firm characteristics that have been documented as affecting ERCs. Previous studies (Kormendi and Lipe, 987; Easton and Zmijewski, 989) have suggested that growth opportuny measured by the market-to-book (MB) value is posively linked to future earnings and higher MB is associated wh greater ERCs (Collins and Kothari, 989). Teoh and Wong (993) use market beta as a proxy for firm risk, which is predicted to be negatively associated wh ERCs. They also consider firm size proxied by the natural logarhm of firm s market value and information environment proxied by the number of analysts forecasts in their empirical model. They show that the ERCs of the Big 8 clients are significantly higher than those of non-big 8 clients during the period , implying that investors perceive the earnings qualy of firms auded by Big 8 audors to be higher. For comparison purposes, we follow Teoh and Wong (993) and use the following specification to examine the relationship between Big N audors and ERCs in the US and China samples: CAR UE 0 UE 4 UE 6 UE 8 UE 0 BIGN MB BETA LMV N UE UE UE fixed year / industry effect UE UE 3 LMV N BIGN MB BIGN BIGN BIGN BETA BIGN () Following Teoh and Wong (993), we use two proxies to measure investors reaction to earnings surprise. We first measure the daily abnormal return as the raw return R minus the market return of the same day R mt, which is proxied by the CRSP equally weighted stock index for the US sample and the CSMAR (China Stock Market Accounting Research) equally weighted return for the China sample. For the US sample, the measure of CAR is calculated as the 300-day accumulated abnormal returns from the T announcement date: CAR log( R Rmt ). We discard firms wh less than 00 t trading days. For the China sample, the first measure of market reaction is defined as the 3-day 2 cumulative abnormal returns around earnings announcement dates: T CAR log( R Rmt ). Alternatively, we use the market model to measure the t 2 In the main tests, we use short-window abnormal returns to measure market response to earnings announcements, following Gul et al. (2003) and Lin et al. (2009). In untabulated tests, we also try the 300-day accumulated abnormal return from the earnings announcement date, but we fail to find any evidence that the ERCs of Big 5 audors are higher than those of their non-big 5 counterparts. One possible explanation is the UE information may be pre-empted due to the management earnings forecasts mandated by the listing rules of the Shanghai/Shenzhen exchanges, which require that listed companies should disclose management earnings forecasts when earnings eher increase/decrease by 50% or change signs (Huang et al., 203).

5 Big N Audors and Earnings Response Coefficients 87 daily abnormal returns. The model is estimated over the period from 256 to 7 days before the earnings announcement date, and a minimum estimation period of 00 days is required. The second measure of market reaction is defined as T CAR 2 log( R ˆ i ˆ i Rmt ). t To be consistent wh Teoh and Wong (993), our first earnings surprise measure (UE) is the difference between actual earnings and forecast consensus in year t scaled by stock price in the US sample. As analyst forecast data are not available in China until 2005 and the coverage is relatively small, we also construct a second earnings surprise measure (UE2) to proxy for earnings surprise, especially in the China sample. UE2 is calculated as the difference between earnings in year t and year t- scaled by stock price on the day prior to the earnings announcement dates, or the fiscal year-end stock price if the former is not available. BIGN is a dummy variable which equals for Big N audors and 0 otherwise. In the China context, we use Big 5 to identify international accounting firms, including Arthur Andersen, PricewaterhouseCoopers, Delote Touche Tohmatsu, Ernst & Young, and KPMG. Following Gul et al. (2003) and Lin et al. (2009), we also use the indicator Top 0 to identify the 0 largest audors each year in terms of total assets auded and DBIG5 to identify the five largest domestic Chinese audors each year in terms of total assets auded. MB is the market-to-book ratio, a proxy for growth and persistence, which is equal to market value of equy divided by book value of equy. BETA is the slope coefficient derived from the market return model to proxy for firm risk. LMV is the natural log of the market equy value of a firm. /N is equal to divided by the number of analyst forecasts in year t. We do not include /N in the ERC regressions based on the China sample due to the limation of analysts forecasts data. Our variable of interest is α 3. A significantly posive α 3 suggests a posive relationship between Big N and ERC. III. Sample and Data For the US sample, we extract analyst forecasting data from IBES and accounting data from COMPUSTAT. Stock return data are sourced from the merged CRSP. Due to data availabily, 3 we restrict our sample period to 983 to 202. Following Teoh and Wong (993), we remove firm-years wh less than 00 trading days return data and observations in four-dig SIC industries where Big N audors have 00% of the market share. We further exclude observations wh missing variables shown in our empirical models. Finally, we winsorise the top and bottom % of continuous variables to migate the outlier effect. Through these procedures, we obtain 38,206 observations. The data for the China sample consist of a dataset over the period 995 to 202 from the CSMAR database. The sample starts in 995, when the Big 5 international audors can be identified in the database. We require that each observation have the necessary CSMAR data on market capalisation and financial statements and the necessary daily price data to compute cumulative abnormal returns. To be consistent wh the lerature, we also exclude firms operating in the finance industry. There are 20,662 firm observations that meet the selection creria. We present the descriptive statistics in Table. As the result in Panel A of Table shows, Big N market share is around 80% in our US sample, suggesting a dominant market posion in the US market. The market share of Big N declines after 2002 because 3 IBES do not provide sufficient data for our analyses before 983. As a result, we start our analyses from 983.

