Abstract
China Journal of Accounting Studies, 2015 Vol. 3, No. 4, 294–319, http://dx.doi.org/10.1080/21697213.2015.1100089 Fair value accounting of financial assets and analyst forecasts Zhe Wei and Jian Xue* School of Economics and Management, Tsinghua University, Beijing 100084, China We examine how listed firms’ financial asset holdings affect analyst forecasts. Using China A-share firms over the period 2009–2011, we find that (1) listed firms sell available-for-sale securities (AFS) to meet or beat analyst forecasts, and (2) that when firms hold AFS, their analyst forecasts are more accurate, less biased and less dispersed. However, whether firms hold trading securities (TS) has no significant effect on either analyst forecast accuracy or forecast dispersion. Further examination indicates that listed firms use AFS to manipulate earnings, and that outside governance by the financial market and legal environment cannot moderate this opportunistic behaviour. Finally, our empirical results show that analysts can see through firms’ earnings management in selling AFS. Thus, our results suggest that standard setters should consider managers’ opportunistic behaviour in derecognising financial assets. Keywords: analyst forecasts; earnings management; fair value accounting; financial assets 1. Introduction China started implementing a new accounting standard (hereinafter ‘CAS’)on1 January 2007. One of the important changes of the new CAS was the introduction of fair value accounting to various assets (financial assets and liabilities, debt restructuring, investment properties, and so on). According to the new CAS, trading securities (here- inafter ‘TS’) and available-for-sale securities (hereinafter ‘AFS’) are measured at their fair value and classified according to the managers’ purpose in holding them. TS and AFS are treated differently at subsequent measurements: the fair value changes on TS go directly through the income statement, while the fair value changes on AFS are recognised directly in equity, through the statement of changes in equity. The cumula- tive gains and losses are recognised in profits or losses when AFS are derecognised. Because cross-holding is prevalent among Chinese listed firms and the historical cost of such holding is low, the measurement of AFS offers managers the opportunity to manipulate earnings: listed firms can meet or beat specific earnings targets by selling AFS. For instance, Youngor Group Company Limited (SH600177) sold 106 million CITIC Securities Company Limited (SH600030) shares and recognised a RMB2.58 billion profit from this deal with respect to a net income of RMB2.38 billion for the 2008 fiscal year. In 2009, the Ministry of Finance of China released Interpretation No. 3 of the Accounting Standards for Business Enterprises and required listed firms to dis- close their capital gains and losses in detail in ‘Other Comprehensive Income’ (OCI) to provide more transparent financial reports. Without changing the subsequent *Corresponding author. Email: xuejian@sem.tsinghua.edu.cn Paper accepted by Liansheng Wu. © 2015 Accounting Society of China China Journal of Accounting Studies 295 measurement of AFS, firms can continue to manipulate earnings through AFS under the new CAS. On 24 July 2014, the International Accounting Standards Board (IASB) issued IFRS 9 to replace the original IAS 39. Under IFRS 9, financial equities are clas- sified either as measured (1) at fair value through profit or loss (FVTPL), or (2) at fair value through other comprehensive income (FVTOCI). The classification is irrevocable and once the AFS are classified as FVTOCI, the (cumulative) change in their fair value will not go through profit or loss when the AFS are derecognised. The IASB believes that IFRS 9 can provide more transparent disclosure about financial assets and moder- ate the opportunistic behaviour of managers. Thus, whether the subsequent measure- ment of financial assets (specifically, that of TS and AFS) affects the valuation of firms remains unclear. Furthermore, should the new CAS converge with IFRS 9? We provide empirical evidence that supports the new measurement proposed in IFRS 9 by examining whether TS and AFS affect analyst forecasts differently. We use sell-side analysts to represent our investors, and provide empirical evidence on whether the accounting measurements of TS and AFS affect investors’ valuations. Our results demonstrate the following: (1) Firms are more likely to meet or beat analyst forecasts only when they hold AFS at the beginning of the fiscal year. This indicates that managers can use AFS to manipulate earnings. (2) When firms hold AFS, their analysts make forecasts with higher accuracy, lower forecast error and less disagree- ment. (3) Whether firms hold TS or not has no impact on the accuracy or dispersion of analyst forecasts. However, the forecast error is smaller when listed firms hold TS. Our research is related to the literature on fair value accounting and analyst fore- casts. Byard, Li, and Yu (2011) examine the effect of mandated IFRS adoption on ana- lyst forecasts based on European listed firms. Their results show that the IFRS adoption can lower analysts’ forecast error and dispersion when the country has strong enforcement. Tan, Wang, and Welker (2011) find that the IFRS adoption can improve the forecast accuracy of analysts from overseas or with overseas working experience. Their results indicate that the IFRS adoption can lower the information asymmetry among investors. Neither Byard et al. (2011) nor Tan et al. (2011) directly examine the effect of fair value accounting on analyst forecasts. Liang and Riedl (2014) compare the analyst forecasts for UK and US investment property firms and find that (1) fair value accounting can increase the information on the balance sheet, and (2) including the changes of fair value in profits and losses lowers the accuracy of analysts’ earnings forecasts. Different from previous studies that compare historical accounting with fair value accounting, we focus on the impact of accounting measurements of fair value assets. Our results indicate that allowing cumulative fair value changes to be recycled can give managers opportunities to manipulate earnings and affect investors’ valuations. Our findings are also related to recent studies on AFS and earnings management. He, Wong, and Young (2012) use China A-share firms to examine the relation between AFS realised gains and losses and TS unrealised gains and losses. They show that Chi- nese listed firms make up the unrealised loss of TS by selling AFS. Ye, Zhou, Li, and Guo (2009) and Gu, Wang and Xue (2015) both find that managers prefer to classify financial assets as AFS when implementing the new CAS. They also demonstrate that Chinese listed firms use AFS to smooth earnings. We expand the sample period from 2007–2008 in previous research to 2009–2011 and provide further evidence to show that managers not only use AFS to smooth earnings but also to meet or beat analyst forecasts. 296 Wei and Xue Our study contributes to the literature in the following ways: first, to the best of our knowledge, we are the first to examine how firms’ financial asset holdings affect analyst forecasts with a large sample. Our results show that Chinese listed firms pick the timing of AFS sales to meet or beat analyst forecasts. Second, previous literature focuses either on the incremental information contained in fair value accounting versus historical accounting, or on whether managers use the fair value measurement of finan- cial assets to smooth earnings. Few studies combine measurements of financial assets on fair value, management incentive and investor response. Our research fills this gap by examining the impact of financial asset holdings on analyst forecasts. Last, our results also suggest that China’s standard setters should converge with IFRS 9 and stop allowing cumulative fair value gains to be recycled when AFS are derecognised, because this would moderate managers’ opportunistic sales of AFS. Thus, we can make the financial reports more transparent, more reflective of managers’ effort and more useful for investors’ valuations. The rest of this paper is organised as follows. Section 2 describes the institution background and related literature. Section 3 develops our hypotheses and describes our research design. Sections 4 and 5 present the empirical results. Section 6 concludes. 