Earnings Management and the Long-Run Underperformance of Firms Following Convertible Bond Offers

Journal of Business Finance & Accounting, 36(1) & (2), 73 98, January/March 2009, 0306-686X doi: 10.1111/j.1468-5957.2008.02120.x Earnings Management and the Long-Run Underperformance of Firms Following Convertible Bond Offers De-Wai Chou, C. Edward Wang, Sheng-Syan Chen and Sandra Tsai Abstract: This paper examines whether the long-run underperformance of convertible bond issuers can be explained by earnings management, as reflected in discretionary current accruals around the time of the offer. Consistent with the earnings management hypothesis, we find that convertible issuers who adjust their discretionary current accruals to report higher net income in the issue year will generally experience inferior operating and stock return performance over the five-year post-issue period. Our findings indicate that there is some temporary overvaluation of convertible issuers by the stock market, but that the resultant disappointed investors will subsequently correct their valuation errors. The similarity of our results to those reported within the prior literature on initial public offers (IPOs) and seasoned equity offers (SEOs) suggests that the earnings management hypothesis is not unique to stock offers, but that it actually extends to convertible bond offers. Keywords: convertibles bond, earnings management, accounting accruals, long-run performance 1. INTRODUCTION It is noted in several of the prior studies that firms experience poor long-run operating and stock price performance following their convertible bond offers, with the average annual stock return of convertible bond issuing firms being about 4 to 8% less than that of non-issuing matched firms over the five-year post-offer period. 1 Many of these prior studies also show that, as compared to matched firms, convertible bond issuers have much greater declines in cash flow and investment-related operating performance over The authors are respectively from the Department of Finance, Yuan Ze University, Taiwan; the Department of Business Administration, National Taipei University, Taiwan; the Department of Finance, National Taiwan University, Taiwan; and the Department of Finance, National Chung Cheng University, Taiwan. They thank Peter F. Pope (editor) and two anonymous referees for helpful comments on earlier drafts. This paper has also benefited from comments by seminar participants at the National Chengchi University and National Central University in Taiwan. (Paper received April 2006, revised version accepted September 2008) Address for correspondence: De-Wai Chou, Department of Finance, Yuan Ze University, Taiwan. e-mail: dwchou@saturn.yzu.edu.tw 1 See for example, Lee and Loughran (1998), Spiess and Affleck-Graves (1999) and Lewis et al. (2001). Journal compilation C 2009 Blackwell Publishing Ltd, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA. 73

74 CHOU, WANG, CHEN AND TSAI the four-year post-offer period (McLaughlin et al., 1998; Lee and Loughran, 1998; and Lewis et al., 2001). In this study, we set out to examine whether aggressive earnings management by income-increasing accounting adjustments around the time of convertible bond offers can explain the subsequent inferior operating and stock price performance of the issuing firms. Legoria et al. (1999) find that firms use discretionary accruals to create a pattern of improved financial performance leading up to the year of the debt issue. Similarly, firm managers also have incentives to manage reported earnings prior to the issuing of convertible bonds in order to minimize the company s risk assessment by creditors. In addition, theory predicts that for some firms, convertible bonds can simultaneously mitigate both debt- and equity-related financing problems. However, Lewis et al. (2001) find no support for any prediction of such investment behavior by firms; indeed, they conclude that firms rationed out of the equity market will respond either by immediately issuing convertible bonds, or by subsequently issuing equity. Munro (1996) provides some preliminary evidence concerning the nature of convertible bonds issuers in the UK, but his study still would not be able to rationalize the use of such issues. From a standpoint of the conversion mechanism embedded in convertible bonds, it is implicit that some element of the nature of issuing convertible bonds is similar to that of issuing common equity; hence, Lewis et al. (1999) provide a methodology for separating issuers into groups, with the securities design suggesting that issuers are debt-like, equity-like or hedgers; that is, convertible bonds can be regarded as straight debt with an embedded call option. It is therefore clear that the economic substance of convertible bonds is an issue of significant importance to investors (Casson, 1998; and Mehran and Homaifar, 1993). If the stock market were to interpret the offer of convertible bonds as being more akin to equity than debt, then the increased likelihood of dilution of existing shareholder wealth would send out a strong