6 88 Du and Zhou of the collapse of Arthur Andersen. During the sample period, the percentage of Big N clients total assets is larger than that of market share, suggesting that Big N clients are larger than non-big N clients. These figures together indicate that Big N audors dominate in the US market. Panel B of Table presents the yearly distribution of the China sample firms and their choice of audors. The market share of Big 5 audors in terms of number of clients varies across years, ranging from 3.9% in 998 to 9.39% in In total, Big 5 audors aud 6.22% of listed firms in China, on average accounting for 43.25% of the total assets in the market, which indicates that the market share of Big 5 audors in China is substantially lower than is in the US. Table Yearly distribution of clients Panel A: The US sample No. of clients Total client assets ($million) Non-Big N Big N % Big N Non-Big N Big N % Big N client Year clients assets % 6,9 5, % % 6,925 33, % % 20,086 62, % % 22,443 57, % % 2, , % % 6,786 37, % % 4,737 43, % % 6, , % %,085 32, % % 06, , % % 0, , % % 33,428,243, % % 43,946,556, % % 34,452,0, % % 34,03,690, % % 53,245,73, % % 62,588 2,79, % % 65,730 2,025, % % 56,362,795, % % 94,755 2,727, % % 74,960,952, % % 79,46 2,786, % % 336,548 7,352, % % 458,65 9,227, % % 508,34,762, % % 567,56 0,635, % % 638,857,978, % % 599,854,30, % % 69,25 3,293, % % 744,926 4,069, % Total % 5,498,045 3,595, %

7 Big N Audors and Earnings Response Coefficients 89 Panel B: The China sample No. of clients Total client assets (RMB billion) Year Non-Big N Big N Big N share (%) Non-Big N Big N Big N share (%) % % % % % % %, % %, % %, % 2002, %, % 2003, %,958, % 2004, % 2,399, % 2005, % 2,823, % 2006, % 2,990, % 2007, % 3,368 2, % 2008, % 4,539 4, % 2009, % 5,49 5, % 200, % 7,472 7, % 20, % 9,654 8, % 202 2, % 2,66 0, % Total 9,377, % 60,325 45, % The table presents the yearly distribution of Big N (5) /non-big N (5) audors market share and the total assets of their clients in the US (China) market. Panel A displays the yearly distribution in the US market; the sample period is from 983 to 202. In Panel B, we report the yearly distribution in China during the sample period 995 to 202. In each year, we report the number of clients, market share (in percentage form), aggregated clients assets (million), and percentage of clients assets of each type of audors. Table 2 reports the descriptive statistics of the cumulative abnormal returns and the control variables used in the ERC regressions partioned on the basis of audor choice. Panel A of Table 2 presents the descriptive statistics based on the US sample. In general, shows that Big N clients have better performance (higher UE), larger size (larger LMV), lower growth (lower MB), lower risk (lower BETA), and more analyst forecasts (lower /N). Panel B of Table 2 presents the China sample s descriptive statistics. As shown, there are considerable variations in these variables between the Big 5 client firms and the Non-Big 5 client firms. Notable differences are that the Big 5 clients are, on average, larger and less risky and display smaller 3-day cumulative abnormal returns. The results in Table 2 also reveal that the characteristics of Big N clients differ in the US and China samples. The pairwise Pearson correlation coefficients among the ERC regression variables are reported in Table 3. Consistent wh prior studies, the correlation coefficients of most of the control variables are significant and in the expected direction. Panel B of Table 3 presents the correlation matrix for the China sample. The two measures of market reaction to earnings surprise are posively and significantly correlated (0.982). They are also posively correlated wh the indicator of Big 5 audors. Specifically, the correlation between CAR (CAR2) and BIG5 is 0.08 (0.07), providing evidence that investors reaction to earnings news is affected by audor choice. We also note that the correlations among several variables are greater than 0.970, particularly the correlation between two-variable and three-variable interactions. For example, UE_BIG5 is highly correlated wh UE_BETA_BIG5 (0.979) and UE_LMV_BIG5 (0.999), indicating that multicollineary could be a problem in multivariate analyses.