2. Institution background and literature review 2.1. Institution background The Ministry of Finance of China issued the new CAS on 15 February 2006 to converge with international accounting standards. The new CAS came into force on 1 January 2007. One of its important changes was the adoption of fair value measure- ment on various assets. In this paper, we focus on the accounting measurements of TS and AFS on fair value according to the new CAS. Before the new CAS was implemented, financial assets were classified as bond investment, equity investment or accounts receivable. They were shown as either long- term or short-term assets on a balance sheet, and all were measured at historical cost. Under the new CAS, all financial assets are classified as TS, AFS, held-to-maturity, and loans and receivables. Held-to-maturity investments and loans and receivables are measured at amortised cost using the effective interest rate. TS and AFS are measured at their fair value. The changes in fair value (the unrealised gains and losses) of TS are directly recognised in profits and losses. However, the unrealised gains and losses of AFS go directly into equity and the cumulative changes in the fair value are recognised in profits and losses when the AFS are derecognised. To ensure the proper implementation of the new CAS, the Ministry of Finance of China issued Interpretation No. 3 of the Accounting Standards for Business Enterprises on 21 June 2009. In this interpretation, the government required additional disclosure about the gains and losses that did not go into profits and losses under ‘Other Compre- hensive Income’ (OCI) (a sub-item under ‘Earnings Per Share’). Thus, Chinese listed firms had to disclose their unrealised gains and losses from AFS in OCI after June The measurements of financial assets are not uncontroversial. Improving classifica- tion and measurement became a main project of the IASB after the recent financial cri- sis. On 24 July 2014, the IASB issued a new version of IFRS 9 with substantial changes in the classification and measurement of financial assets. We focus on the sub- sequent measurement of FVTOCI financial equities. Under IFRS 9, the cumulative fair China Journal of Accounting Studies 297 value changes on FVTOCI financial equities will not go into profits and losses when derecognised. Does this change benefit outside investors? Shall the new CAS converge with IFRS 9 here? These are the two main questions this paper tries to answer. 2.2. Fair value accounting and unrealised gains or losses The recent financial crisis sparked a lively debate about fair value accounting. Whether fair value accounting is better than historical accounting remains controversial. Support- ers of fair value accounting suggest that the fair value of assets provides value relevant information (e.g. Barth, 1994; Eccher, Ramesh, & Thiagarajan, 1996; Kothari, 2001) and helps investors make investment decisions. Most empirical research about fair value accounting examines the value relevance of assets. Barth (1994) uses commercial banks to demonstrate that fair value accounting provides incremental information. Wang, Xue, and Li (2011) find that the fair value of financial assets has information in addition to their historical cost using Chinese data. Xu and Zeng (2013) further demon- strate that the disclosure of unrealised gains and losses of AFS has an impact on firms’ stock market prices. Liang and Riedl (2014) do not directly test the incremental infor- mation of fair value accounting. They examine the impact of fair value accounting on analyst forecasts. Their results show that the net asset forecasts for UK investment property firms are more accurate than those for US investment property firms. Their results also demonstrate that incorporating unrealised gains and losses into profits and losses reduces the accuracy of analysts’ earnings forecasts by increasing earnings volatility. Other related literature shows that the adoption of fair value accounting may improve the information environment and analyst forecast accuracy, as well as reducing the analyst forecast dispersion, but only when the country has strong enforcement (e.g. Byard et al., 2011; Tan et al., 2011). The unrealised gains and losses generated from the changes in financial asset fair value can either go directly into profit and loss (TS) or into equity (AFS) and be dis- closed in OCI. Hirst and Hopkins (1998) designed experiments and found that buy-side analysts cannot properly reflect the information of AFS unrealised gains and losses in their valuations when the unrealised gains and losses do not go into profits and losses. Tarca, Brown, Hancock, Woodliff, Bradbury and van Zijl (2008) used their experiments to show that prohibiting the recycling of cumulative unrealised gains and losses into profits and losses can help the investor better understand the information available. Our research adds empirical evidence to this previous experimental literature. 2.3. Financial assets and earnings management Since the work of Burgstahler and Dichev (1997), which demonstrated the wide practice of earnings management, researchers have paid much attention to the subject. Earlier studies focus on the measurement issue of earnings management and use abnor- mal accruals as indicators of earnings management (e.g. Dechow, Sloan, & Sweeney, 1995; Jones, 1991; Wang, Li, & Chen, 2015). However, it is difficult to identify which item of accruals is used and the calculation of abnormal accruals is also unsatisfactory (Bernard & Skinner, 1996; Ecker, Francis, Olsson, & Schipper, 2011). Many researchers use specific industry companies as research samples to investigate how firms manipulate earnings. Liu and Ryan (2006) find that commercial banks do this through provisions for loan losses and loan charge-offs. In contrast, Dong, Ryan and Zhang (2009) find that banks manipulate earnings by reclassifying unrealised 298 Wei and Xue holding gains or losses from marketable securities. These studies explore how banks manipulate earnings, but the results are limited to the banking industry and are difficult to extend to other industries. McNichols and Wilson (1988) find that firms use provi- sions for bad debt to manipulate earnings. Teoh, Wong, and Rao (1998) suggest that initial public offering (IPO) firms adopt a straight line depreciation method to increase earnings before IPO. Roychowdhury (2006) demonstrate that firms can also manipulate earnings through real activities (e.g. overproduction, cut-off R&D expense). Bartov (1993), Poitras, Wilkins, and Kwan (2002) and Herrmann, Inoue, and Thomas (2003) use data from the US, Singapore and Japan and find that firms smooth earnings through the sale of specific assets (e.g. fixed assets). The accounting measurement of AFS offers managers an opportunity to manipulate earnings at lower cost: the cumulative unrealised gains or losses are recognised in profit and loss once the AFS are sold. Managers can choose the time to sell AFS to manipu- late earnings. By selling AFS to achieve this, (1) managers face lower litigation risks, and (2) the impact of the AFS sale on future performance is weaker than other real activity earnings management. Similar to our research, He et al. (2012) use China A-share firms over 2007–2008 and find that firms sell AFS when they make losses from TS; Ye et al. (2009) find that Chinese firms are more likely to reclassify financial assets as AFS when the new CAS is implemented. They also find that firms smooth earnings through the sale of AFS. Similarly, Gu et al. (2015) find that Chinese firms use AFS to avoid losses. All these studies use the same sample period from 2007 to 2008. The Ministry of Finance of China issued Interpretation No. 3 of the Accounting Standards for Business Enterprises on 21 June 2009. The extra disclosure requirement may moderate earnings management through AFS. Thus, whether previous findings can be extended to a later period is still questionable. We have extended our sample period to 2009–2011 to examine whether the stricter disclosure requirement can moderate managers’ opportunistic behaviour through selling AFS. 2.4. Financial assets, earnings management and analyst forecasts Analysts are important information intermediaries in financial markets and their fore- casts can affect investors’ decisions. Therefore, managers have incentives to meet or beat analyst forecasts through earnings management (Degeorge, Patel, & Zeckhauser, 1999; Matsumoto, 2002). Herrmann et al. (2003) find that Japanese firms manipulate earnings to meet or beat previous forecasts by selling fixed assets and marketable secu- rities. Hunton, Libb, and Mazza (2006) also find, in their experimental test, that man- agers use AFS to meet or beat analyst forecasts. Our study adds empirical evidence to the experimental results of Hunton et al. (2006) by showing that China A-share firms meet or beat analyst forecasts through the sale of AFS. Different from sales of other assets, the cumulative fair value changes in AFS have already been included in stockholders’ equity and disclosed in OCI, so it is likely that analysts