negative signal to the market (Purdy, 1977). Furthermore, investors might expect to observe unusual phenomenon during season equity offers; managers will, for example, often engage in upward earnings management in the pre-issuance period to fool the market into setting a better offer price (Marquardt and Wiedman, 2004). We therefore expect to find a negative relationship between conversion probability and long-term stock price performance. Very little empirical evidence has thus far been presented to indicate whether the poor long-term operating and stock price performance of issuers following convertible and non-convertible security offers may simply be attributable to common factors. Thus, similarities in operating performance following the issue of both convertible bonds and common equity motivate us to examine whether the earnings management associated with equity offers may also occur in the case of the issuance of convertible bonds. 2 Rangan (1998) and Teoh et al. (1998a and 1998b) note that stock issuers are found to have both unusually high income-increasing accounting adjustments around the time of the offer, and unusually poor operating and stock return performance in the subsequent post-offer period; similarly, stock issuers are also found to exhibit inferior performance when they make unusually large income-increasing accounting adjustments around the time of the offer. 2 See for example, Loughran and Ritter (1997) and Lewis et al. (2001).

EARNINGS MANAGEMENT AND LONG-RUN UNDERPERFORMANCE 75 This suggests that stock issuers will generally engage in the upward management of earnings in order to increase the offer proceeds, and that investors can often misinterpret the transitory increases in earnings reported at the time of stock offers, resulting in their substantial overvaluation of such issues. In the post-offer period, when high earnings cannot be sustained, the market is invariably disappointed and subsequently corrects the balance by lowering its valuation. However, this earnings management hypothesis may not be unique to stock offers, since it may be that it also extends to convertible bond offers. Firms which are overvalued are likely to issue securities of any type; thus, it is likely that convertible offers, like stock offers, provide a signal of the overvaluation of the firm. The inability of investors to unravel earnings management surrounding convertible offers is therefore a potential source of the issuers subsequent poor performance. In this study, we use discretionary current accruals to measure the manipulation of earnings determined under management discretion, adopting the methods developed in Teoh et al. (1998a and 1998b). We report evidence consistent with the earnings management hypothesis on a sample of convertible bond offers which took place between 1981 and 1998, and find a discernible increase in discretionary current accruals prior to the issue date, a peak during the issue year, and a subsequent decline thereafter. Our results indicate that in the five-year period following convertible bond offers, issuers experienced declines in asset-scaled operating income before depreciation plus interest income (OIBD/assets) and return on assets (ROA). This post-issue decline in OIBD/assets and ROA is particularly pronounced for those issuers who adopted aggressive earnings management in the issue year. Our evidence suggests that managers advance their accruals to increase their reported net income during the convertible bond issue period. We also find that discretionary current accruals around the offer can predict the underperformance of post-issue stock returns. We find that during the five-year post-issue period, the stock performance of those convertible issuers who adopted aggressive earnings management in the issue year was inferior to that of issuers who adopted conservative earnings management; these results hold for various measures of abnormal returns, benchmarks and accumulation periods. We further show that postissue abnormal stock returns have a significantly negative correlation with discretionary current accruals in the issue year. Although our evidence suggests that discretionary current accruals have significant influences on subsequent stock returns for convertible bond issuers, we are unable to determine any robust relationship between conversion probability and post-issue stock performance, indicating that the issues may not be classified by the market on the basis of their economic substance. The similarity of our results to those on seasoned equity offers reported in Rangan (1998) and Teoh et al. (1998b) provide support for the earnings management hypothesis. If equity financing is substituted by convertible financing, the managers of the issuing firms still have incentives to manipulate their reported earnings to increase the offer proceeds, and investors have a continuing tendency to overvalue such new issues. Therefore, income-increasing accounting adjustments around the time of convertible offers can explain the subsequent inferior performance of the issuers. The remainder of this paper is organized as follows. Section 2 describes our sample selection procedure and methodology, followed in Section 3 by our examination