8 90 Du and Zhou Table 2 Descriptive statistics of variables Panel A: The US market Big N Variables N Min P P25 Median P75 P99 Max Mean Mean difference Median difference (T-stat) (Z-stat) 0 CAR 7, , (-5.20***) (-5.7***) 0 UE 7, , (-3.65***) (-0.9) 0 MB 7, , (-2.54***) (-23.7***) 0 LMV 7, , (-60.67***) (-59.64***) 0 BETA 7, , (-27.00***) (-26.***) 0 /N 7, , (60.84***) (59.34***) Panel B: The China market Big N Variables N Min P P25 Median P75 P99 Max Mean Mean difference (T-stat) Median difference (Z-stat) 0 CAR (-.77*) (-3.4***) 0 CAR (-.667*) (-3.24***) 0 UE (-0.76) (-3.932***) 0 MB (-3.583***) (2.762***) 0 BETA (8.25***) (9.949***) 0 LMV ( ***) ( ***) The table presents the descriptive statistics of variables for the Big N and non-big N samples in the US and China markets. For each variable, we report descriptive statistics for the Big N and non-big N samples. We also report the difference of each variable between these two groups using T-stat (variable mean difference) and Z-stat (variable median difference). In the US market, CAR is the cumulative abnormal return, calculated as continuously compounded daily abnormal return. In the China market, CAR is the 3-day cumulative abnormal returns around earnings announcement dates. CAR2 is the 3-day cumulative abnormal returns using the market model. In the US market, UE is the difference between actual earnings and forecast consensus in year t scaled by stock price, while UE in the China market sample is calculated as the difference between earnings in year t and year t- scaled by book value of equy. Big N is a dummy variable which equals for Big N audors and 0 otherwise. BETA is the slope derived from the market return model. LMV is the log value of the market equy value of a firm. MB is the market-to-book ratio, which is equal to market value of equy divided by book value of equy. /N is equal to divided by the number of analysts forecasts in year t. *, **, and *** indicate the 0%, 5%, and % significance level respectively.

9 Big N Audors and Earnings Response Coefficients 9 Table 3 Correlation matrix Panel A: The US sample CAR UE BIGN BIGNUE MBUE MBBIGNUE BETAUE BETABIGNUE LMVUE LMVBIGNUE /NUE UE BIGN BIGNUE MBUE MBBIGNUE BETAUE BETABIGNUE LMVUE LMVBIGNUE /NUE /NBIGNUE Panel B: China sample CAR CAR2 BIG5 UE UE_BIG5 UE_MB UE_MB_BIG5 UE_BETA UE_BETA_BIG5 UE_LMV UE_LMV_BIG5 CAR.000 CAR BIG UE UE_BIG UE_MB UE_MB_BIG UE_BETA UE_BETA_BIG UE_LMV UE_LMV_BIG This table reports the correlation among variables in the US and China market samples. Bold text indicates 5% significance or better.

