Voluntary disclosure of balance sheet information in quarterly earnings announcements $

Transcription

1 Journal of Accounting and Economics 33 (2002) Voluntary disclosure of balance sheet information in quarterly earnings announcements $ Shuping Chen a, Mark L. DeFond b, *, Chul W. Park c a School of Business, University of Washington, Seattle, WA , USA b Leventhal School of Accounting, University of Southern California, Los Angeles, CA , USA c Hong Kong University of Science and Technology, Clear Water Bay, Kowloon, Hong Kong Received 22 August 2000; received in revised form 23 October 2001 Abstract We investigate a pervasive voluntary disclosure practice managers including balance sheets with quarterly earnings announcements. Consistent with expectations, we find that managers voluntarily disclose balance sheets when current earnings are relatively less informative, or when future earnings are relatively more uncertain. Specifically, balance sheet disclosures are more likely among firms: (1) in high technology industries; (2) reporting losses; (3) with larger forecast errors; (4) engaging in mergers or acquisitions; (5) that are younger; and (6) with more volatile stock returns. This is consistent with managers disclosing balance sheets in response to investor demand for value relevant information to supplement earnings. r 2002 Elsevier Science B.V. All rights reserved. JEL classification: M41; D82; G12 Keywords: Capital market; Voluntary disclosure 1. Introduction Empirical research investigating voluntary management disclosure tends to focus on earnings disclosures, management forecasts, and to a lesser extent, overall $ The authors appreciate helpful comments from Mingyi Hung, S.P. Kothari (the editor), K.R. Subramanyam, and an anonymous referee. We also would like to thank Stan Levine (of First Call) for data, and Ming He Bai and Yuan Zhang for their help in collecting data. *Corresponding author. Tel.: ; fax: address: defond@marshall.usc.edu (M.L. DeFond) /02/$ - see front matter r 2002 Elsevier Science B.V. All rights reserved. PII: S (02)

2 230 S. Chen et al. / Journal of Accounting and Economics 33 (2002) disclosure levels. 1 However, a relatively pervasive voluntary disclosure practice that is not investigated in prior research is management s inclusion of balance sheet information along with quarterly earnings announcements. The purpose of this study is to describe the characteristics of voluntary balance sheet disclosures, and to investigate management s incentives to include balance sheet information in their quarterly earnings announcements. We examine all quarterly earnings announcements included in the Wall Street Journal ProQuest database for the 12 quarters ending with the third quarter of Our analysis finds that 52% of the 2,551 firms in our sample include a balance sheet in at least one quarterly earnings announcement, and that 37% of the 23,086 quarterly earnings announcements we identify include a balance sheet. We also find that the percentage of quarterly earnings announcements containing balance sheets grows from 31% to 46% over the period we analyze. In addition, once firms begin including balance sheets in their quarterly earnings announcements they tend to continue, with just 35% ceasing to include balance sheets once they initiate disclosure. Therefore, we conclude that including balance sheets in earnings announcements is a pervasive voluntary disclosure practice that is growing over the period we examine. We argue that managers have incentives to voluntarily include balance sheet information along with quarterly earnings announcements when current earnings are relatively less informative, or when future earnings are relatively more uncertain. In these settings investors are likely to have a greater demand for additional value relevant information such as balance sheets to help assess firm value. We test our conjecture by developing several hypotheses that identify settings in which investors are likely to find current earnings relatively less informative, or future earnings relatively more uncertain. Consistent with our predictions, a logit analysis finds that managers are more likely to disclose balance sheet information along with quarterly earnings announcements for firms: (1) in high technology industries; (2) reporting losses; (3) with larger forecast errors; (4) engaging in mergers or acquisitions; (5) that are younger; and (6) with more volatile stock returns. These findings are consistent with managers disclosing balance sheet information when investors demand additional value relevant information to help assess firm value. To investigate the robustness of our logit results we also examine the price earnings and the returns earnings associations of firms with and without voluntary balance sheet disclosures. Comparing the R 2 in regressions of quarterly earnings on share price across disclosure and non-disclosure observations indicates that earnings explain a significantly lower proportion of share value among firms that make balance sheet disclosures. This is consistent with firms supplementing their earnings disclosures with balance sheet information when earnings are less value-relevant. However, comparing the R 2 in regressions of quarterly earnings on stock returns across disclosure and non-disclosure observations indicates that there is no difference in the explanatory power of earnings across the two groups. Thus, our 1 For example, Skinner (1994), Frankel et al. (1995), McNichols and Wilson (1995), Lang and Lundholm (1996), Botosan (1997), Tasker (1998), and Frankel et al. (1999).