76 CHOU, WANG, CHEN AND TSAI of whether discretionary current accruals surrounding convertible bond offers can predict operating underperformance in the post-offer period. In Section 4, we investigate the predictability of post-offer stock return underperformance with discretionary current accruals around the time of the offer. The final section presents the conclusions drawn from this study. 2. SAMPLE SELECTION AND METHODOLOGY (i) Sample Design and Characteristics The sample adopted for this study is based upon US convertible bond offers which took place between 1981 and 1998, as recorded in the Securities Data Company s New Issues Database. In order to qualify for the final sample, the issues had to meet the following criteria: (a) The common stock of the issuing firm must have been listed on the New York Stock Exchange (NYSE), the American Stock Exchange (AMEX) or the Nasdaq, and must have securities returns available from the Center for Research in the Securities Prices (CRSP) tapes; (b) The issuing firm must not be a regulated utilities or financial institution; (c) The issuing firm must have financial information available from Compustat; and (d) The issuing firm must not have engaged in any other convertible bond offers in the five-year period prior to the current issue date. 3 Our final sample comprised of 312 convertible bond offers. Sample distribution details are provided in Table 1, by calendar year and by industry. As Panel A shows, the largest number of convertible bond offers in one year was 1986, when there were 57 offers (18.3%), followed by 1987 with 34 offers (10.9%) and 1985 with 25 offers (8%). The high level of issue activity in the years 1985 to 1987 reflects the convertible offers reported in Lee and Loughran (1998), McLaughlin et al. (1998) and Lewis et al. (2001). Panel B of Table 1 shows that the issuers represent a broad cross-section of industries, thereby indicating that convertible bond financing is not specific to a small group of industries. The top three industries, in terms of the number of convertible offers, were the computer equipment and services industries (20.5%), wholesale and retail industries (13.1%) and electric and electronic equipment industries (9%). (ii) Measuring Earnings Management If earnings management is used to increase earnings, the increase can be accomplished through the acceleration of the recognition of revenue, or a delay in the recognition 3 The inclusion of this final criterion is essentially because, as in Spiess and Affleck-Graves (1999) and Lewis et al. (2001), the focus in this study is on the examination of stock price and operating performance over several years.

EARNINGS MANAGEMENT AND LONG-RUN UNDERPERFORMANCE 77 Table 1 Sample Distributions of Convertible Bond Offers, by Year and Industry Panel A: Sample Distribution by Year Year Number Percent of Sample 1981 18 5.8 1982 15 4.8 1983 17 5.4 1984 13 4.2 1985 25 8.0 1986 57 18.3 1987 34 10.9 1988 8 2.6 1989 15 4.8 1990 10 3.2 1991 12 3.8 1992 20 6.4 1993 22 7.1 1994 6 1.9 1995 10 3.2 1996 13 4.2 1997 12 3.8 1998 5 1.6 Total 312 100.0 Panel B: Sample Distribution by Industry Industry Group Two-digit SIC Code Number Percent of Sample Computer equipment and services 35,73 64 20.5 Wholesale and retail 50,51,52,53,54,56,57,59 41 13.1 Electric and electronic equipment 36 28 9.0 Manufacturing 30,32,33,34 24 7.7 Transportation 37,39,40,41,42,45 24 7.7 Chemicals and allied products 28 21 6.7 Eating and drinking establishments 58 14 4.5 Scientific instruments 38 13 4.2 Health services 80 12 3.8 Oil and gas extraction 13 12 3.8 Paper and paper products 24,25,26,27 11 3.5 Other 48 15.4 Notes: This table summarizes the sample distributions of 312 convertible bond offers recorded in Securities Data Company s New Issues Database during the period from 1981 to 1998. The sample selection criteria used are: (i) the common stock of the issuing firm is listed on the New York Stock Exchange (NYSE), the American Stock Exchange (AMEX), or the Nasdaq, and has security returns available from the Center for Research in Securities Prices (CRSP) tape; (ii) the issuing firm is not a regulated utility or a financial institution; (iii) the issuing firm must have financial information available from Compustat; and (iv) the issuing firm must have had no other convertible bond offers in the five years prior to the issue date. The industries (defined by Compustat two-digit SIC codes) listed in Panel B each have 10 or more convertible bond offers. of expenses, relative to cash flows. 4 The differences between recognized revenue and cash received, or between recognized expenses and cash expenditure, result in accruals or deferrals. Since the basis of earnings management lies in the difference between cash flows and earnings, the analysis of accruals, which is the difference 4 Earnings management can also be accomplished through changes in accounting methods and changes in capital structure, such as debt defeasance and debt-equity swaps.