10 92 Du and Zhou IV. Empirical Results 4. The US sample The main results of the US sample are reported in Table 4. Column in Table 4 shows that the coefficients of BIGN*UE are wh less than 5% significance level, suggesting that Big N clients are associated wh higher ERC. The coefficient in Column 2 in Table 2 is still posive but not significant. These results together show that during , investors perceive financial reports released by Big N clients as more credible compared wh those released by non-big N clients. The results are generally consistent wh the findings in Teoh and Wong (993). In Table 5, we present the partion results. The results show that the significantly posive relationship between Big N and ERC only exists in the pre-sox era. These results are consistent wh those in prior studies (e.g. Boone et al., 200; Cassell et al., 203; Chang et al., 200). Table 4 The regression results of the US sample Variables UE=UE UE=UE2 Coefficient T-value Coefficient T-value UE 2.733*** *** 2.86 BIGN 0.0** *** 4.46 BIGN*UE 0.573** UE*MB *** 5.3 UE*MB*BIGN *** *** UE*BETA UE*BETA*BIGN -0.37** UE*LMV -0.94*** *** 4.52 UE*LMV*BIGN *** UE*/N *** UE*/N*BIGN *** Years Adj R-square Observations The table reports the regression results concerning the relationship between Big N and ERC. In column (), we report the results using UE as the proxy for earnings surprise. In column (2), we present the results using UE2 as the earnings surprise proxy. CAR is the cumulative abnormal return for firm i, which is calculated as the continuously compounded daily abnormal return in year t. In this study, our daily abnormal return is equal to the difference between stock return and the CRSP equally weighted stock index. Our principal earnings surprise measure (UE) is the difference between actual earnings and forecast consensus in year t scaled by stock price. The second earnings surprise measure (UE2) is calculated as the difference between earnings in year t and year t- scaled by book value of equy. Big N is a dummy variable which equals for Big N audors and 0 otherwise. BETA is the slope derived from the market return model. LMV is the log value of the market equy value of a firm. MB is the market-to-book ratio, which is equal to market value of equy divided by book value of equy. /N is equal to divided by the number of analysts forecasts in year t. *, **, and *** indicate the 0%, 5%, and % significance level respectively. The selection bias problem in aud research has attracted increasing concerns in recent years (Lennox et al., 202). 4 To address the potential self-selection concern, we 4 Although there is a debate on the comparative mers of Heckman (976) and the matching method (Lennox et al., 202), this discussion is outside the scope of this study. Teoh and Wong (993) use the matching sample approach. In our study, however, we follow recent studies in the area of audor choice lerature (for details, see Lennox et al., 202) and the Heckman two-stage approach.

11 Big N Audors and Earnings Response Coefficients 93 Table 5 The effect of Big N on ERC in the US market -partion analyses Panel A: The effect of Big N on ERC before SOX Variables UE=UE UE= UE2 Coefficient T-value Coefficient T-value UE.645*** BIGN BIGN*UE.56** *** 2.6 UE*MB 0.054*** ** 2.46 UE*MB*BIGN -0.35*** * -.87 UE*BETA *** 3.0 UE*BETA*BIGN *** UE*LMV *** 5.85 UE*LMV*BIGN *** -4.3 UE*/N ** UE*/N*BIGN ** Years Adj R-square Observations Panel B: The effect of Big N on ERC after SOX Variables UE=UE UE=UE2 Coefficient T-value Coefficient T-value UE 2.85*** *** 3.39 BIGN 0.034*** *** 8.62 BIGN*UE UE*MB * *.90 UE*MB*BIGN 0.044** ** UE*BETA UE*BETA*BIGN -0.62* UE*LMV *** *** 5.22 UE*LMV*BIGN 0.09** UE*/N ** ** UE*/N*BIGN Years Adj R-square Observations The table reports the regression results concerning the relationship between Big N and ERC in the preand post-sox eras. We report the regression results in the pre-sox era in Panel A and the results in the post-sox era in Panel B. In column (), we report the results using UE as the proxy for earnings surprise. In column (2), we present the results using UE2 as the earnings surprise proxy. CAR is the cumulative abnormal return for firm i, which is calculated as the continuously compounded daily abnormal return in year t. In this study, our daily abnormal return is equal to the difference between stock return and the CRSP equally weighted stock index. Our principal earnings surprise measure (UE) is the difference between actual earnings and forecast consensus in year t scaled by stock price. The second earnings surprise measure (UE2) is calculated as the difference between earnings in year t and year t- scaled by book value of equy. Big N is a dummy variable which equals for Big N audors and 0 otherwise. BETA is the slope derived from the market return model. LMV is the log value of the market equy value of a firm. MB is the market-to-book ratio, which is equal to market value of equy divided by book value of equy. /N is equal to divided by the number of analysts forecasts in year t. *, **, and *** indicate the 0%, 5%, and % significance level respectively. use Heckman s two-stage procedure to estimate the coefficients (Heckman, 976). The first stage obtains an inverse Mills ratio (IMR) from the logistic regression on the