3 S. Chen et al. / Journal of Accounting and Economics 33 (2002) market tests find mixed evidence that balance sheet disclosures are made when earnings are relatively less informative. Our study contributes to the literature that examines voluntary management disclosures and the usefulness of balance sheet information in valuing securities. While theory and empirical evidence suggest that the balance sheet is an important source of value relevant information, and a large body of literature documents that share price is associated with balance sheet information, prior research provides little evidence on whether firms are voluntarily forthcoming with balance sheet information when it is most likely to be useful beyond earnings information. We shed light on this issue by investigating firms voluntary balance sheet disclosures and provide evidence consistent with managers making these disclosures in response to investor demand for additional value-relevant information to supplement reported earnings. The next section motivates our hypotheses, Section 3 presents our research design, and Section 4 discusses our sample and presents our hypotheses tests. Section 5 presents an analysis of market variables, Section 6 presents robustness checks and Section 7 summarizes our findings. 2. Hypothesis development 2.1. The demand for balance sheet information to supplement reported earnings Dye (1985) argues that managers have incentives to make voluntary accounting disclosures when market participants find the disclosures useful in assessing firm value. 2 We expect investors to find voluntary balance sheet disclosures relatively more useful in assessing firm value when current earnings are less informative, or when future earnings are more uncertain. In these settings, investors are likely to demand additional value relevant disclosures to supplement the information contained in earnings. For example, because current earnings are less informative when firms report losses, investors are more likely to demand additional value relevant disclosure to help assess firm value. Similarly, because future earnings are more uncertain among firms whose operations are less predictable (such as younger firms), investors are more likely to demand additional disclosures when they evaluate younger firms (Lang, 1991). A logical management response to these demands is to disclose balance sheet information because balance sheets provide value relevant information that compliments earnings information (Ou and Penman, 1989; Ohlson, 1995; Barth et al., 1998). Thus, our overall prediction is that managers are more likely to voluntarily disclose balance sheet information in settings where current earnings are relatively less informative, or when future earnings are relatively more uncertain. A setting in which we expect current earnings to be less informative and future earnings more uncertain is in high technology industries. Firms in high technology 2 See Verrecchia (2001) for a review of the theoretical literature on voluntary disclosure.

4 232 S. Chen et al. / Journal of Accounting and Economics 33 (2002) industries tend to make large investments in intangibles such as research and development, human capital, and brand-name development. In this setting earnings from traditional accounting models are likely to be less informative (Collins et al., 1997; Francis and Schipper, 1999; Lev and Zarowin, 1999). High-tech firms also operate in rapidly changing environments that make their future operations, and hence future earnings, relatively more uncertain. While balance sheet information is also problematic in valuing intangibles and in resolving future uncertainty, analysts find various balance sheet accounts particularly useful in valuing high technology companies. For example, cash balances are important in assessing the ability of high technology companies to enter new markets, to make new product launches, and to survive until the next round of financing (Bank, 1999; Scott, 2000). Similarly, inventory and receivables management is particularly critical for these firms due to the uncertainty of their operating environment and the untried nature of their products and customer base (Palazzo, 1999; Ramstad, 1996; McGough, 2000). Therefore, because we expect high-tech firms current earnings to be less informative and their future earnings to be more uncertain, and because investors find balance sheet information particularly useful in assessing their performance, our first hypothesis is (in alternative form): Hypothesis 1. Managers of firms in high technology industries are more likely to disclose balance sheet information in their quarterly earnings announcements. We also expect that managers are more likely to disclose balance sheet information when firms report losses. In the presence of a loss, earnings fail in their traditional role as an indicator of future earnings (Hayn, 1995; Collins et al., 1997). Moreover, because losses cannot be sustained indefinitely, firms experiencing losses are more likely to liquidate, making their abandonment value more relevant in assessing shareholder value. This makes balance sheet information relatively more useful because prior research demonstrates that firms exit values can be reasonably expressed as a weighted average of balance sheet items (Berger et al., 1996; Collins et al., 1999). Hence, our second hypothesis is (in alternative form): Hypothesis 2. Managers of firms reporting losses are more likely to disclose balance sheet information in their quarterly earnings announcements. Balance sheet information is likely to be useful in interpreting the valuation implications of earnings when reported earnings differ from market expectations. In this setting managers are likely to have incentives to guide market participants in understanding why earnings diverge from expectations, as well as the valuation implications of the divergence. Balance sheet disclosures can provide this guidance because balance sheet accounts can be useful in interpreting reported earnings (McGough and Podd, 1999). For example, working capital accounts provide investors with value relevant information about the nature of reported accruals (DeFond and Park, 2001). Thus, our third hypothesis is (in alternative form):

5 S. Chen et al. / Journal of Accounting and Economics 33 (2002) Hypothesis 3. Managers of firms with quarterly earnings that deviate from analysts forecasts are more likely to disclose balance sheet information in their quarterly earnings announcements. Investors are also likely to have a relatively greater demand for balance sheet information when firms engage in merger and acquisition activity. Mergers and acquisitions are likely to impact firms future operating activities, which in turn are likely to increase the uncertainty related to their future earnings. Balance sheet disclosure in this setting is likely to be useful in helping investors assess the impact of the merger and acquisition activity on future earnings. For example, the total asset number can be used to predict the normal component of future earnings (Ohlson, 1995). Thus, our fourth hypotheses is (in alternative form): Hypothesis 4. Managers of firms that engage in mergers or acquisitions during the quarter are more likely to disclose balance sheet information in their quarterly earnings announcements. Another factor that is likely to impact the demand for value relevant information is the firm s age. We expect greater uncertainty about younger firms earnings because their production activities are likely to be less predictable. This is consistent with Lang (1991) who argues that firms with greater uncertainty about future earnings such as younger firms are likely to reap greater benefits from additional disclosure. Therefore, our fifth hypothesis is (in alternative form): Hypothesis 5. Managers of younger firms are more likely to disclose balance sheet information in their quarterly earnings announcements. Stock return volatility is also likely to be associated with balance sheet disclosures. High stock return volatility is consistent with greater uncertainty about future earnings, because stock price is a function of expected future earnings. Since investors are likely to have a greater demand for information when future earnings are more uncertain, we expect that managers of these firms have greater incentives to voluntarily disclose additional value relevant information. Thus, our sixth hypotheses is (in alternative form): Hypothesis 6. Managers of firms with more volatile stock returns are more likely to disclose balance sheet information in their quarterly earnings announcements Proprietary costs and balance sheet disclosures Verrecchia (1983) analyzes voluntary disclosure in the context of accounting information and argues that full voluntary disclosure will not always occur. He demonstrates that when private information disclosure results in proprietary costs, the market is likely to interpret non-disclosure with less suspicion because the costs