78 CHOU, WANG, CHEN AND TSAI between cash flows and earnings, provides insights into earnings management practices. It should be noted, however, that accrual items are not subject to equal manipulation or management. Long-term accrual and deferral items relate to accounting adjustments to long-term assets or liabilities (such as depreciation), and these are quite difficult to manage or adjust, essentially because the accounting choices relating to long-term assets remain constant over several years. Conversely, short-term accruals, which are accounting adjustments to short term assets (such as changes in accounts receivable), are much easier to manage, simply because accounting choices are made over a much shorter time horizon. Since short-term accruals are more readily subject to management, the focus of this study, as in many of the recent studies, is on short-term accruals. 5 The computation of accruals in this study is based upon the definition of accruals provided by Perry and Williams (1994), in which total accruals are computed as the change in non-cash working capital (excluding current maturities of long-term debt, less total depreciation expenses for the current period). 6 Their definition is similar to that of Jones (1991), differing only by the exclusion of the adjustment made for income tax. Perry and Williams (1994) elected to include income tax in their model essentially because income tax accruals could well represent an important component of an earnings management strategy. Earnings management is revealed by an abnormal level of accruals relative to a firm s business activities. Since a regression model can be used to estimate expected accruals, any deviations from the expected accruals would suggest that they were subjected to management discretion, which might well be attributable to earnings management. We follow the methodology of Teoh et al. (1998a and 1998b) to estimate the expected current accruals from a modification of the Jones (1991) model. Expected accruals are estimated from a cross-sectional regression of current accruals on the change in sales in a given year using an estimation sample which includes all firms with the same two-digit SIC code as the convertible issuer, whilst excluding the actual issuer and other convertible issuers. At least 30 firms are required with exactly the same two-digit SIC code in order to ensure the preciseness of the estimated coefficients obtained from the regression. In addition to the adoption of the appropriate SIC code filter, all non-ordinary common stocks, such as ADRs, closed-end funds and REITs are also excluded from the estimation sample. A cross-sectional regression is then performed for each fiscal year, with all of the variables being scaled by lagged total assets in order to reduce the potential heteroskedasticity within the data. The fitted current accruals of the issuer are then calculated for a specific year using the estimated coefficients from the corresponding regression and the change in sales, net of the issuer s change in trade receivables, for that year. 7 The fitted current accruals, which are then referred to as non-discretionary current accruals (NDCA), are the asset-scaled proxies for non-manipulated accruals dictated by 5 See for example, Garcia et al. (2005) and Peasnell et al. (2005). 6 These are also adopted by Teoh et al. (1998a and 1998b) who provide a detailed description of the definition and construction of accrual measures, which includes the specific Compustat items used to construct the accrual measures (Teoh et al., 1998b). We follow their description and definition in the construction of our accrual estimates in this study. 7 The change in trade receivables is subtracted from the change in sales to allow for the possibility of sales manipulation.