see through managers’ incentive to sell AFS. However, whether analysts can do this is not well studied or fully investigated. Burgstahler and Eames (2003) find that analysts take earnings management into account when they make earnings forecasts, but that analysts cannot identify which firms manipulate their earnings. Hirst and Hopkins (1998) find that buy-side analysts cannot see through firms’ earnings management when firms sell AFS. Using sell-side analysts’ forecasts, our research shows that, to some extent, analysts can see through firms’ earnings management in selling AFS. China Journal of Accounting Studies 299 3. Hypotheses development and research design 3.1. Hypotheses development TS held by non-financial industry firms in China are stocks, funds, bonds and financial derivatives. AFS are mainly composed of stocks and funds. Therefore, TS are similar to AFS. Whether a financial asset is classified as TS or AFS is mainly determined by the intention of managers. According to the new CAS, the unrealised gains and losses of TS go directly into net income each year. The unrealised gains and losses of AFS go into equity and the cumulative unrealised gains and losses are recognised as net income when AFS are derecognised. The different accounting rules for TS and AFS make it possible for managers to manipulate earnings through AFS but not TS. Differ- ent from earnings management by accruals, earnings management through AFS does not violate any accounting standards or regulations. Thus, the litigation risk is lower for managers who manipulate earnings through AFS. Moreover, manipulating earnings through AFS has less impact on firms’ future performance compared with other real activity earnings management (e.g. sale of fixed assets or goodwill, cut-off R&D expense). Overall, earnings management through AFS costs the least compared with other methods. Firms manipulate earnings to meet various targets: He et al. (2012) find that Chi- nese firms make up unrealised losses from TS by selling AFS. Gu et al. (2015) show that firms sell AFS to avoid losses. Previous literature indicates that meeting or beating analyst forecasts is also an important earnings target for listed firms (e.g. Matsumoto, 2002). Whether firms in China meet or beat analyst forecasts by selling AFS is an unanswered question. If firms do manipulate earnings to meet or beat analyst forecasts through sales of AFS, then firms with (or with more) AFS are more likely to meet or beat analyst forecasts. In the meantime, firms with more realised gains from AFS sales are more likely to meet or beat analyst forecasts. However, if firms do not use AFS to meet or beat analyst forecasts, then whether to hold AFS, the amount of the holding and the realised gains from selling them should have no relation to the likelihood of meeting or beating analyst forecasts. Stated formally, our hypothesis H1a (in null form) is as follows: H1a：Whether firms hold AFS, hold more AFS or have more realised gains from selling AFS is not related to whether they meet or beat analyst forecasts. Managers can also choose when to sell TS to realise more gains. However, the unre- alised gains and losses are already recognised in net income; therefore it is impossible for firms to manipulate earnings by selling TS. Although TS and AFS are similar in nature, their different accounting rules mean they have different effects on an income statement. So whether firms hold (or hold more) TS or have more realised gains from TS has no effect on their meeting or beating analyst forecasts. Our hypothesis H1b is stated (in null form) as follows: H1b: Whether firms hold TS, hold more TS or have more realised gains from selling TS is not related to whether they meet or beat analyst forecasts. Does the decision of firms as to when to sell AFS increase the information asymmetry between insiders and outsiders? Does this behaviour affect investors’ decision? Using sell-side analysts as representative of outside investors, we try to investigate the impact of AFS holdings on analyst forecasts. There are three possible scenarios: (1) Firms do not manipulate earnings by selling AFS. In this case, the information asymmetry for 300 Wei and Xue firms with AFS is higher and it is difficult for analysts to predict when they will sell AFS. Thus, analyst forecasts become less accurate and more dispersed. (2) Firms manipulate earnings to meet or beat analyst forecasts by selling AFS, and analysts can- not see through this opportunistic behaviour by managers. In this scenario, we predict that the dispersion among analyst forecasts is larger when firms hold (more) AFS. However, analyst forecasts would be more accurate and less biased if firms held AFS in these circumstances. (3) Firms manipulate earnings by selling AFS and analysts can see through this opportunistic behaviour. In this case, their forecasts would be more accurate, less biased and less dispersed when firms hold AFS. However, when firms hold more AFS, we do not have any prediction about analyst forecast accuracy, forecast error or dispersion. Assume Firm A holds AFS while Firm B does not. Then, because analysts know that Firm A will use AFS to smooth earnings and Firm B cannot, their forecasts for firm A will be more accurate and less dispersed. If both Firms A and B hold AFS and Firm A holds more than Firm B, then they can both smooth earnings by selling AFS. We cannot predict that Firm A’s analyst forecasts would be more accurate than Firm B’s. To summarise, how holding AFS affects analyst forecasts is an empirical question and our hypothesis H2a (in null form) is stated as follows: H2a: Whether to hold AFS, and how much to hold, is not related to analyst forecast accuracy, bias or dispersion. Firms cannot manipulate earnings by selling TS because their unrealised gains or losses go directly into net income each year. Fair value changes in TS cannot be controlled by managers, including unrealised gains and losses of TS in net income increases firms’ earnings volatility (Liang & Riedl, 2014). Therefore, it is more difficult for ana- lysts to make forecasts when firms hold TS, and their forecasts are less accurate and more dispersed. However, firms hold TS to offset their operating risk. Firms can reduce their earnings volatility by selling TS, and analyst forecasts are more accurate and less dispersed when firms hold TS. The two scenarios may coexist and we do not have a clear prediction of the effects of holding TS on analyst forecasts. Our last hypothesis, H2b (in null form), is as follows: H2b: Whether to hold TS, and how much to hold, is not related to analyst forecast accu- racy, bias or dispersion. 3.2. Research design We use the following models to test our hypotheses. First, we use the logistic regression model (1) to test whether firms manipulate earnings to meet or beat analyst forecasts by selling TS or AFS. MTBT ¼ a þ a Financial Instrument =Financial Instrument Gains i;t 0 1 i;t1 i;t þ b Controls þ e (1) j;i;t i;t MTBT is a dummy variable indicating whether firm i meets or beats its analyst fore- i;t casts; Financial Instrument is the holding of financial assets of firm i in year t, i,t-1 which is used as D_TS , D_AFS , TS or AFS in the regression. D_TS i,t-1 i,t-1 i,t-1 i,t-1 i,t-1 (D_AFS ) is a dummy variable indicating whether firm i holds TS (AFS) at the i,t-1 beginning of year t. TS (AFS )is defined as the fraction of total TS (AFS) to total i,t-1 i,t-1 assets. D_URG_TS , D_RG_TS , D_RG_AFS , URG_TS , RG_TS and RG_AFS i,t i,t i,t i,t i,t i,t China Journal of Accounting Studies 301 are used as Financial Instrument Gains in the regression, respectively. D_URG_TS i,t i,t (D_RG_TS )[D_RG_AFS ] is a dummy variable indicating whether firm i has unre- i,t i,t alised (realised) [realised] gains and losses from holding TS (selling TS) [selling AFS] in year t. URG_TS (RG_TS )[RG_AFS ] is a continuous variable, which is the frac- i,t i,t i,t tion of unrealised gains or losses from TS (realised gains or losses from TS) [realised gains or losses from AFS] to total assets for firm i at the beginning of year t.If firms manipulate earnings to meet or beat analyst forecasts through AFS, α is significantly positive when D_AFS , D_RG_AFS or RG_AFS are used in the regression. As i,t-1 i,t i,t discussed earlier, firms cannot use TS to manipulate earnings because unrealised gains or losses from TS go directly into net income. α is insignificant when D_TS , 1 i,t-1 D_URG_TS , D_RG_TS or RG_TS are used in the regression. We control for firm i,t i,t i,t characteristics that affect analyst forecasts (Barth, Kasznik, & McNichols, 2001; Lang & Lundholm, 1996). We also control for firm size (Ln(TA )), unexpected earnings sur- –1 prise (Surprise), past performance (ROA ), change in performance (Change_ROA ), i,t–1 