12 94 Du and Zhou determinants of audor choice in Wang et al. (2008). Specifically, the first-stage regression model is specified as follows: BIGN SIZE 0 LEVERAGE 4 REC 7 GROWTH NYSEAMX 8 2 CACL 5 ROA INVENTORY 3 6 (2) To conserve space, Table 6 only reports the results of the second-stage regression. Untabulated results of the first-stage regression show that firms wh larger size, greater growth, lower prof, healthy financial status (larger current ratio), higher receivable level, and lower inventory are more likely to hire Big N audors. In addion, firms that are incorporated in major exchanges such as NYSE and AMEX are more likely to choose high qualy audors. We use the IMR to adjust the self-selection problem. The IMR derived from the first-stage regression is used in the second-stage analysis. Table 6 consistently shows that Big N clients exhib higher, but not statiscally significant, ERC. The results suggest that the main results in Table 4 are likely to be affected by the selection problem. The results are consistent wh recent concerns regarding the selection problem in aud research (Lennox et al., 202). Table 6 Sensivy test of the US sample two-stage analyses Variables UE=UE UE= UE2 Coefficient T-value Coefficient T-value UE 2.803*** *** 2.70 BIGN BIGNUE UE*MB *** 3.63 UE*MB*BIGN *** *** UE*BETA *** 2.67 UE*BETA*BIGN * -.64 UE*LMV *** 2.90 UE*LMV*BIGN ** UE*/N -.247*** * -.9 UE*/N*BIGN IMR *** *** -2.3 Years Adj R Observations In this table, we address the selection problem by using Heckman s two-stage least square approach. SIZE is firm size measured as the natural log of total assets. GROWTH is the annual growth of sales revenues. ROA is net income scaled by total assets. LEVERAGE is total liabilies divided by total assets. CACL is current assets divided by current liabilies. REC is receivables scaled by total assets. INVENTORY is year-end inventory divided by total assets. NYSEAMX is an indicator equal to if the client firm is incorporated in the NYSE or AMEX and 0 otherwise. We use IMR (inverse Mills ratio) to adjust the self-selection issue in examining the effect of Big N on ERC. 4.2 The China sample Panel A of Table 7 presents the estimates from the OLS regression of equation () for the China sample. All of the continuous variables used in the regressions are

13 Big N Audors and Earnings Response Coefficients 95 winsorised at the 5% level due to some outliers. To measure market reaction to earnings surprise, models to 3 use CAR and models 4 to 5 use CAR2. Estimates from models and 4 using the full sample and models 3 and 6 using the sample suggest that Big 5 clients have greater market reactions. However, as shown in models to 6, there is no statistically significant relationship between the CARs and earnings surprise UE nor between the CARs and the interaction term of UE and Big5. The results are inconsistent wh our expectations and wh prior studies (Gul et al., 2003; Lin et al., 2009) that find higher ERCs for firms auded by the 0 largest audors in China in relatively short and earlier sample periods. One possible explanation for this is that the long sample period may incorporate many instutional reforms related to the aud market in China which may bring some noise to the empirical tests. Another explanation is that Big 5 aud firms are not perceived by Chinese investors as better qualy audors. For example, Liu et al. (2004) find that the clients of the Big 5 audors tend to be less conservative compared wh those auded by non-big 5 audors in China. Table 7 The regression results of the China sample Panel A: Regressions in the pre- and post-200 periods () (2) (3) (4) (5) (6) VARIABLES CAR CAR CAR CAR2 CAR2 CAR2 Full sample Full sample Constant -0.03*** *** -0.00*** -0.02** *** (-6.22) (-3.26) (-3.24) (-2.40) (-0.9) (-3.06) BIG ** ** 0.003** ** (2.22) (0.72) (2.25) (2.28) (0.62) (2.36) UE (-0.38) (-0.8) (-.49) (0.82) (-0.0) (-0.6) UE_BIG (.0) (.) (0.4) (0.42) (.9) (-0.30) UE_MB (0.95) (-0.2) (0.63) (0.8) (-0.3) (0.47) UE_MB_BIG (-0.67) (0.03) (-0.74) (-0.86) (-0.02) (-0.92) UE_BETA * (-0.0) (.62) (.35) (0.4) (.80) (.22) UE_BETA_BIG (-0.77) (0.66) (-.05) (-0.57) (0.64) (-0.79) UE_LMV (0.56) (0.3) (.32) (-0.77) (0.06) (-0.0) UE_LMV_BIG (-0.70) (-.40) (0.03) (-0.0) (-.45) (0.78) Year effect Yes Yes Yes Yes Yes Yes Industry effect Yes Yes Yes Yes Yes Yes Observations 20,662 3,536 6,082 20,662 3,536 6,082 R-squared Adj. R-squared The table reports the regression results concerning the relationship between Big N and ERC in the China market. CAR is the 3-day cumulative abnormal returns around earnings announcement dates. CAR2 is the 3-day cumulative abnormal returns using the market model. BIG5 is a dummy variable which equals if the audor is one of top 5 audors in China and 0 otherwise. BETA is the slope derived from the market return model. LMV is the log value of the market equy value of a firm. MB is the market-to-book ratio, which is equal to market value of equy divided by book value of equy. Robust t-statistics are reported in parentheses. *, **, and *** indicate the 0%, 5%, and % significance level respectively.