EARNINGS MANAGEMENT AND LONG-RUN UNDERPERFORMANCE 79 business conditions. It is presumed that the remaining portion of the current accruals, referred to as discretionary current accruals (DCA), are not dictated by either firm or industry conditions; thus, they provide the proxies for the manipulation of earnings determined at the discretion of managers. Using the estimation sample, we run the following cross-sectional regression: TA j,t /AT j,t 1 = α 0 (1/AT j,t 1 ) + α 1 ( Sales j,t /AT j,t 1 ) + ε j,t, (1) where TA is total current accruals; AT is total assets; Sales is the change in sales; j indicates the firms in the estimation sample; and t indicates the year. Non-discretionary (or expected) current accruals for firm i are estimated by: NDCA i,t = ˆα 0 (1/AT i,t 1 ) + ˆα 1 [( Sales i,t TR i,t )/AT i,t 1 ], (2) where ˆα 0 is the estimated intercept; ˆα 1 is the slope coefficient for convertible bond issuing firm i; and TR i,t is the change in trade receivables for year t for convertible bond issuing firm i. The increase in trade receivables is then subtracted from the change in sales to allow for the possibility of credit sales manipulation by the issuer. Discretionary current accruals, DCA i,t, for convertible bond issuing firm i for year t, are then estimated as: DCA i,t = TA i,t /AT i,t 1 NDCA i,t. (3) These discretionary current accruals are used as the measure of abnormal accruals, which in this study, is also the proxy for earnings management. (iii) Measuring Conversion Probability We use standard Black-Scholes assumptions and follow the model of Lewis et al. (1999) to estimate conversion probability; in their model, the decision on the choice of security issues is treated as a financing problem as opposed to a dichotomous debt-only or equityonly financing choice. Managers will usually endogenously determine the amount of debt and equity to be included in their incremental capital sourcing decisions. We assume the underlying common stock follows a diffusion process described by geometric Brownian motion. This probability is then estimated as N (d 2 ) where N ( ) is the cumulative probability under a standard normal distribution function. The probability is denoted as CP. The model is as follows: CP = ln(s/x) + (r div σ 2 /2)T σ, (4) T where S is the current price of the underlying common stock; X is the conversion price; r is the continuously compounded yield estimated from a 10-year US Treasury bond on the issue date; div is the issuing firm s continuously compounded dividend yield for the fiscal year-end immediately preceding the offer date; σ is the standard deviation of the continuously compounded common equity return estimated over the 240 to 40 trading-day period prior to the issue date; and T is the number of years until the maturity of the bond. The left-hand-side variable CP can take a continuous value

80 CHOU, WANG, CHEN AND TSAI between 0 and 1; therefore, as CP approaches 1, the more likely the issuer is to issue equity-like securities. (iv) Measuring Operating Performance We measure the operating performance of convertible bond issuers over the period from year 5 to year +5 using the two accounting ratios adopted in Lewis et al. (2001), including operating income before depreciation plus interest income scaled by total assets (OIBD/assets), and return on assets (ROA). We measure the abnormal operating performance of the issuing firms as compared with the performance of our control samples, following the procedure suggested by Barber and Lyon (1996) to construct the performance-matched samples to control for industry and economic conditions, as well as the mean-reversion tendency following abnormal pre-event performance. (v) Measuring Long-run Stock Price Performance The debate goes on within the asset pricing literature with regard to the appropriate method of measurement for long-run stock returns, with contrasting viewpoints having been proposed on the topic in various studies. 8 We use four methods in this study, buyand-hold returns (BHAR), cumulative abnormal returns (CAR), the three-factor model of Fama and French (1993) and a four-factor model which includes the three factors of Fama-French in conjunction with a one-year momentum factor, as proposed in Carhart (1997). Fama (1998) suggests that the adoption of the buy-and-hold abnormal returns methodology can be problematic because the distribution of long-term buy-and-hold returns is skewed. Although it is subsequently suggested by Lyon et al. (1999) that cumulative abnormal returns are less skewed than buy-and-hold abnormal returns, they nevertheless note that the skewness problem still remains. In order to address this skewness problem, in our test of the statistical significance of the cumulative abnormal returns, in addition to the conventional cross-sectional t-statistic, we also use a skewness-adjusted t-statistic, as derived by Hall (1992) and similar to the procedure described in Lyon et al. (1999). We calculate the cumulative abnormal returns relative to three benchmarks, the CRSP equally-weighted market index, the CRSP value-weighted market index, and a size and book-to-market matched control sample; the procedure for measuring the last of these follows that of Lyon et al. (1999). We first identify all of the firms within the CRSP database with an equity market value between 70% and 130% of the equity market value of a sample firm, and then select from this set of firms the firm with the book-to-market ratio which is closest to that of the sample firm. Firm size is the firm s total common equity market value measured on the first day of the issue month. Book-to-market ratio is the firm s equity book value divided by its equity market value, measured at the fiscal year end prior to the issue. Fama (1998) and Lyon et al. (1999) note that the stock returns of firms announcing a specific corporate event are usually correlated, whilst Lyon et al. (1999) also show that cumulative abnormal returns are affected by a problem of cross- sectional dependence. In order to solve this problem, we also use the monthly calendar-time portfolio 8 Examples include, amongst others, Barber and Lyon (1997), Fama (1998), Mitchell and Stafford (2000) and Loughran and Ritter (2000).