i,t earnings volatility (std_ROA ) and growth (Growth_ROA ), average days i,t-1~t-5 i,t-1~t-3 between analyst forecasts and earnings announcement date (Ln(Day+1) ), average i,t firm-specific experience (Ln(Exp_firm+1) ) and average industry-specific experience i,t (Ln(Exp_ind+1) ). i,t We use the following ordinary least squares (OLS) model to test hypotheses H2a and H2b: Analyst Forecast ¼ c þ c Financial Instrument þ b Controls þ (2) i;t i;t1 j;i;t i;t 0 1 j Analyst_Forecast is the analyst forecast characteristics for firm i in year t, including i,t forecast accuracy (the absolute forecast error, Accy), forecast error (Bias) and forecast dispersion (Dispersion). Financial Instrument and Controls are the same as those in i,t-1 model (1). Definitions of all variables are in Table 1. Predictions for coefficients of Financial Instrument , γ are different in different scenarios. In the first scenario i,t-1 1 when firms do not manipulate earnings by selling AFS, the coefficients of D_AFS i,t-1 and AFS are positively significant. In the second scenario when firms manipulate i,t-1 earnings to meet or beat analyst forecasts by selling AFS and analysts cannot see through their opportunistic behaviour, coefficients of D_AFS and AFS are nega- i,t-1 i,t-1 tively significant when the dependent variable is Accy or Bias; when the dependent variable is Dispersion, the coefficient is insignificantly different from zero. In the last scenario, when firms use AFS to manipulate earnings and analysts see through the managers’ incentive, the coefficient of D_AFS is negatively significant for all vari- i,t-1 ables. In this scenario, we do not have any prediction for the coefficients of AFS i,t-1. Also, the coefficients of D_TS and TS can be either positive, negative or insignif- i,t-1 i,t-1 icantly different from zero. 4. Sample selection and empirical results 4.1. Sample selection We begin with all China A-share listed firms over the period 2009–2011 as our research sample. We start from 2009 because the Ministry of Finance of China issued Interpretation No. 3 of the Accounting Standards for Business Enterprises on 21 June 2009. Since 2009, all firms have had to disclose capital gains or losses from AFS in OCI. Interpretation No. 3 of the Accounting Standards for Business Enterprises makes the information on AFS more transparent than in the period 2007–2008 when China 302 Wei and Xue Table 1. Variable definitions. Variables Symbols Definitions A：Dependent Variables Analyst forecast accuracy Accy Absolute consensus forecast error: jForecast Eps j=Forecast : i;t i;t i;t Analyst forecast error Bias Consensus forecast signed error: ðForecast Eps Þ=Forecast : i;t i;t i;t Analyst forecast dispersion Dispersion Analyst forecast standard deviation, scaled by consensus forecast. Meet or beat analyst MTBT Equals 1 if actual earnings meet or beat analyst forecasts forecasts; zero otherwise. Consensus forecast Forecast Average analyst forecast. B：Independent Variables TS dummy variable D_TS Equals 1 if firms hold trading securities (TS) at the -1 beginning of the year, zero otherwise. AFS dummy variable D_AFS Equals 1 if firms hold available-for-sale securities -1 (AFS) at the beginning of the year, zero otherwise. TS TS Total TS at the beginning of the year, scaled by -1 beginning total assets. AFS AFS Total AFS at the beginning of the year, scaled by -1 beginning total assets. Unrealised gains and D_URG_TS Equals 1 if firms record any unrealised gains or losses losses from TS dummy from TS, zero otherwise. Realised gains and losses D_RG_AFS Equals 1 if firms record any realised gains or losses from TS dummy from TS, zero otherwise. Realised gains and losses D_RG_AFS Equals 1 if firms record any realised gains or losses from AFS dummy from AFS, zero otherwise. Unrealised gains and URG_TS Unrealised gains and losses from TS, scaled by losses from TS beginning total assets. Realised gains and losses RG_TS Realised gains and losses from TS, scaled by from TS beginning total assets. Realised gains and losses RG_AFS Realised gains and losses from AFS, scaled by from AFS beginning total assets. C：Control Variables Total assets TA Total assets at the beginning of the fiscal year. -1 Unexpected earnings Surprise Change in the earnings per share, scaled by current earnings per share: (EPS -EPS )/EPS . t t-1 t Return on assets ROA Last period earnings, scaled by beginning total assets. -1 Change in ROA Change_ROA Current ROA minus last period ROA: ROA -ROA . t t-1 Performance volatility std_ROA Standard deviation of past 5 year ROAs. Growth Growth_ROA Average ROA growth for past 3 years. Forecast interval Day Days between analyst forecast and actual earnings announcement date. Firm-specific experience Exp_firm Average years analysts follow specific firm. Industry-specific Exp_ind Average years analysts follow specific industry. experience started implementing its new CAS. Given that information about realised and unrealised gains or losses from TS and AFS is hand collected, we limited our sample to the year 2011. Other financial data and analyst forecast data are from the China Stock Market and Accounting Research (CSMAR) database. We use the last forecast from each analyst before the fiscal year-end. Then we merge the analyst forecast data with financial data and get 4,821 observations. Our tests require us to calculate the China Journal of Accounting Studies 303 forecast dispersion, so we eliminate 1,077 firm-year observations with one or two analyst forecasts only. Finally, we eliminate observations without sufficient financial data to do the tests. Our final sample includes 1,883 observations. Table 2 exhibits the descriptive statistics of this sample. As Table 2 shows, 29% of our observations hold TS and 32% hold AFS; 46% of the observations with TS hold AFS at the same time. Around 42% of the observations with AFS hold TS in the same year. The average holding of TS is 1% relative to total assets and the percentage for AFS is 3.55%. These results indicate that (1) firms hold more AFS than TS, and (2) our research question is relevant among Chinese listed firms. The average absolute forecast error is 33% for our full sample. The average absolute forecast error for TS-holding firms is 31%, which is not significantly different from the full sample. When firms hold AFS at the beginning of the fiscal year, the average absolute forecast error is 29%. The average absolute forecast error for AFS-holding firms is 4% smaller than that for the full sample and the difference is significant at a 10% α-level. The average forecast error of the full sample is 23% (significantly larger than 0). This shows that the analysts are overoptimistic, which is consistent with the literature. The average forecast errors for our subsamples are smaller than for the full sample which demonstrates that analysts are more cautious when making forecasts for TS- and AFS-holding firms. The average forecast dispersion of subsamples (0.26) is not significantly different from that of the full sample (0.27). Twenty six percent of all firm-year observations meet or beat their consensus analyst forecast; 28% of TS-holding observations meet or beat their consensus analyst forecast in our sample. The likelihood of meeting or beating analyst forecasts rises to 30% when firms hold AFS, which is consistent with our hypothesis that firms can manipulate earnings with AFS to meet or beat analyst forecasts. For the control variables, the average interval from analyst forecasts to fiscal year- end is 166 days; analysts follow a specific firm for half a year on average, whereas they follow a specific industry for 2 years 6 months on average. Untabulated results show that there is no significant difference between firms with and without financial assets in any firm’s characteristics except size. The average size of firms with financial assets is significantly larger than that of firms without financial assets. We discuss the empirical results in the next section. 