14 96 Du and Zhou Panel B: Regressions whout three-term interactions () (2) (3) (4) (5) (6) VARIABLES CAR CAR CAR CAR CAR CAR Full full Constant *** *** -0.00*** -0.02** *** (-6.22) (-3.24) (-3.35) (-2.39) (-0.88) (-3.7) D 0.003** ** 0.003** ** (2.32) (0.7) (2.37) (2.36) (0.6) (2.49) UE * (-0.8) (0.28) (-.73) (0.95) (0.47) (-0.47) UE_D 0.072* ** * (.72) (0.6) (.38) (2.0) (0.42) (.75) UE_MB (0.75) (-0.05) (0.40) (0.56) (-0.07) (0.2) UE_BETA * ** (-0.8) (.96) (.08) (-0.04) (2.4) (0.96) UE_LMV * (0.44) (-0.9) (.65) (-0.85) (-0.47) (0.28) Year effect Yes Yes Yes Yes Yes Yes Industry effect Yes Yes Yes Yes Yes Yes Observations 20,662 3,536 7,26 20,662 3,536 7,26 R-squared Adj. R-squared The table reports the regression results concerning the relationship between Big N and ERC in the China market. The results are based on a modified model in which three-term interactions are excluded. CAR is the 3-day cumulative abnormal returns around earnings announcement dates. CAR2 is the 3-day cumulative abnormal returns using the market model. D is a dummy variable which equals if the audor is one of top 5 audors in China and 0 otherwise. BETA is the slope derived from the market return model. LMV is the log value of market equy value of a firm. MB is the market-to-book ratio, which is equal to market value of equy divided by book value of equy. Robust t-statistics are reported in parentheses. *, **, and *** indicate the 0%, 5%, and % significance level respectively. However, untabulated results show that the variance inflation factors (VIF) from the two main regressions are que high: the VIF values for some continuous variables are over 500, an indication that there is multicollineary in the main regressions. As Panel B of Table 3 reveals that the correlation coefficients among some variables in equation () are near to, particularly those between the two-term and three-term interactions, we turn to a modified model by excluding all the three-term interactions in the original regressions of Teoh and Wong (993). 5 Panel B of Table 7 compares the results using the model whout the three-term interaction variables. The results indicate that using the modified model, the estimated coefficients of UE_Big5 become significant in both the full sample period and the post-200 period. In the above tests, the indicator of Big 5 audors is the proxy for perceived aud qualy in the China sample. To check the robustness of our results, we also repeat the ERC analyses using alternative proxies for perceived aud qualy. Gul et al. (2003) and Lin et al. (2009) find that the client firms of the top 0 audors in China have higher ERCs in the and periods, respectively. Using the indicator of the top 0 large audors based on total assets auded, we fail to find any significant 5 Multicollineary could be more of a concern in the China sample due to the fact that Big 5 audors dominate the market share of large companies: the 6.22% of Big 5 clients on average account for 43.25% of total client assets during the period 995 to 202 (see Panel B of Table ).