EARNINGS MANAGEMENT AND LONG-RUN UNDERPERFORMANCE 81 approach, as suggested by Fama (1998) and Mitchell and Stafford (2000), to estimate the three- and four-factor models. 3. ACCRUALS AND OFFER/POST-OFFER PERFORMANCE (i) Total and Discretionary Current Accruals Around Convertible Bond Offers Panel A of Table 2 divides our sample of convertible issuers into four quartiles, by their DCA in the issue year, with each quartile containing 78 firms. Following Teoh et al. (1998a and 1998b), we label the quartile of issuers with the highest DCA as aggressive issuers and the quartile of issuers with the lowest DCA as conservative issuers. The aggressive quartile has a DCA of at least 8.8%, the second quartile has a DCA of 1.4% to 8.8%, the third quartile has a DCA of 2.7% to 1.4%, and the conservative quartile has a DCA below 2.7%. The mean and median values of the market value of common equity (MVE), bookto-market ratio (BM), net income (NI), and total assets (AT) in the issue year within each quartile are reported in Panel B of Table 2. The two quartiles with the highest DCA comprise of smaller firms than the two more conservative quartiles, although this relationship is not monotonic across the four quartiles. In contrast, there is no systematic pattern between DCA and the book-to-market ratio. The mean value of DCA is 0.057 (5.7% of the total assets), which is statistically significant at the 0.01 level (t = 4.31). The mean value can be interpreted as discretionary current accruals comprising of 5.7% of the total assets for the year of the convertible bond offer. This value, as a percentage of total assets, is similar to the results in Teoh et al. (1998a) on a sample of seasoned equity offering (SEO) firms, in which they report that discretionary current accruals comprise of an average of 5.59% of the total assets. The mean DCA (0.057) comprises of a larger percentage of total assets than net income, indicating that without the inclusion of discretionary accruals, the average net income could well be negative. The median DCA (0.0138, which is 1.38% of total assets) is also significant at the 0.01 level. With the median assets for the sample being $241.18 million, the median accruals amount to $3.33 million, which is about one-third of the median net income (Table 2). Using the aggressive quartile to present another example, with the median assets being $129.9 million and DCA being 0.1893, the median accruals amount to $24.42 million, which is three times as much as the median net income. Not only is the median DCA statistically significant, but its effect as a proportion of the total median net income is economically significant. Following Teoh et al. (1998a), we take this evidence as an indication that managers are engaging in the upward management of earnings prior to the convertible bond offer. Table 3 presents the mean total current accruals (TA) and DCA over the 11-year period surrounding the convertible bond offers (years 5 to +5) for the whole sample and for the aggressive and conservative quartiles. Both the mean TA and DCA values, measured as the percentage of total assets at the end of year 1, tend to increase prior to the issuing year, and then decrease following the issuing year, with the highest mean for TA being 9.75% and the highest for DCA being 5.7%. Both the mean TA and DCA values for years 1 to+1 are significantly positive at the 5% level or better. After adjusting for control firms, the mean values of TA and DCA still exhibit similar