6 234 S. Chen et al. / Journal of Accounting and Economics 33 (2002) of disclosure can exceed the benefits to shareholders when proprietary costs are sufficiently large. This suggests that consideration of proprietary costs may reduce management incentives to make voluntary balance sheet disclosures. However, Verrecchia (1983) also observes that management decisions to make accounting disclosures are typically not decisions of disclosure versus non-disclosure, but rather decisions of accelerated versus delayed disclosure. This is the case with respect to the balance sheet disclosures examined in this study because managers are required to disclose balance sheet information in their 10-Q and 10-K filings. Thus, the disclosure decision is whether to voluntarily include balance sheet information in quarterly earnings announcements that are made on average about 1 month in advance of the mandatory 10-Q and 10-K filings. 3 In addition, the balance sheet information disclosed with earnings announcements does not include detailed footnote information (see Appendix A). Therefore, we believe it is unlikely that proprietary costs are a consideration in management s decision to disclose balance sheet information in quarterly earnings announcements. 3. Research design We test our hypotheses by estimating the coefficients in the following logit regression: Balance Sheet Disclosure Indicator ¼ a 0 þ a 1 ðhigh-tech DummyÞþa 2 ðloss DummyÞ þ a 3 ðabsolute Forecast Error DummyÞ þ a 4 ðmergers & Acquisitions DummyÞ þ a 5 ðageþþa 6 ðreturn VolatilityÞ þ a 7 ðlogðmarket ValueÞÞ þ a 8 ðanalyst CoverageÞ þ a 9 ðmarket-to-book RatioÞþe; ð1þ where Balance Sheet Disclosure Indicator=1 if the quarterly earnings announcement includes balance sheet information and 0 otherwise. High-Tech Dummy=1 if the firm reports Compustat SIC codes (Drugs), (R&D services), (programming), (computers), (electronics), or (precise measurement instruments), and 0 otherwise. Loss Dummy=1 if quarterly net income before extraordinary items (from Compustat) is negative during the current quarter and 0 otherwise. Absolute Forecast Error Dummy=1 if the absolute value of the forecast error (defined as reported earnings minus the most recent consensus mean analysts 3 Based on a sample of 591 observations drawn from our test sample, the earnings announcement precedes the required 10-Q or 10-K disclosure by a mean (median) of 29 (26) days.

7 S. Chen et al. / Journal of Accounting and Economics 33 (2002) forecast from the First Call database) is larger than one cent during the current quarter, and 0 otherwise. Mergers &Acquisitions Dummy=1 if the firm reports merger or acquisition activity during the current quarter (as reported in Compustat), and 0 otherwise. Age=the year of the current quarterly announcement minus the first year the firm is publicly traded (according to the CRSP database). Return Volatility=the standard deviation of stock returns over the prior 250 days, where at least 100 days of stock returns are required for inclusion in the sample. log(market Value)=natural log of the firm s market value at the end of the current quarter. Analyst Coverage=number of analysts following (from the First Call database) at the end of the current quarter. Market-to-Book Ratio=ratio of market-to-book value at the end of the current quarter. Forecast errors are computed using forecasts and reported earnings in the First Call database, and financial statement variables (aside from earnings used to compute forecast errors) are obtained from the Compustat database. 4 The SIC classification used to identify high technology industries is consistent with Kasznik and Lev (1995). Log(Market Value), Analyst Coverage, and the Market-to-Book Ratio are included as control variables because prior research finds that they influence firms discretionary disclosure decisions (Lang and Lundholm, 1993; Kasznik and Lev, 1995; Tasker, 1998; Frankel et al., 1999). The Market-to-Book Ratio is also expected to capture growth prospects, a potentially correlated omitted variable with respect to the High Tech Dummy variable. 4. Data and empirical results 4.1. Initial sample of all firms with earnings announcements We initially identify all quarterly earnings announcements during the 12 fiscal quarters from fourth quarter 1992 through third quarter 1995 included in the ProQuest Wall Street Journal Abstracts database. 5 We identify firms that include balance sheets in their press releases using the Academic Universe database. 4 We also use the split factors in the First Call database to put all variables on an equivalent per-share basis at the time of the earnings announcement. However, some split factors in earlier periods are missing. We talked with the director of quantitative research at First Call who confirmed that the split factors in the First Call database are essentially identical to those in the quarterly Compustat database. Thus, when a firm s split factors are missing in the First Call database, we use the appropriate split factors from Compustat. 5 All firms do not report quarterly earnings in each of the 12 quarters because of newly entering firms, exiting firms, some firms not always making press releases, and earnings announcements missing from the ProQuest database. Restricting our sample to firms with complete disclosure for each of the 12 quarters greatly reduces the sample size and induces a survivorship bias.