15 Big N Audors and Earnings Response Coefficients 97 association between ERCs and perceived aud qualy in Panel A of Table 8. In addion, we also examine the difference in perceived aud qualy between the international Big 5 audors and the five largest domestic audors (DBig5). The results in Panel B of Table 8 indicate that the Big 5 clients have higher ERCs than the clients of local large audors. Table 8 Sensivy test of the China sample alternative Big N indicators Panel A: Regressions using Top 0 audors () (2) (3) (4) (5) (6) (7) (8) VARIABLES CAR CAR CAR CAR CAR2 CAR2 CAR2 CAR2 full full full full Constant -0.03*** -0.03*** *** -0.00*** -0.02** -0.02** *** (-6.24) (-6.24) (-3.23) (-3.37) (-2.44) (-2.43) (-0.88) (-3.9) D (0.96) (.0) (0.0) (.08) (.22) (.36) (0.23) (.32) UE ** (0.04) (-0.5) (0.32) (-2.7) (.8) (0.55) (0.48) (-0.95) UE_D (-0.5) (0.97) (0.38) (0.6) (-.08) (.) (0.56) (0.32) UE_MB (0.66) (0.53) (-0.00) (0.5) (0.50) (0.28) (-0.05) (-0.08) UE_BETA ** ** (0.32) (-0.29) (.99) (0.93) (0.43) (-0.8) (2.9) (0.78) UE_LMV ** (0.05) (0.82) (-0.25) (2.6) (-.9) (-0.40) (-0.50) (0.84) UE_MB_D (-0.8) (-0.20) UE_BETA_D (-0.99) (-0.94) UE_LMV_D (0.87) (.45) Year effect Yes Yes Yes Yes Yes Yes Yes Yes Industry effect Yes Yes Yes Yes Yes Yes Yes Yes Observations 20,662 20,662 3,536 7,26 20,662 20,662 3,536 7,26 R-squared Adj. R-squared Panel B: Regressions using Big 5 and local Big 5 indicators () (2) (3) (4) () (2) (3) (4) VARIABLES CAR CAR CAR CAR CAR2 CAR2 CAR2 CAR2 full full full full Constant -0.03*** *** *** -0.00*** -0.02** -0.02** *** (-6.23) (-6.22) (-3.25) (-3.34) (-2.42) (-2.40) (-0.90) (-3.7) D 0.003** 0.003** ** 0.003** 0.003** ** (2.9) (2.30) (0.74) (2.32) (2.27) (2.37) (0.64) (2.46) D (0.0) (0.09) (0.28) (-0.5) (0.23) (0.30) (0.38) (0.05) UE * (0.08) (-0.6) (0.28) (-.77) (.26) (0.97) (0.48) (-0.50)

16 98 Du and Zhou UE_D * ** * (0.8) (.72) (0.7) (.28) (0.8) (2.0) (0.44) (.66) UE_D (-0.93) (0.20) (0.08) (-0.35) (-.24) (0.24) (0.6) (-0.29) UE_MB (0.6) (0.76) (-0.04) (0.38) (0.42) (0.57) (-0.04) (0.20) UE_BETA * ** (0.22) (-0.7) (.94) (.08) (0.30) (-0.03) (2.3) (0.96) UE_LMV * (0.04) (0.4) (-0.9) (.70) (-.24) (-0.88) (-0.48) (0.32) UE_MB_D (-0.62) (-0.79) UE_MB_D (0.75) (0.96) UE_BETA_D (-0.84) (-0.62) UE_BETA_D (-0.5) (-0.42) UE_LMV_D (-0.47) (0.7) UE_LMV_D (.05) (.34) Year effect Yes Yes Yes Yes Yes Yes Yes Yes Industry effect Yes Yes Yes Yes Yes Yes Yes Yes Observations 20,662 20,662 3,536 7,26 20,662 20,662 3,536 7,26 Adj. R-squared The table reports the regression results concerning the relationship between Big N and ERC in the China market. In this table, we define D (2) as a dummy variable which equals if an audor is ranked in the top 0 (5) and 0 otherwise. CAR is the 3-day cumulative abnormal returns around earnings announcement dates. CAR2 is the 3-day cumulative abnormal returns using the market model. BETA is the slope derived from the market return model. LMV is the log value of the market equy value of a firm. MB is the market-to-book ratio, which is equal to market value of equy divided by book value of equy. Robust t-statistics are reported in parentheses. *, **, and *** indicate the 0%, 5%, and % significance level respectively. To migate the self-selection concern, we use Heckman s two-stage procedure to estimate the coefficients (Heckman, 976). The first stage obtains an IMR following the logistic regression on the determinants of audor choice in Wang et al. (2008). Specifically, the first-stage regression model is specified as BIG5 0 SIZE 2 GROWTH 3 ROA 4 LEVERAGE 5 CACL 6 OTHREC 7 INVENTORY 8 SOE LARGEHOLDING 9, (3) where SIZE is the natural log of total assets, GROWTH is the annual growth of sales revenues, ROA is net income scaled by total assets, LEVERAGE is total liabilies divided by total assets, and CACL is current assets divided by current liabilies.