4.2. Empirical results 4.2.1. Financial assets and meeting or beating analyst forecasts We examine whether firms holding financial assets affect the likelihood of meeting or beating analyst forecasts with model (1) in this section. Table 3 exhibits our regression results. The dependent variable is MTBT in Table 3. In Columns 1 and 2 of the table, we test whether holding financial assets is related to meeting or beating analyst fore- casts. In Columns 3 and 4, we examine the relation between financial assets gains and losses and meeting or beating analyst forecasts. Column 1 shows that whether firms hold TS or not does not affect the odds of firms meeting or beating analyst forecasts. However, it is more likely that a firm can meet or beat analyst forecasts when it holds AFS at the beginning of the fiscal year ða ¼ 0:300; p\0:05Þ. Untabulated D AFS results show that the odds of AFS-holding firms meeting or beating analyst forecasts is 1.35 times larger than that of firms without financial assets. As shown in Column 2, the more AFS a firm holds, the more likely it will meet or beat its analyst forecasts 304 Wei and Xue Table 2. Descriptive statistics. Full Sample D_TS =1 D_AFS =1 -1 -1 Variable N Mean Median N Mean Median N Mean Median Accy 1,883 0.33 0.20 553 0.31 0.19 601 0.29 0.18 Bias 1,883 0.23 0.15 553 0.19 0.13 601 0.17 0.11 Dispersion 1,883 0.27 0.19 553 0.26 0.19 601 0.26 0.17 MTBT 1,883 0.26 0 553 0.28 0 601 0.30 0 Day 1,883 166.89 163.00 553 164.75 161.44 601 164.84 161.55 EXPY_firm 1,883 0.77 0.67 553 0.75 0.67 601 0.84 0.79 EXPY_ind 1,883 2.34 2.33 553 2.25 2.25 601 2.32 2.33 b 10 10 10 10 10 10 TA (Yuan) 1,883 2.14×10 0.50×10 553 3.65×10 0.61×10 601 4.51×10 0.81×10 -1 TS 1,883 0.28% 0.00% 553 0.95% 0.10% 601 0.62% 0.00% -1 AFS 1,883 1.10% 0.00% 553 1.97% 0% 601 3.55% 1.31% -1 Surprise 1,883 0.42 0.18 553 0.54 0.21 601 0.50 0.20 ROA 1,883 0.08 0.07 553 0.08 0.07 601 0.08 0.07 -1 Change_ROA 1,883 0.00 0.00 553 0.00 0.00 601 0.00 0.00 std_ROA 1,883 0.04 0.00 553 0.04 0.03 601 0.04 0.03 Growth_ROA 1,883 0.00 0 553 0.00 0 601 0.00 0.00 D_URG_TS 1,883 0.27 0 553 0.72 1 601 0.35 0 D_RG_TS 1,883 0.32 0 553 0.74 1 601 0.40 0 D_RG_AFS 1,883 0.25 0 553 0.34 0 601 0.73 1 URG_TS 1,883 0.03% 0.00% 553 0.08% 0.00% 601 0.05% 0.00% RG_TS 1,883 0.03% 0.00% 553 0.07% 0.00% 601 0.03% 0.00% RG_AFS 1,883 0.18% 0.00% 553 0.29% 0.00% 601 0.50% 0.00% D_TS 1,883 0.29 0 553 1 1 601 0.42 0 -1 D_AFS 1,883 0.32 0 553 0.46 0 601 1 1 -1 All variables are defined in Table 1. All continuous variables are winsorised at top and bottom 1%. a: we use the log transform of this variable as the independent variable in our regression. b: we use the log transform of this variable as the control variable in our regressions. China Journal of Accounting Studies 305 Table 3. Financial assets and meet or beat analyst forecasts. MTBT MTBT MTBT MTBT (1) (2) (3) (4) D_TS 0.058 -1 (0.46) D_AFS 0.300** -1 (2.37) TS −2.582 -1 (-0.55) AFS 3.255** -1 (2.13) D_URG_TS −0.283* (-1.81) D_RG_TS 0.517*** (3.53) D_RG_AFS 0.365** (2.79) URG_TS 28.638 (1.19) RG_TS 15.249 (0.64) RG_AFS 12.039* (1.92) Ln(Day+1) −1.202*** −1.208*** −1.211*** −1.222*** (-6.53) (-6.57) (-6.59) (-6.58) Ln(Expy_firm+1) 0.111 0.112 0.161 0.131 (0.49) (0.49) (0.70) (0.57) Ln(Expy_ind+1) −0.169 −0.172 −0.211 −0.218 (-0.69) (-0.71) (-0.86) (-0.89) Ln(TA ) −0.031 −0.011 −0.034 0.006 -1 (-0.55) (-0.21) (-0.62) (0.11) Surprise 0.021 0.023 0.022 0.021 (0.97) (1.07) (1.00) (0.97) ROA 0.105 −0.027 −0.154 −0.018 -1 (0.08) (-0.02) (-0.12) (-0.01) Change_ROA −0.180 −0.261 −0.144 −0.226 (-0.16) (-0.24) (-0.13) (-0.20) std_ROA −0.586 −0.188 −0.712 −0.762 (-0.36) (-0.12) (-0.44) (-0.46) Growth_ROA 3.747 3.949 3.987 4.361 (1.21) (1.28) (1.28) (1.38) Constant 5.214*** 4.870*** 5.291*** 4.606*** (3.26) (3.05) (3.33) (2.88) Year Fixed Yes Yes Yes Yes Industry Fixed Yes Yes Yes Yes Observations 1,883 1,883 1,883 1,883 Pseudo R 0.073 0.072 0.081 0.075 All variables are defined in Table 1. All continuous variables are winsorised at top and bottom 1%. The z-statistics in () are calculated with robust errors; ***, **, * indicate significance at the 0.01, 0.05 and 0.10 levels, respectively, in two-tailed tests. The dependent variable is MTBT. The independent variables in Columns 1 and 2 are financial assets. The independent variables in Columns 3 and 4 are gains and losses from financial assets. (a ¼ 3:255; p\0:05). It is easier to manipulate earnings with AFS to meet or beat AFS analyst forecasts when the firm holds more AFS. Thus, firms with more AFS are more likely to meet or beat analyst forecasts. Among the control variables, only Ln(Day+1) i,t 306 Wei and Xue is significantly different from zero. The results indicate that the earlier the analyst forecasts are made, the more difficult it is for firms to meet or beat analyst forecasts. As discussed in Section 3, if firms manipulate earnings to meet or beat analyst fore- casts through the sale of AFS, we can observe that firms with more gains from AFS are more likely to meet or beat analyst forecasts. Table 3 Column 3 shows that if firms generate realised gains and losses from AFS, they are more likely to meet or beat ana- lyst forecasts (a ¼ 0:365; p\0:01). The results in Column 3 also indicate that D RG AFS firms realising gains and losses from TS are more likely to meet or beat analyst forecasts (a ¼ 0:517; p\0:01). However, firms are less likely to meet or beat D RG TS analyst forecasts if they have any unrealised gains or losses from TS. The results show that firms can realise more gains from TS by choosing the right time to sell them. Column 4 shows that neither realised gains and losses nor unrealised gains and losses from TS are related to the odds of firms meeting or beating analyst forecasts. In contrast, the coefficient of RG_AFS is positive and significant at a 10% α-level. Overall, the results in Table 3 are consistent with our suggestion that firms can manipulate earnings to meet or beat analyst forecasts through AFS but not TS. Firms not only use AFS to smooth earnings (He et al., 2012; Ye et al., 2009), they also manipulate earnings with AFS to meet or beat analyst forecasts. 4.2.2. Financial assets and analyst forecasts We test hypotheses H2a and H2b with model (2) in this section. Table 4 presents our regression results. The dependent variable for Table 4, Columns 1 and 2, is forecast accuracy (Accy). Columns 3 and 4 (5 and 6) use forecast dispersion (error) Dispersion (Bias) as the dependent variable. Column 1 shows that whether firms hold TS or not does not affect their analyst forecast accuracy. The results in Column 2 also indicate that the amount of TS that firms hold has no effect on their analyst forecast accuracy. The results are different from those of Liang and Riedl (2014). The reason is that our sample is different from theirs. In their study, the unrealised gains and losses only increase the performance volatility of investment property firms. In our sample, as dis- cussed previously, TS can be used to both offset performance volatility and make it easier for analysts to forecast firms’ future performance. Different from TS, analyst forecast accuracy for AFS-holding firms is 3.7% higher (which means that the forecast accuracy is improved by 10%) than that of firms without financial assets. We do not find evidence that the more AFS firms have, the more accurate are their analyst forecasts. With reference to the control variables, the results are consistent with previous research: the earlier the analyst forecasts are made, the less accurate they are; past per- formance is positively related to analyst forecast accuracy; performance volatility is negatively related to analyst forecast accuracy; firm size is negatively related to analyst forecast accuracy, but not significantly. In Table 4, Columns 3 and 4, we examine the impact of the holding of financial assets on analyst forecast dispersion. The results show that a firm’s holding of TS has no effect on the dispersion. Similar to the results in Table 4, Columns 1 and 2, these results also indicate that the analyst forecast dispersion is 8% lower for AFS-holding firms and the amount of AFS has no significant effect on forecast dispersion. These results are consistent with scenario 3, described earlier, and demonstrate that analysts can, to some extent, see through firms smoothing earnings with AFS. China Journal of Accounting Studies 307 Table 4. Financial assets and analyst forecast characteristics Accy Accy Dispersion Dispersion Bias Bias (1) (2) (3) (4) (5) (6) D_TS −0.030 −0.008 −0.052** -1 (-1.40) (-0.62) (-2.12) D_AFS −0.037* −0.022* −0.079*** -1 (-1.66) (-1.70) (-3.07) TS 0.023 −0.260 0.197 -1 (0.04) (-0.58) (0.26) AFS −0.030 −0.079 −1.099*** -1 (-0.11) (-0.62) (-4.18) Ln(Day+1) 0.302*** 0.304*** 0.101*** 0.101*** 0.374*** 0.376*** (8.56) (8.60) (5.03) (5.05) (9.07) (9.15) Ln(Expy_firm+1) 0.001 0.007 0.021 0.023 0.008 0.010 (0.03) (0.16) (0.90) (0.96) (0.18) (0.21) Ln(Expy_ind+1) −0.057 −0.054 −0.085*** −0.084*** −0.090 −0.084 (-1.08) (-1.05) (-3.35) (-3.32) (-1.51) (-1.41) Ln(TA ) −0.010 −0.015 0.002 −0.001 0.012 0.004 -1 (-0.97) (-1.51) (0.35) (-0.09) (1.02) (0.35) Surprise −0.009** −0.010*** −0.001 −0.002 −0.013** −0.014** (-2.54) (-2.60) (-0.49) (-0.53) (-2.31) (-2.44) ROA −2.078*** −2.084*** −1.140*** −1.133*** −1.480*** −1.455*** -1 (-8.76) (-8.71) (-8.22) (-8.16) (-5.60) (-5.45) Change_ROA 0.509** 0.517** −0.011 −0.014 0.250 0.283 (2.22) (2.23) (-0.09) (-0.11) (0.94) (1.05) std_ROA 1.753*** 1.723*** 0.991*** 0.989*** 1.637*** 1.544*** (5.12) (5.02) (6.16) (6.11) (4.33) (4.05) Growth_ROA −0.041 −0.037 −0.430 −0.435 −0.295 −0.374 (-0.08) (-0.07) (-1.56) (-1.58) (-0.50) (-0.63) Constant −0.788*** −0.703** −0.114 −0.065 −1.780*** −1.655*** (-2.85) (-2.53) (-0.63) (-0.35) (-5.50) (-5.11) Year Fixed Yes Yes Yes Yes Yes Yes Industry Fixed Yes Yes Yes Yes Yes Yes Observations 1,883 1,883 1,883 1,883 1,883 1,883 Adjusted R 0.137 0.135 0.124 0.122 0.114 0.114 All variables are defined in Table 1. All continuous variables are winsorised at top and bottom 1%. The t-statistics in () are calculated with robust errors. ***, **, * indicate signif- icance at the 0.01, 0.05 and 0.10 levels, respectively, in two-tailed tests. The dependent variable in Columns 1 and 2 is Accy. The dependent variable in Columns 3 and 4 is Dis- persion. The dependent variable in Columns 5 and 6 is Bias. 308 Wei and Xue Finally, in Table 4, Columns 5 and 6, we examine how firms’ financial asset holdings affect analyst forecast errors. The average forecast error for TS-holding firms is 5.2% lower (about 25% of the average forecast error), relative to that of firms with- out TS or AFS. The forecast error for AFS-holding firms is 7.9% lower. The results in Table 4, Column 6, indicate that how much TS firms hold has no effect on analyst forecast errors. Different from Columns 2 and 4, the more AFS firms hold, the lower the average analyst forecast error. Referring to the control variables, the earlier the ana- lyst forecasts are made the larger is the average forecast error; average forecast error is lower if firms perform better last year; and the forecast error is larger if firms are more volatile. The results exhibited in Table 4 demonstrate that firms can use AFS to manip- ulate earnings and analysts are able to see through the opportunistic behaviour. Our empirical results indicate that the different subsequent measurements of TS and AFS have different effects on analyst forecasts. The results in Tables 3 and 4 are con- sistent with firms using AFS to manipulate earnings to meet or beat analyst forecasts. The results in Table 4 also demonstrate that analysts can, to an extent, see through firms’ opportunistic behaviour with AFS. 5. Additional tests 5.1. AFS sales and meeting or beating analyst forecasts We find that firms are more likely to meet or beat analyst forecasts when they hold AFS or they realise more gains from AFS. We take this as indirect evidence that firms manipulate earnings with AFS to meet or beat analyst forecasts. There is no direct evi- dence that firms sell AFS when their earnings fall below analyst forecasts. We designed the following model, equation (3), to examine whether firms sell AFS when earnings before AFS-realised gains and losses fall below analyst forecasts. RG AFS ¼ d þ d Diff þ b Control þ e (3) i;t 0 1 i;t k i;t k¼1 RG_AFS is realised gains and losses from AFS for firm i in year t, scaled by i,t beginning total assets. Diff =NIBAG -Forecast , where NIBAG (net income before i,t i,t i,t AFS-realised gains and losses) is earnings per share less realised gains and losses per share from AFS. Forecast is the consensus analyst forecast for firm i year t. Control i,t variables are the same as those in models (1) and (2). If firms sell AFS to meet or beat analyst forecasts, they will increase realised gains from AFS when their pre-managed earnings are below analyst forecasts. Thus, we predict that δ is negative and signifi- cant. The regression result is presented in Table 5. As shown there, δ is –0.025, signif- icant at a 1% α-level and consistent with our prediction. The results in Tables 3 and 5 show that firms do manipulate earnings to meet or beat analyst forecasts with AFS. 5.2. Self-selection bias As Table 2 shows, firms with financial assets are not significantly different from firms without financial assets. Our results may still face a self-selection problem. For instance, firms with more stable cash flows are more likely to hold financial assets. Meanwhile, the earnings of firms with more stable cash flows may be easier to predict. We use Heckman’s two-stage regression to control for possible self-selection bias. First, we run the following model, equation (4), to predict whether firms hold financial assets: China Journal of Accounting Studies 309 Table 5. Pre-managed earnings and realised gains and losses from AFS. RG_AFS Diff −0.025*** (–8.56) Ln(Day+1) 0.002 (0.99) Ln(Expy_firm+1) −0.005* (–1.71) Ln(Expy_ind+1) 0.005 (1.57) Ln(TA ) −0.001** -1 (–2.15) Surprise −0.000 (–1.20) ROA 0.019 -1 (0.94) Change_ROA 0.022 (0.98) std_ROA −0.013 (–0.58) Growth_ROA −0.046 (–1.48) Constant 0.012 (0.86) Year Fixed Yes Industry Fixed Yes Observations 601 Adjusted R 0.300 All variables are defined in Table 1. All continuous variables are winsorised at top and bottom 1%. The t- statistics in () are calculated with robust errors. ***, **, * indicate significance at the 0.01, 0.05 and 0.10 levels, respectively, in two-tailed tests. CFO NCA Lev i;t i;t1 i;t1 D FS ¼ a þ a Log TA þ a þ a þ a þ a ROA i;t 0 1 i;t1 2 3 4 5 i;t1 TA TA TA i;t1 i;t1 i;t1 þ a std ROA þ e 6 i;t1 i;t (4) D_FS is a dummy variable indicating whether firm i holds financial assets in year i,t t.If firm i holds financial assets in year t, D_FS equals 1; otherwise, D_FS i,t i,t equals 0. CFO is the net cash flow from firm i’s operation in year t. NCA is i,t i,t-1 firm i’s total non-current assets at the beginning of year t. Lev is firm i’s total i,t-1 liabilities at the beginning of year t, scaled by total assets. The other variables are as defined in model (1). Untabulated results show that the χ for our selection model is 65.32 and significant at a 1% α-level. We put the calculated inverse mills ratio in our models (1) and (2) and re-examine the regression results in Tables 3 and 4. The results in Table 6 are consistent with those in Tables 3 and 4 except that firms’ holdings of AFS are not significantly related to analyst forecast accuracy. In conclusion, the results in Table 6 demonstrate that our results still hold after controlling for self-selection bias. 310 Wei and Xue Table 6. Heckman two-stage regressions. Panel A: Financial assets and meet or beat analyst forecasts. MTBT MTBT MTBT MTBT (1) (2) (3) (4) D_TS 0.039 -1 (0.97) D_AFS 0.074* -1 (1.81) TS 0.255 -1 (0.23) AFS 0.454 -1 (1.39) D_URG_TS −0.083** (-2.40) D_RG_TS 0.113*** (3.29) D_RG_AFS 0.064** (2.08) URG_TS 4.668 (0.94) RG_TS 2.137 (0.39) RG_AFS 2.142*** (2.61) Control Yes Yes Yes Yes Year Fixed Yes Yes Yes Yes Industry Fixed Yes Yes Yes Yes Observations 1,834 1,834 1,834 1,834 Wald Chi 86.39*** 88.74*** 106.16*** 95.65*** (Continued) China Journal of Accounting Studies 311 Table 6. (Continued). Panel B: Financial assets and analyst forecast characteristics. Accy Accy Dispersion Dispersion Bias Bias (1) (2) (3) (4) (5) (6) D_TS −0.043 −0.033 −0.103* -1 (–1.25) (–1.51) (–1.83) D_AFS −0.044 −0.050** −0.122** -1 (–1.25) (–2.24) (–2.11) TS 0.275 −0.304 0.656 -1 (0.27) (–0.47) (0.42) AFS 0.220 0.045 −0.766* -1 (0.76) (0.25) (–1.71) Control Yes Yes Yes Yes Yes Yes Year Fixed Yes Yes Yes Yes Yes Yes Industry Fixed Yes Yes Yes Yes Yes Yes Observations 1,834 1,834 1,834 1,834 1,834 1,834 Wald Chi 134.16*** 129.25*** 168.10*** 162.03*** 62.73*** 65.95*** All variables are defined in Table 1. All continuous variables are winsorised at top and bottom 1%. The test-statistics in () are calculated with robust errors. ***, **, * indicate sig- nificance at the 0.01, 0.05 and 0.10 levels, respectively, in two-tailed tests. (a) The sample size is smaller for additional data requirements. 312 Wei and Xue 5.3. Financial assets and analyst experience So far, we have examined how firms’ financial asset holdings affect analyst forecast characteristics. Now we further examine whether financial asset holdings affect analyst characteristics. We change the dependent variables in model (2) to analyst firm and industry experience. The regression results are presented in Table 7. The results in Table 7 indicate that the analysts following TS-holding firms are less experienced in terms of following specific firms and industries. Including unrealised gains and losses from fair value changes adds to the difficulty of forecasting (Liang & Riedl, 2014). Experienced analysts face more reputation risk in following TS-holding firms and are therefore less likely to follow TS firms. The results in Table 7 also rule out the possibility that our results in Tables 3 and 4 are caused by analysts’ self-selection. 