8 236 S. Chen et al. / Journal of Accounting and Economics 33 (2002) Table 1 presents descriptive information on the pattern of disclosure among our sample firms and the appendix presents examples of balance sheet disclosures from two earnings announcements included in our sample. Panel A of Table 1 presents the number of firms and observations with earnings announcements, in total and for each of the 12 quarters we examine, along with the proportion of balance sheet disclosures. The first column reports that of the 2,551 firms in our sample, 52% (1,328) make at least one balance sheet disclosure. The second column reports that of the 23,086 quarterly announcements in our sample, 37% (8,641) include balance sheet disclosures. The remaining columns of Panel A report that the number of firms with earnings announcements range from 1,653 to 2,059 over the 12 quarters we examine, with the proportion of balance sheet disclosures increasing from 31% to 46%. Thus, panel A of Table 1 indicates that over half of the firms in our sample make at least one balance sheet disclosure, and that the proportion of firms disclosing balance sheets increases by approximately 50% over the period we analyze. While firms in high technology industries constitute only 30% of the firms in our sample, they make 43% of the balance sheet disclosures. 6 Given the disproportionately large representation by firms in high technology industries, we separately investigate how this group influences the increasing trend in balance sheet disclosures. Thus, Panel B of Table 1 separately analyzes the high technology firms in our sample. The first column of Panel B reports that 63% of these firms include at least one balance sheet disclosure. Though not reported directly in the table, this compares with 48% of the sample firms in non-high technology industries. Panel B also reports that the relative frequency of high technology firms making balance sheet disclosures is increasing over the period we analyze, from 47% during fourth quarter 1992 to 64% during third quarter However, firms in non-high technology industries also increase their relative disclosure frequency, from 24% during fourth quarter 1992 to 38% during third quarter 1995 (not reported directly in the table). Thus, although high technology firms are more inclined to disclosing balance sheet information throughout the period analyzed, their presence does not solely drive the increasing relative frequency found in Panel A. The analysis in Panel C of Table 1 examines the likelihood that a given firm will disclose balance sheet information by presenting the distribution of the ratio of each firm s total balance sheet disclosures divided by its total earnings announcements. This analysis indicates that the mean and median percentage of each firms earnings announcements that contain balance sheets is 37% and 9%, respectively; and the upper and lower quartiles are 88% and 0%, respectively. Panel D of Table 1 reports the number of breaks in the disclosure sequences of firms that include at least one balance sheet disclosure. A break in disclosure occurs when a firm reports an earnings announcement that does not include a balance sheet, in a quarter immediately following a quarter that does include a balance sheet. This 6 While not directly reported in a table, these proportions can be calculated from the data in panels A and B of Table 1. The proportion of high technology firms=30%=764c2,551. The proportion of high technology firms with balance sheet disclosures to the total number of firms with balance sheet disclosures=43%=(55% * 6,669)C(37% * 23,086).

9 Table 1 Descriptive analysis of the number of earnings announcements, and the proportion that includes voluntary balance sheet disclosures, during the period fourth quarter 1992 through third quarter 1995 Panel A: Analysis of entire sample Number of earnings announcements Proportion that discloses balance sheets (%) Total firms Panel B: Analysis of High-Tech firms Number of High-Tech firms earnings announcements Proportion of High-Tech firms that discloses balance sheets(%) Total observations Observations/firms per quarter Qtr 4 Qtr 1 Qtr 2 Qtr 3 Qtr 4 Qtr 1 Qtr 2 Qtr 3 Qtr 4 Qtr 1 Qtr 2 Qtr 3 2,551 23,086 1,653 1,726 1,797 1,852 1,921 1,952 2,036 2,023 2,041 2,059 2,025 2, , S. Chen et al. / Journal of Accounting and Economics 33 (2002)

10 238 Table 1 (continued) Panel C: Distribution of the balance sheet disclosure ratio, computed by dividing each firm s frequency of balance sheet disclosure by the frequency of its quarterly reports over the sample period (n ¼ 2; 551) Mean Standard deviation Median Upper quartile Lower quartile 37% 43% 9% 88% 0% Panel D: Number of breaks in disclosure after disclosing balance sheet for at least one quarter Number of breaks Number of Firms , Proportion of firms (%) S. Chen et al. / Journal of Accounting and Economics 33 (2002)