17 Big N Audors and Earnings Response Coefficients 99 OTHREC is other receivables scaled by total assets. INVENTORY is year-end inventory divided by total assets. SOE is an indicator which equals if the client firm is a state-owned enterprise. LARGEHOLDING is the ownership held by the largest shareholder. We then add the IMR from the first-stage regression as an addional regressor in the second stage to repeat the major ERC analyses. To conserve space, Table 9 only presents the results from the second-stage regression models. The estimated coefficients of the ERCs are not significant, indicating that we should interpret the results reported in Panel B of Table 5 wh caution. Table 9 Sensivy test of the China sample two-stage analyses () () (3) (4) VARIABLES CAR CAR CAR2 CAR2 Constant (0.27) (0.27) (0.62) (0.62) BIG (-0.94) (-0.75) (-0.74) (-0.59) UE (-0.55) (-0.72) (0.45) (0.24) UE_BIG (0.30) (.30) (-0.29) (.53) UE_MB (0.68) (0.24) (0.66) (0.20) UE_BETA (.4) (0.69) (.04) (0.6) UE_LMV (0.37) (0.66) (-0.68) (-0.37) UE_MB_BIG (-.04) (-.6) UE_BETA_BIG (-.42) (-.7) UE_LMV_BIG (0.29) (0.88) IMR *** *** (-0.94) (-4.) (-0.74) (-4.4) Year effect Yes Yes Yes Yes Industry effect Yes Yes Yes Yes Observations 0,39 0,39 0,39 0,39 R-squared Adj. R-squared In this table, we address the selection problem by using Heckman s two-stage least square approach. CAR is the 3-day cumulative abnormal returns around earnings announcement dates. CAR2 is the 3-day cumulative abnormal returns using the market model. SIZE is the natural log of total assets. GROWTH is the annual growth of sales revenues. ROA is net income scaled by total assets. LEVERAGE is total liabilies divided by total assets. CACL is current assets divided by current liabilies. OTHREC is other receivables scaled by total assets. INVENTORY is year-end inventory divided by total assets. SOE is an indicator which equals if the client firm is a state-owned enterprise. LARGEHOLDING is the ownership held by the largest shareholder. The IMR value derived from a first-stage regression is used to address the selection problem in the second stage.

18 200 Du and Zhou V. Conclusion In this paper, we follow Teoh and Wong (993) to investigate the market reaction to the perceived aud qualy of financial statements using long historical data in the US and China. On the basis of a US sample covering the period , we make some interesting findings in these two markets. First, we find that aud markets in the US and China do change over time and that the market structure is different in these two markets. The market share of large audors in the US is dominant compared wh that in China. Second, in the US market, we find evidence consistent wh Teoh and Wong s (993) argument that investors perceive the financial statements auded by Big N audors to be of higher qualy than those auded by Non-Big N audors. The qualy differentiation however becomes insignificant in the post-sox period. Third, we test a modified Teoh and Wong (993) model in China and find weak evidence that the ERCs of Big 5 clients are higher than those of Non-Big 5 clients. However, when multicollineary is of less concern, the model performs favourably compared alongside the original model based on the China sample during the period 995 to 202. Taken together, our study suggests that the findings in Teoh and Wong (993) hold but are likely to change due to aud industry and regulation changes. By comparing the results from these two markets, we also find that the results from the China market are weaker than those from the US market. This difference may be explained by differences in market developments and research design. Open Access. This article is distributed under the terms of the Creative Commons Attribution License which perms any use, distribution, and reproduction in any medium, provided the original author(s) and the source are creded. References Boone, J. P., Khurana, I. K., and Raman, K. K. (200), Do the Big 4 and the Second-Tier Firms Provide Auds of Similar Qualy?, Journal of Accounting and Public Policy 29 (4): Cassell, C. A., Giroux, G., Myers, L. A., and Omer, T. C. (203), The Emergence of Second Tier Audors in the US: Evidence from Investor Perceptions of Financial Reporting Credibily, Journal of Business Finance and Accounting 40 (3-4): Chan, K. H., Lin, K. Z., and Wong, B. (200), The Impact of Government Ownership and Instutions on the Reporting Behavior of Local Audors in China, Journal of International Accounting Research 9 (2): -20. Chang, H., Cheng, C. A., and Reichelt, K. J. (200), Market Reaction to Audor Swching from Big 4 to Third-Tier Small Accounting Firms, Auding: A Journal of Practice and Theory 29 (2): Chen S., Sun, Y., and Wu, D. (200), Client Importance, Instutional Improvements, and Aud Qualy in China: An Office and Individual Audor Level Analysis, The Accounting Review 85 (): DeAngelo, L. (98), Audor Size and Aud Qualy, Journal of Accounting and Economics 3 (3): DeFond, M. L., Wong, T. J., and Li, S. (999), The Impact of Improved Audor Independence on Aud Market Concentration in China, Journal of Accounting and Economics 28 (3):