82 CHOU, WANG, CHEN AND TSAI Table 2 Summary Statistics of Convertible Bond Issuers Panel A: Discretionary Current Accruals Whole Aggressive Q1 Q2 Q3 Conservative Q4 Sample (DCA 8.8%) (8.8% > DCA 1.4%) (1.4% > DCA 2.7%) (DCA < 2.7%) DCA (%) Mean 5.70 30.37 4.49 0.64 11.41 (<0.01) (<0.01) (<0.01) (<0.01) (<0.01) Median 1.38 18.93 4.07 0.61 6.79 (<0.01) (<0.01) (<0.01) (<0.01) (<0.01) Panel B: Firm Characteristics Whole Sample Aggressive Q1 Q2 Q3 Conservative Q4 MVE ($ million) Mean 4,228.61 2,046.27 4,362.01 4,222.14 4,252.83 Median 219.34 123.34 193.63 296.31 251.40 BM (%) Mean 49.60 37.68 49.39 62.64 49.24 Median 41.50 32.37 43.99 56.41 38.88 NI ($ million) Mean 42.80 15.58 43.06 88.97 20.60 Median 9.65 7.35 8.56 13.47 12.11 AT ($ million) Mean 1,435.61 627.02 895.26 2,570.79 1,649.36 Median 241.18 129.90 208.33 398.35 289.52 Notes: This table presents summary statistics for discretionary current accruals (DCA), market value of common equity (MVE), book-to-market ratio (BM), net income (NI), and total assets (AT), classified by DCA at the year of convertible bond offers. We follow the methodology of Teoh et al. (1998a and 1998b) to estimate DCA. The aggressive and conservative quartiles contain the firms with the highest and smallest discretionary current accruals (DCA) at year 0, where year 0 is the fiscal year of the convertible offering. P-values are reported in parentheses.

EARNINGS MANAGEMENT AND LONG-RUN UNDERPERFORMANCE 83 Table 3 Mean Total Current Accruals and Discretionary Current Accruals Around Convertible Bond Offers Year Relative to the Issue Year 5 4 3 2 1 0 1 2 3 4 5 Panel A: Full Sample TA/assets 1.44 3.18 3.50 4.77 8.52 9.75 4.92 2.45 1.95 1.60 0.33 (0.89) (3.20) a (3.14) a (4.32) a (4.85) a (7.39) a (6.28) a (4.04) a (2.02) b (3.56) a (0.53) Control-firm- adjusted TA 2.92 1.43 3.76 1.51 6.68 6.08 3.92 1.29 0.42 3.33 4.76 ( 0.99) (0.75) ( 1.05) (1.00) (2.98) a (3.58) a (3.79) a (1.29) (0.29) (1.14) ( 1.69) c DCA/assets 0.94 0.17 0.39 1.09 3.70 5.70 2.79 0.32 0.14 0.66 0.97 (0.52) (0.17) ( 0.31) (1.10) (2.18) b (4.47) a (3.62) a (0.59) ( 0.15) (0.94) ( 1.68) c Control-firm- adjusted DCA 0.73 0.27 4.85 1.50 5.15 4.77 2.66 0.19 0.39 3.07 6.33 ( 0.21) (0.13) ( 1.64) c (0.79) (2.32) b (2.99) a (2.45) b ( 0.22) ( 0.28) (1.00) ( 1.93) c Panel B: Aggressive Quartile TA/assets 6.81 2.84 3.06 11.19 12.26 32.08 8.46 2.54 4.60 2.24 2.49 (1.89) c (0.87) (0.89) (4.09) a (4.09) a (8.17) a (3.60) a (2.14) b (1.66) c (2.29) b (1.54) Control-firm- adjusted TA 0.55 1.07 0.64 6.63 12.33 24.52 5.11 1.63 5.15 5.33 9.89 ( 0.05) ( 0.26) ( 0.10) (1.66) c (3.51) a (4.73) a (1.96) b ( 0.68) (1.26) (1.53) ( 0.86) DCA/assets 8.92 2.54 2.28 5.74 2.62 25.36 6.70 0.69 1.72 2.99 0.22 (2.44) b ( 0.75) ( 0.50) (2.55) b (0.76) (8.30) a (3.05) a (0.63) (0.58) (1.25) (0.19) Control-firm- adjusted DCA 1.66 4.61 2.10 8.14 5.80 23.41 4.55 1.91 1.61 8.08 12.31 ( 0.13) ( 0.99) ( 0.44) (1.40) (1.34) (5.29) a (1.73) c ( 0.81) (0.37) (1.64) ( 0.91)