11 S. Chen et al. / Journal of Accounting and Economics 33 (2002) analysis provides an indication of whether the decision to disclose is likely to be made on a quarterly basis, or whether firms continue to disclose once they begin. Panel D indicates that 65% of the disclosure firms do not have a break in disclosure. Thus, once firms begin disclosing, they tend to continue. This suggests that it may be useful to explore factors that are associated with managers decisions to initiate disclosure and we perform this investigation in Section Sample of firms with information to test hypotheses To test our hypotheses we exclude observations that are missing data required to compute our independent variables, observations in the top and bottom 1% of the Return Volatility and log(market Value), and observations in the top 1% and all negative values of the Market-to-Book Ratio. These restrictions result in a sample of 14,672 earnings announcements made by 2,170 firms, including 5,835 (40%) with balance sheet disclosures made by 1,093 (50%) firms. The percentage of observations and firms with balance sheet disclosures in our test sample are similar to that found in the larger sample examined in panel A of Table 1 (37% and 52%, respectively), suggesting that our hypotheses test sample is reasonably representative of that group. Table 2 reports descriptive statistics for the independent variables used in the logit analysis, along with other descriptive variables. The far right column reports the Table 2 Descriptive statistics of sample observations, by disclosure type. The disclosure sample=5,835 and the non-disclosure sample=8,837 Variables Sample Mean Median s t-stat a z-stat Variables of interest (1) High-Tech Dummy Disclosure Non-Disclosure (2) Loss Dummy Disclosure Non-Disclosure (3) Absolute Forecast Error Dummy Disclosure Non-Disclosure (4) Mergers & Acquisition Dummy Disclosure Non-Disclosure (5) Age Disclosure Non-Disclosure (6) Return Volatility Disclosure Non-Disclosure Control variables (1) Log (Market Value) Disclosure Non-Disclosure (2) Analyst Coverage Disclosure Non-Disclosure (3) Market-to-Book Disclosure Non-Disclosure

12 240 S. Chen et al. / Journal of Accounting and Economics 33 (2002) Table 2 (continued) Variables Sample Mean Median s t-stat a z-stat Other variables (1) Return On Assets Disclosure Non-Disclosure (2) Market Value ($ million) Disclosure Non-Disclosure (3) Stock Return Disclosure Non-Disclosure (4) Short Window Stock Return Disclosure Non-Disclosure (5) Forecast Error Disclosure Non-Disclosure (6) Absolute Forecast Error Disclosure Non-Disclosure a t-tests refer to differences in means and z-statistics refer to Wilcoxon two-sample tests. Variable definitions Variables of interest High-Tech Dummy=1 if the firm reports Compustat SIC codes (Drugs), (R&D services), (programming), (computers), (electronics), or (precise measurement instruments), and 0 otherwise. Loss Dummy=1 if quarterly net income before extraordinary items (from Compustat) is negative during the current quarter and 0 otherwise. Absolute Forecast Error Dummy=1 if the absolute value of the forecast error (defined as reported earnings minus the most recent consensus mean analysts forecast from the First Call database) is larger than one cent during the current quarter, and 0 otherwise. Mergers &Acquisitions Dummy=1 if the firm reports mergers and acquisitions activity during the current quarter (as reported in Compustat), and 0 otherwise. Age=the year of the current quarterly announcement minus the first year the firm is publicly traded (according to the CRSP database). Return Volatility=the standard deviation of stock returns over the prior 250 days, where at least 100 days of stock returns are required for inclusion in the sample. Control variables: log(market Value)=natural log of the firm s market value at the end of the current quarter. Analyst Coverage=number of analysts following (from the First Call database) at the end of the current quarter. Market-to-Book Ratio=ratio of market-to-book value at the end of the current quarter. Other variables: Return on assets=income before extraordinary items scaled by average total assets. Market value=firm s market value at the end of the current quarter. Stock return=raw stock return during the 60 days prior to the current quarter earnings announcement. Short window stock return=two-day raw return compounded over the day of the earnings announcement and the following day. Forecast error=reported earnings minus the most recent consensus mean analysts forecast from the First Call database. Absolute Forecast Error=absolute value of Forecast Error.

13 S. Chen et al. / Journal of Accounting and Economics 33 (2002) results of t-tests comparing the means across the disclosure and non-disclosure groups, and Wilcoxon 2-sample tests comparing the medians. The first six variables are the variables of interest in our logit analysis and hence represent univariate tests of our hypotheses. All six of these variables are significantly different across the two groups in the predicted direction with p-values o1%, providing support for all of our hypotheses at the univariate level. The next three variables in Table 2 are control variables used in our logit analysis, and indicate that the disclosure firms tend to have greater analyst coverage and larger market-to-book ratios. The next set of variables we analyze includes ROA and some market-related variables. This analysis finds that the disclosure firms have smaller mean ROAs and market values, but that the medians of these variables are insignificantly different across the two groups. The disclosure firms also tend to report larger stock returns during the 60 days prior to the earnings announcement, but their two-day returns around the earnings announcement are insignificantly different from the non-disclosure firms. In addition, while the disclosure firms mean forecast error is smaller, the median forecast error is larger; and the mean and median absolute values of the forecast error are larger. Larger absolute forecast errors are consistent with the univariate tests of the Absolute Forecast Error Dummy reported above. Table 3 reports the Pearson and Spearman correlation coefficients among the independent variables used in the logit analysis. Consistent with prior research, log(market Value) is highly positively correlated with Analyst Coverage, with a Pearson (Spearman) coefficient of 0.74 (0.72). The Return Volatility variable is also highly correlated with Age, log(market Value), and Analyst Coverage Hypotheses tests with all available observations Table 4 presents the logit tests of our hypotheses. Because many sample firms appear multiple times during the 12 quarters we analyze, we run separate regressions for each quarter and examine the means of the 12 estimated coefficients. Panel A presents the means of the coefficients along with tests of their significance. This analysis finds that each of our hypotheses is supported with p-values o1% (twotailed). Specifically, our sample firms are more likely to disclose balance sheet information when they are in high technology industries, when they report losses, when the absolute value of the forecast error is larger, subsequent to mergers or acquisitions, when they are relatively younger, and when their stock returns are more volatile. The coefficients on the control variables indicate that balance sheet disclosures are relatively more likely among large firms with higher Analyst Coverage and lower Market-to-Book ratios. Since the results on log(market Value) and Market-to-Book are inconsistent with our univariate findings in Table 2, and because Table 3 finds that the correlations associated with these variables are relatively high, we perform the procedure recommended in Allison (1999) to assess whether multi-collinearity impacts our logit model estimates. While not reported in a table, we find evidence of multi-collinearity with respect to log(market Value) and Market-to-Book. Therefore, we rerun the analysis in Panel A of Table 4 after dropping log(market Value),