5.4. Financial market and litigation environment development The development levels among the various provinces in China are quite different. Thus, we can study how outside governance affects firms’ earnings management with AFS by examining how the financial market and litigation environment affect our results. Table 7. Financial assets and analyst experience. Exp_Firm Exp_Firm Exp_Ind Exp_Ind (1) (2) (3) (4) D_TS −0.056*** −0.043*** -1 (–3.88) (–3.38) D_AFS −0.013 0.005 -1 (–0.88) (0.39) TS −0.034 0.291 -1 (–0.07) (0.75) AFS −0.485*** −0.040 -1 (–2.72) (–0.28) Ln(TA ) 0.119*** 0.116*** 0.035*** 0.034*** -1 (21.75) (21.58) (6.95) (6.78) Surprise −0.004 −0.004 −0.003 −0.003 (–1.41) (–1.53) (–1.38) (–1.45) ROA 2.266*** 2.274*** 0.545*** 0.537*** -1 (18.22) (18.26) (5.66) (5.54) Change_ROA −0.705*** −0.679*** −0.113 −0.093 (–5.92) (–5.66) (–0.96) (–0.79) std_ROA −0.793*** −0.850*** −0.137 −0.181 (–4.44) (–4.74) (–0.86) (–1.14) Growth_ROA −2.318*** −2.373*** −0.140 −0.158 (–7.11) (–7.23) (–0.49) (–0.55) Constant −2.348*** −2.297*** 0.202* 0.218* (–19.13) (–18.80) (1.74) (1.87) Year Fixed Yes Yes Yes Yes Industry Fixed Yes Yes Yes Yes Observations 1,883 1,883 1,883 1,883 Adjusted R 0.329 0.326 0.307 0.303 All variables are defined in Table 1. All continuous variables are winsorised at top and bottom 1%. The t-statistics in () are calculated with robust errors. ***, **, * indicate significance at the 0.01, 0.05 and 0.10 levels, respectively, in two-tailed tests. Dependent variables are analyst firm-specific experience LnðExp firm þ 1Þ and analyst industry-specific experience LnðExp ind þ 1Þ. China Journal of Accounting Studies 313 As discussed previously, if firms manipulate earnings with AFS to meet or beat analyst forecasts, and if analysts can see through firms’ opportunistic behaviour, the forecast accuracy (dispersion) is higher (smaller) when firms hold AFS. If the finan- cial market in which firms are located is better developed, (1) firms care more about their capital market performance, are more likely to manipulate earnings to meet or beat analyst forecasts and are more likely to use AFS to manage earnings; (2) analysts are also more likely to see through firms’ earnings management with AFS. We predict that the effect of AFS holdings on analyst forecast accuracy and dispersion is stronger when firms are located in provinces where the financial mar- kets are better developed. How the development of litigation environments affects our results is unclear. On the one hand, investor protection is better where the liti- gation environment is better developed and firms are less likely to manipulate earn- ings. On the other hand, firms face more litigation risk in accrual earnings management when they are located in provinces with a better developed litigation environment, and they are more likely to use real activities to manage earnings (Zang, 2011). In this case, firms are more likely to use AFS to manage earnings. We add two dummy variables, FINMKT and LAW and their cross term with Finan- cial Instrument / Financial Instrument Gains , in models (1) and (2) to test i,t-1 i,t-1 how development of the financial market and litigation environment affect our results. We use marketisation indexes (Fan, Wang, Zhang & Zhu 2003) to measure financial market and litigation environment development. We rank all provinces in the same year. If the financial market (litigation environment) development level is above the median, then FINMKT (LAW) is equal to one; otherwise, it is equal to zero. We report our results in Table 8 and, for brevity, we do not report the coeffi- cients of control variables in Table 8. As shown in Table 8, Panel A, Column 1, the coefficient of D_AFS is 0.525 -1 (positive and significant), which shows that firms manipulate earnings to meet or beat analyst forecasts with AFS. The cross term D_AFS *FINMKT is negative but -1 not significantly different from zero. Similarly, in Table 8, Column 2, the coefficient of D_AFS *LAW is also negative and insignificant. Because FINMKT and LAW are -1 highly correlated, we put them together in Column 3. The results remain qualita- tively the same as in Columns 1 and 2. We use the dummy variable D_RG_AFS in Table 8 Columns 4 and 6. The results show that the coefficient of D_RG_AFS is no longer significant. The results in Table 8, Panel A, indicate that financial market and litigation environment development do not affect firms’ earnings management with AFS. In Table 8, Panel B, we re-examine our results in Table 4. The coefficient of D_AFS *FINMKT is negative and significant when the dependent variable is fore- -1 cast accuracy or dispersion, and it is not significantly different from zero when the dependent variable is forecast error. The results also show that the coefficient of D_AFS- *LAW is positive and significant when we control for financial market development. These results, together with those in Table 8, Panel A, demonstrate that financial market development does not affect firms’ decisions to manipulate earnings with AFS, and analysts are more likely to see through firms’ earnings management with AFS when the financial market is better developed. In contrast, firms are less likely to manipulate earnings with AFS when the litigation environ- ment is better developed. In conclusion, outside governance cannot moderate firms’ earnings management with AFS. 314 Wei and Xue Table 8. Financial market and litigation environment development. Panel A: For meet or beat analyst forecasts. MTBT MTBT MTBT MTBT MTBT MTBT D_TS −0.002 −0.068 −0.073 -1 (–0.01) (–0.22) (–0.23) D_AFS 0.525* 0.678** 0.663* -1 (1.69) (2.09) (1.90) D_TS *FINMKT 0.067 0.010 -1 (0.21) (0.02) D_AFS *FINMKT −0.316 0.002 -1 (–0.94) (0.00) D_TS *LAW 0.136 0.130 -1 (0.41) (0.29) D_AFS *LAW −0.499 −0.490 -1 (–1.42) (–1.08) D_URG_TS −0.338 −0.208 −0.320 (–0.87) (–0.54) (–0.76) D_RG_TS 0.220 −0.041 0.040 (0.65) (–0.12) (0.11) D_RG_AFS 0.277 0.510 0.402 (0.72) (1.38) (0.99) D_URG_TS*FINMKT 0.048 0.432 (0.11) (0.76) D_RG_TS*FINMKT 0.371 −0.392 (0.99) (–0.73) D_RG_AFS*FINMKT 0.039 0.534 (0.10) (0.88) D_URG_TS*LAW −0.165 −0.461 (–0.39) (–0.83) D_RG_TS*LAW 0.731** 1.024** (1.99) (1.99) D_RG_AFS*LAW −0.228 −0.648 (–0.58) (–1.10) China Journal of Accounting Studies 315 LAW 0.440** 0.329 0.226 0.160 (2.47) (1.31) (1.25) (0.65) FINMKT 0.394** 0.159 0.194 0.093 (2.21) (0.63) (1.10) (0.38) Constant 5.209*** 4.986*** 5.023*** 5.493*** 5.496*** 5.545*** (3.26) (3.12) (3.13) (3.43) (3.43) (3.44) Control Yes Yes Yes Yes Yes Yes Year Fixed Yes Yes Yes Yes Yes Yes Industry Fixed Yes Yes Yes Yes Yes Yes Observations 1,883 1,883 1,883 1,883 1,883 1,883 Pseudo R 0.076 0.077 0.077 0.084 0.087 0.088 Panel B: For analyst forecast characteristics. Accy Accy Accy Dispersion Dispersion Dispersion Bias Bias Bias D_TS −0.054 −0.018 −0.032 0.033 0.030 0.042 −0.107* −0.045 −0.072 -1 (–1.15) (–0.36) (–0.61) (1.18) (1.04) (1.37) (–1.95) (–0.80) (–1.21) D_AFS 0.057 −0.076 −0.013 0.042 −0.039 0.002 −0.065 −0.186*** −0.143** -1 (1.06) (–1.33) (–0.23) (1.36) (–1.16) (0.07) (–1.05) (–2.83) (–2.04) D_TS *FINMKT 0.035 0.066 −0.048 −0.044 0.073 0.122 -1 (0.65) (0.92) (–1.55) (–1.05) (1.19) (1.47) D_AFS *FINMKT −0.106* −0.208*** −0.071** −0.139*** −0.008 −0.137 -1 (–1.84) (–2.82) (–2.10) (–3.23) (–0.12) (–1.61) D_TS *LAW −0.015 −0.058 −0.046 −0.013 −0.006 −0.089 -1 (–0.28) (–0.79) (–1.44) (–0.31) (–0.10) (–1.05) D_AFS *LAW 0.049 0.182** 0.024 0.115** 0.136* 0.220** -1 (0.80) (2.31) (0.68) (2.51) (1.92) (2.43) LAW −0.022 −0.011 −0.022 −0.033 −0.089*** −0.050 (–0.76) (–0.28) (–1.31) (–1.38) (–2.70) (–1.06) FINMKT −0.025 −0.016 −0.010 0.014 −0.092*** −0.054 (–0.88) (–0.38) (–0.62) (0.59) (–2.77) (–1.13) Constant −0.796*** −0.761*** −0.744*** −0.135 −0.100 −0.100 −1.735*** −1.691*** −1.669*** (–2.91) (–2.77) (–2.71) (–0.84) (–0.62) (–0.62) (–5.51) (–5.35) (–5.29) (Continued) 316 Wei and Xue Table 8. (Continued). Panel B: For analyst forecast characteristics. Control Yes Yes Yes Yes Yes Yes Yes Yes Yes Year Fixed Yes Yes Yes Yes Yes Yes Yes Yes Yes Industry Fixed Yes Yes Yes Yes Yes Yes Yes Yes Yes Observations 1,883 1,883 1,883 1,883 1,883 1,883 1,883 1,883 1,883 Adjusted R 0.139 0.136 0.140 0.129 0.125 0.131 0.117 0.117 0.119 All variables are defined in Table 1. All continuous variables are winsorised at top and bottom 1%. The test-statistics in () are calculated with robust errors. ***, **, * indicate sig- nificance at the 0.01, 0.05 and 0.10 levels, respectively, in two-tailed tests. FINMKT equals 1 if firms are located in provinces where financial market development levels are above the median level; zero otherwise. LAW equals 1 if firms are located in provinces where litigation environment development levels are above the median level; zero otherwise. China Journal of Accounting Studies 317 5.5. Year fixed effect and industry fixed effect In Table 3, the coefficients of control variables are not significantly different from zero except for that of Ln(Day+1). The reason is that we control both year and industry fixed effects in our regressions. We also run the regressions without fixed effects. Untabulated results show that firms are less likely to meet or beat analyst forecasts when the change in return on assets (ROA) is larger, and firms with higher growth are more likely to meet or beat analyst forecasts. 