14 242 Table 3 Correlation matrix of the independent variables used in the logit analysis; Pearson correlations in the upper diagonal, and Spearman correlations in the lower diagonal. Total observations=14,672 a High- Tech Dummy Loss Dummy Absolute Forecast Error Dummy Mergers & Acquisitions Dummy Age Return Volatility Log (Market Value) Analyst Coverage High-Tech Dummy Loss Dummy Absolute Forecast Error Dummy Mergers & Acquisitions Dummy Age Return Volatility Log (Market Value) Analyst Coverage Market-to-Book Ratio a See Table 2 for variable definitions. Correlations coefficients are significant at px0:0001 when jrjx0:03; and significant at px0:05 level with 0:03 > jrj > 0:02: Marketto-Book Ratio S. Chen et al. / Journal of Accounting and Economics 33 (2002)

15 S. Chen et al. / Journal of Accounting and Economics 33 (2002) Table 4 Mean coefficients of 12 logistic regressions by quarter; dependent variable=1 if firm discloses balance sheet, and 0 otherwise a Independent variables Predicted Sign Mean coefficients No. of positive coefficients t-statistics b Panel A: Total sample High-Tech Dummy *** Loss Dummy *** Absolute Forecast Error Dummy *** Merger & Acquisition Dummy *** Age *** Return Volatility *** Log (Market Value)? *** Analyst Coverage? ** Market-To-Book? *** Panel B: Initiation sample c High-Tech Dummy *** Loss Dummy ** Absolute Forecast Error Dummy Merger & Acquisition Dummy Age *** Return Volatility *** Log (Market Value)? *** Analyst Coverage? Market-To-Book? ** ***¼ po0:01 (two-tailed). **¼ po0:05 (two-tailed). *¼ po0:10 (two-tailed). See Table 2 for variable definitions. a The coefficients in the table are estimated from the following logit model: Balance Sheet Disclosure Indicator ¼ a 0 þ a 1 ðhigh-tech DummyÞþa 2 ðloss DummyÞ þ a 3 ðabsolute Forecast Error DummyÞ þ a 4 ðmergers & Acquisitions DummyÞ þ a 5 ðageþþa 6 ðreturn VolatilityÞ þ a 7 ðlogðmarket ValueÞÞþa 8 ðanalyst CoverageÞ þ a 9 ðmarket-to-book RatioÞþe: The total observations used to estimate the 12 annual regressions for the full sample equal 14,672, including 5835 with balance sheet disclosures. The total observations used to estimate the 12 annual regressions for the initiation sample equal 6667, including 485 with balance sheet disclosures. b t-statistics=the mean of the coefficients/the standard error of the coefficients (Fama and MacBeth, 1973; Bernard, 1987). c The initiation sample consists of all balance sheet disclosure observations immediately preceded by a non-disclosure quarter, and control firms in the same year and quarter that never disclose balance sheet information.

16 244 S. Chen et al. / Journal of Accounting and Economics 33 (2002) then again after dropping Market-to-Book, and again after dropping both log(market Value) and Market-to-Book. This analysis finds that all six hypotheses continue to be supported at the 1% level (two-tailed) in all three additional regressions, suggesting that multi-collinearity does not impact our primary findings Hypotheses tests on initial balance sheet disclosures Panel D of Table 1 reports that once firms begin making balance sheet disclosures, 65% tend to continue. This suggests that there may be factors in the initial disclosure quarter that motivate managers to begin disclosing balance sheet information. We investigate this issue by examining the sub-sample of our disclosure observations that initiate disclosure during the period we analyze. We construct an initiation sample by defining the initiation quarter as a quarter in which a firm discloses balance sheet information that is immediately preceded by a quarter in which the firm does not disclose balance sheet information. Because we do not have information on the reporting behavior of firms that appear in the first quarter of our analysis, this analysis is limited to 11 quarters. Our control group consists of all of the other firms in the sample, in the same quarter and year, that do not disclose balance sheet information during the entire period we analyze. 7 Table 4, Panel B presents the means of the coefficients from the 11 regressions estimated for the initiation sample. This test finds support for four of our six hypotheses, with p-values o5% (two-tailed). Specifically, our sample firms are more likely to initiate balance sheet disclosures when they are in high technology industries, when they report losses, when they are relatively younger, and when they have larger stock return volatility. We do not find that initiation of balance sheet disclosure is associated with the absolute value of the forecast error or with merger and acquisition activity. However, we note that there are some limitations to the initiation analysis. One is that we are unable to identify the true initiation quarter because we only have data for 12 quarters. That is, firms may actually initiate disclosure for the first time during a quarter prior to the period covered by our analysis. Another limitation is that limiting the analysis to the initiation sample greatly reduces our sample size. The average sample size used to estimate the logit coefficients is 606 per quarter for the initiation sample versus 1,222 for the full sample. Therefore, given these limitations, the results of the initiation sample should be interpreted with caution. 5. Market analysis 5.1. Analysis of price earnings associations The results of our hypotheses tests are consistent with managers disclosing balance sheet information when current earnings are relatively less informative, or when 7 We also define the control firms to be the non-disclosure firms matched on year, quarter and size. This reduces the sample size and produces weaker associations in the logit analysis.

17 Table 5 Market-based tests Panel A: Mean coefficients and adjusted R 2 of 12 price earnings regressions by-quarter with t-statistics in parentheses a Price=b 0 +b 1 (Earnings Per Share)+e Price=b 0 +b 1 (Earnings Per Share)+b 2 (Book Value Per Share)+e All observations After dropping loss observations Intercept Earnings Book Value Adj. R 2 (%) Intercept Earnings Book Value Adj. R 2 (%) B/S disclosers (39.35)*** (6.55)*** (21.47)*** (12.98)*** Non-disclosers (26.86)*** (8.94)*** (8.50)*** (9.47)*** t-statistic for test of difference between the two adjusted R 2 : 2.26** t-statistic for test of difference between the two adjusted R 2 : 2.73** B/S disclosers (30.01)*** (5.52)*** (27.76)*** (18.77)*** (9.73)*** (15.12)*** Non-disclosers (12.53)*** (7.18)*** (17.51)*** (7.23)*** (10.54)*** (19.02)*** t-statistic for test of difference between the two adjusted R 2 : 5.93*** t-statistic for test of difference between the two adjusted R 2 : 5.14*** Panel B: Mean Coefficients and Adjusted R 2 of 12 returns return earnings regression, by-quarter with t-statistics in parentheses a Returns[ 60,+1]=b 0 +b 1 (Earnings)+e All observations After dropping loss observations Intercept Earnings Adj. R 2 Intercept Earnings Adj. R 2 B/S disclosers (4.79)*** (7.72)*** 2.55 (0.14) (6.39)*** 4.20 Non-disclosers (3.34)*** (7.15)*** 2.53 ( 1.15) (7.69)*** 4.55 t-statistic for test of difference between the two adjusted R 2 : 0.02 t-statistic for test of difference between the two adjusted R 2 : 0.24 ***=po0:01 (two-tailed). **=po0:05 (two-tailed). *=po0:10 (two-tailed). a t-statistics=the mean of the coefficients/the standard error of the coefficients. S. Chen et al. / Journal of Accounting and Economics 33 (2002)

18 246 S. Chen et al. / Journal of Accounting and Economics 33 (2002) future earnings are relatively more uncertain. We further investigate this finding by examining the value relevance of the earnings of the disclosure firms versus the nondisclosure firms using the price earnings regression in Ohlson (1995), which is assumed to measure the value relevance of earnings. If the balance sheet disclosure firms earnings are relatively less value-relevant, we expect the relation between earnings and price to be relatively weaker for these firms, providing them with an incentive to supplement their earnings announcements with balance sheet disclosures. We test this prediction by estimating the coefficients in the following two regressions for the disclosure and the non-disclosure samples using quarterly data Price ¼ b 0 þ b 1 ðearnings Per ShareÞþe; Price ¼ b 0 þ b 1 ðearnings Per ShareÞþb 2 ðbook Value Per ShareÞþe: ð2þ ð3þ If our predictions are correct, we expect the balance sheet disclosure sample s R 2 to be lower. Table 5 Panel A compares the mean coefficients from 12 quarterly regressions, for each regression model, across the disclosure and non-disclosure samples. We run two analyses for each regression, one with all available observations and one after dropping firms that report losses. We drop loss firms because Collins et al. (1999) find that loss firms have a smaller mean coefficient on earnings and a larger mean coefficient on book value when compared to firms reporting positive earnings. We compare R 2 across the two groups using a t-statistic equal to the difference in the means of the R 2 divided by the pooled standard error of the R 2,asin Ball et al. (2000). Consistent with our prediction, the R 2 are significantly smaller for the balance sheet disclosure sample at the 5% level (two-tailed). Thus, the analysis in Panel A is consistent with managers making voluntary balance sheet disclosures when earnings are relatively less value-relevant Analysis of returns earnings associations We further investigate differences in the relation between earnings and share value by comparing the following earnings return regression across the disclosure and non-disclosure samples using quarterly data Returns½ 60; þ1š ¼b 0 þ b 1 ðearningsþþe: ð4þ Raw returns are accumulated from 60 days prior to the earnings announcement to the day prior to the earnings announcement, and earnings are scaled by price at the beginning of the return window, with both variables truncated at the top and bottom 1%. Panel B of Table 5 compares the mean coefficients from 12 quarterly regressions across the disclosure and non-disclosure samples before and after dropping loss firms. Surprisingly, the R 2 is slightly larger for the disclosure sample when all observations are used, but the result reverses when loss observations are dropped. However, t-tests indicate that these differences are not significant at conventional levels. Thus, in contrast to our price earnings analysis reported in Panel A, our