6. Conclusion This paper examines how firms’ financial asset holdings affect analyst forecasts, and provides empirical evidence to standard setters that subsequent measurements of finan- cial assets can affect investors’ valuations. We find firms are more likely to meet or beat analyst forecasts when they hold AFS. We also find that realised gains and losses from AFS are positively related to the likelihood of meeting or beating analyst fore- casts. However, firms’ TS holdings (gains and losses from TS) have no significant effects on firms’ likelihood to meet analyst forecasts. We also find that analyst forecast accuracy (dispersion) [error] is higher (smaller) [smaller] when firms hold AFS. These results show that allowing cumulative unrealised gains and losses from AFS to recycle into net income leaves room for firms to manage earnings with AFS. Our results also demonstrate that analysts can, to some extent, see through firms’ earnings management with AFS. Finally, we find that analysts are more likely to see through firms’ earnings management with AFS when the firms are located in areas where the financial market is better developed. Firms are less likely to use AFS to manage earnings when they are located in areas with litigation environments that are better developed. Our results show that prohibiting the recycling of cumulative unrealised gains and losses from AFS into net income can moderate firms’ opportunistic behaviour. Our results suggest that China’s standard setters should converge with IFRS 9 to protect investors. Acknowledgements We appreciate the comments from anonymous referees and editors, and from Liansheng Wu and Jason Xiao, which significantly improved this paper. The authors are responsible for any remaining errors. Disclosure statement No potential conflict of interest was reported by the authors. Funding Financial support from the National Natural Science Fund (Project Nos. 71322201 and 71272025) is gratefully acknowledged. Notes 1. The Youngor case discussed in Section 1 demonstrates that realised gains from AFS can greatly influence firms’ performance. Gu, Wang and Xue (2015) find that 23.45% of firms in their sample avoid losses by selling AFS. 2. Managers can choose their timing to sell TS to realise more gains. 3. For example, airline companies and oil companies sign future contracts. 318 Wei and Xue References Barth, M. E. (1994). Fair value accounting: Evidence from investment securities and the market valuation of banks. Accounting Review, 69,1–25. Barth, M. E., Kasznik, R., & McNichols, M. F. (2001). Analyst coverage and intangible assets. Journal of Accounting Research, 39,1–34.10.1111/1475-679X.00001 Bartov, E. (1993). The timing of asset sales and earnings manipulation. Accounting Review, 68, 840–855. Bernard, V. L., & Skinner, D. J. (1996). What motivates managers’ choice of discretionary accru- als? Journal of Accounting and Economics, 22, 313–325.10.1016/S0165-4101(96)00431-4 Burgstahler, D., & Dichev, I. (1997). Earnings management to avoid earnings decreases and losses. Journal of Accounting and Economics, 24,99–126.10.1016/S0165-4101(97)00017-7 Burgstahler, D. C., & Eames, M. J. (2003). Earnings management to avoid losses and earnings decreases: Are analysts fooled?* Contemporary Accounting Research, 20, 253–294.10.1506/ BXXP-RGTD-H0PM-9XAL Byard, D., Li, Y., & Yu, Y. (2011). The effect of mandatory IFRS adoption on financial analysts’ information environment. Journal of Accounting Research, 49,69–96.10.1111/joar.2011.49. issue-1 Dechow, P. M., Sloan, R. G., & Sweeney, A. P. (1995). Detecting earnings management. Accounting Review, 70, 193–225. Degeorge, F., Patel, J., & Zeckhauser, R. (1999). Earnings management to exceed thresholds. Journal of Business, 72(1), 1–33.10.1086/jb.1999.72.issue-1 Dong, M., Ryan, S., & Zhang, X. J. (2009). Historical-cost or fair-value accounting: Analysis of the reclassification of unrealized holding gains and losses for marketable securities. University of Lausanne and University of California (Berkeley) Working Paper. Eccher, E. A., Ramesh, K., & Thiagarajan, S. R. (1996). Fair value disclosures by bank holding companies. Journal of Accounting and Economics, 22,79–117.10.1016/S0165-4101(96) 00438-7 Ecker, F., Francis, J., Olsson, P., & Schipper, K. (2011). “Peer firm selection for discretionary accruals models.” Unpublished paper, Duke University. Fan, G., Wang, X. L., Zhang, L. W., & Zhu, H. P. (2003). Marketion Index for China’s Provinces. Economic Research Journal, (3), 1–1. (In Chinese). Gu, Z., Y. Wang, & J. Xue. (2015). A forward looking decision: Earnings management by selling available for sale securities. Working paper. He, X., Wong, T. J., & Young, D. (2012). Challenges for implementation of fair value accounting in emerging markets: Evidence from China. Contemporary Accounting Research, 29, 538–562.10.1111/care.2012.29.issue-2 Herrmann, D., Inoue, T., & Thomas, W. B. (2003). The sale of assets to manage earnings in Japan. Journal of Accounting Research, 41,89–108.10.1111/joar.2003.41.issue-1 Hirst, D. E., & Hopkins, P. E. (1998). Comprehensive income reporting and analysts’ valuation judgments. Journal of Accounting Research, 36,47–75.10.2307/2491306 Hunton, J. E., Libby, R., & Mazza, C. L. (2006). Financial reporting transparency and earnings management. Accounting Review, 81, 135–157.10.2308/accr.2006.81.1.135 Jones, J. J. (1991). Earnings management during import relief investigations. Journal of Account- ing Research, 29, 193–228.10.2307/2491047 Kothari, S. P. (2001). Capital markets research in accounting. Journal of Accounting and Economics, 31, 105–231.10.1016/S0165-4101(01)00030-1 Lang, M. H., & Lundholm, R. J. (1996). Corporate disclosure policy and analyst behavior. Accounting Review, 71, 467–492. Liang, L., & Riedl, E. J. (2014). The effect of fair value versus historical cost reporting model on analyst forecast accuracy. Accounting Review, 89, 1151–1177.10.2308/accr-50687 Liu, C. C., & Ryan, S. G. (2006). Income smoothing over the business cycle: Changes in banks’ coordinated management of provisions for loan losses and loan charge-offs from the pre- 1990 bust to the 1990s boom. Accounting Review, 81, 421–441.10.2308/accr.2006.81.2.421 Matsumoto, D. A. (2002). Management’s incentives to avoid negative earnings surprises. Accounting Review, 77, 483–514.10.2308/accr.2002.77.3.483 McNichols, M., & Wilson, G. P. (1988). Evidence of earnings management from the provision for bad debts. Journal of Accounting Research, 26,1–31. China Journal of Accounting Studies 319 Poitras, G., Wilkins, T., & Kwan, Y. S. (2002). The timing of asset sales: Evidence of earnings management? Journal of Business Finance & Accounting, 29, 903–934. Roychowdhury, S. (2006). Earnings management through real activities manipulation. Journal of Accounting and Economics, 42, 335–370.10.1016/j.jacceco.2006.01.002 Tan, H., Wang, S., & Welker, M. (2011). Analyst following and forecast accuracy after mandated IFRS adoptions. Journal of Accounting Research, 49, 1307–1357.10.1111/joar.2011.49.issue- Tarca, A., Hancock, P., Woodliff, D., Brown, P., Bradbury, M., & Van Zijl, T. (2008). Identifying decision useful information with the matrix format income statement. Journal of International Financial Management & Accounting, 19(2), 184–217. Teoh, S. H., Wong, T. J., & Rao, G. R. (1998). Are accruals during initial public offerings opportunistic? Review of Accounting Studies, 3, 175–208.10.1023/A:1009688619882 Wang, H., Li, Q., & Chen, Y. (2015). Earnings management, business cycle, and product market competition. China Journal of Accounting Studies, 3, 136–157.10.1080/21697213.2015. Wang, Y. T., Xue, J., & Li, L. (2011). Does fair value measurement model have value relevance? Empirical evidence from financial assets investigation. China Accounting Review, (4), 383–398. (In Chinese). Xu, J. C., & Zeng, X. Y. (2013). The presentation format of comprehensive income and the value-relevance of fair value based on the available for sale investments. Accounting Research, (1), 20–27. (In Chinese). Ye, J. F., Zhou, L., Li, D. M., & Guo, L. (2009). Management motivation, accounting policy options and earnings management – An empirical study on the classification of financial assets for listed companies under the new accounting standards. Accounting Research, (3), 25–30. (In Chinese). Zang, A. Y. (2011). Evidence on the trade-off between real activities manipulation and accrual-based earnings management. Accounting Review, 87, 675–703.
Journal
China Journal of Accounting Studies
– Taylor & Francis
Published: Oct 2, 2015
Keywords: analyst forecasts; earnings management; fair value accounting; financial assets