19 S. Chen et al. / Journal of Accounting and Economics 33 (2002) returns earnings analysis does not find evidence that the balance sheet disclosure firms have less informative earnings. 6. Robustness checks Except where noted, we find that our hypotheses continue to be supported after the following robustness checks: Alternative classification scheme for high technology industries: We redefine the High-Tech Dummy variable using the classification scheme used in Francis and Schipper (1999). Alternative specification of Loss Dummy: We redefine the Loss Dummy as net loss (consistent with Burgstahler and Dichev, 1997). Alternative specification of Absolute Forecast Error Dummy: We redefine the Absolute Forecast Error Dummy as a continuous measure. Sensitivity to dropping observations without analysts forecasts: Our sample firms are restricted to the First Call database in order to compute the Absolute Forecast Error Dummy and Analyst Coverage variables. To test whether our results are sensitive to this constraint, we drop this restriction and rerun our hypotheses tests without the Absolute Forecast Error Dummy and Analyst Coverage variables. Alternative specification of Return Volatility: We replace Returns Volatility with Earnings Volatility, measured as the standard deviation of the return on stockholders equity (ROE) over the 5 years prior to the earnings announcement. The data requirement to estimate the standard deviation of past earnings reduces our sample to 9,793 observations. The results find that the mean coefficient on the Mergers and Acquisitions Dummy is not significant at conventional levels. Including a dummy variable for special items: Because special items, such as asset write-downs, often impact the balance sheet we include a dummy variable indicating whether a special item is reported during the quarter. This analysis finds that the mean coefficient on the special item dummy is not significant at conventional levels. Including a dummy variable for subsequent debt or equity issuance: A potential motivation for balance sheet disclosure is to reduce information asymmetry between management and market participants prior to public offerings. Thus, we add a dummy variable indicating whether the firm issues public debt or equity during the four quarters subsequent to the earnings announcement. This analysis finds that the mean coefficient on the security issuance variable is not significant at conventional levels. Including a variable for working capital: Because market participants may demand balance sheet information in order to help understand working capital, we include working capital (scaled by sales) as an independent variable. This analysis finds that the mean coefficient on the working capital variable is not significant at conventional levels.

20 248 S. Chen et al. / Journal of Accounting and Economics 33 (2002) Sensitivity to excluding firms in high technology industries: Because high technology firms make a disproportionately large number of the balance sheet disclosures, we investigate whether these firms drive our results by dropping high technology firms from our sample. This analysis finds that the significance of the mean coefficients on the Loss Dummy and Age drops to the 10% level (two-tailed), while the mean coefficients on the Absolute Forecast Error Dummy and the Merger and Acquisitions Dummy are not significant at conventional levels. 8 Therefore, the results on the Absolute Forecast Error Dummy and Mergers and Acquisitions variables appear concentrated among firms in the high technology industries. 7. Summary We describe the recent trend in voluntary balance sheet disclosures and test hypotheses that attempt to explain why managers voluntarily include balance sheet information in quarterly earnings announcements. We perform our analyses using all earnings announcements in the Wall Street Journal Proquest database from the fourth quarter 1992 through the third quarter 1995, and find that 52% of the firms include balance sheets in their earnings announcements, and that the relative frequency of this disclosure grows from 31% to 46% over the period we analyze. Our overall prediction is that managers are more likely to voluntarily disclose balance sheet information when current earnings are relatively less informative, or when future earnings are relatively more uncertain. Consistent with our expectations, we find that balance sheet disclosures are more likely among firms: (1) in high technology industries; (2) reporting losses; (3) with larger forecast errors; (4) engaging in mergers or acquisitions; (5) that are younger; and (6) with more volatile stock returns. Corroborating the results in our hypotheses tests, we also find that quarterly earnings explain a significantly lower proportion of share value among firms that make balance sheet disclosures. However, we do not find that quarterly earnings explain a significantly different proportion of stock returns among firms with balance sheet disclosures. Overall, our findings are consistent with voluntary balance sheet disclosures being motivated by investor demand for additional value relevant information to supplement reported earnings. Our study contributes to the literature that examines the usefulness of balance sheet information in valuing securities by identifying the circumstances under which market participants are likely to demand, and managers are likely to provide, voluntary balance sheet disclosures. 8 Further investigation finds that the relative frequency of merger and acquisitions activity among the 4,342 high technology firm-quarters is 7%, compared with 2.7% among the 10,330 non-high technology firm-quarters.