84 CHOU, WANG, CHEN AND TSAI Table 3 (Continued) Year Relative to the Issue Year 5 4 3 2 1 0 1 2 3 4 5 Panel C: Conservative Quartile TA/assets 2.65 4.24 6.03 4.64 13.31 1.23 4.38 5.18 2.49 2.65 4.24 ( 0.70) (2.04) b (2.55) b (1.73) c (2.44) b ( 0.71) (2.98) a (3.41) a (1.99) b ( 0.70) (2.04) b Control-firm- adjusted TA 10.70 2.41 0.24 3.40 10.49 4.09 3.89 5.77 0.07 10.70 2.41 ( 1.38) (0.60) (0.06) ( 1.43) (1.45) ( 1.79) c (1.90) c (2.33) b (0.03) ( 1.38) (0.60) DCA/assets 2.26 2.42 3.22 1.57 7.60 7.94 0.13 1.50 0.87 2.26 2.42 ( 0.67) (1.02) (1.64) (0.60) (1.46) ( 4.16) a (0.08) (1.19) ( 0.68) ( 0.67) (1.02) Control-firm- adjusted DCA 13.21 1.63 3.39 3.34 7.91 7.18 0.07 2.92 1.11 13.21 1.63 ( 1.37) (0.38) ( 0.88) ( 1.27) (1.15) ( 3.23) a (0.03) (1.54) ( 0.49) ( 1.37) (0.38) Notes: This table reports the total current accruals (TA) and discretionary current accruals (DCA) (in percentage) over the 11-year period surrounding convertible bond offers (years 5 to years +5). We follow the methodology of Teoh et al. (1998a and 1998b) to estimate TA and DCA. Using the estimation sample, we run the following cross-sectional regression: TA j,t /AT j,t 1 = α0(1/at j,t 1) + α1( Sales j,t /AT j,t 1) + ε j,t, where TA is total current accruals, AT is total assets, Sales is the change in sales, j indicates the firms in the estimation sample, and t indicates the year. Nondiscretionary (or expected) current accruals for firm i, is estimated as: NDCAi,t = ˆα0(1/AT i,t 1 ) + ˆα1[( Salesi,t TRi,t )/ATi,t 1], where ˆα0 is the estimated intercept, ˆα1 is the slope coefficient for convertible bond issuing firm i, and TRi,t is the change in trade receivables for year t for convertible bond issuing firm i. TRi,t is subtracted from the change in sales to allow for the possibility of credit sales manipulation by the issuer. Discretionary current accruals, DCAi,t, for convertible bond issuing firm i for year t is then estimated as: DCAi,t = TAi,t /ATi,t 1 NDCAi,t. Aggressive issuers are the quartile of issuers with the highest DCA and conservative issuers are the quartile of issuers with the lowest DCA. T-tests are used to test the hypotheses that the means of TA and DCA will be equal to zero. T-statistics are reported in parentheses. a, b, and c represent 1%, 5% and 10% significance levels, respectively.

EARNINGS MANAGEMENT AND LONG-RUN UNDERPERFORMANCE 85 patterns, with the values in years 1 to +1 being significantly positive at the 5% level or better. The results of the aggressive quartile in Table 3 show that both the mean TA and DCA values are even higher, with the highest mean for TA being 32.08% and the highest for DCA being 25.36%. After adjusting for control firms, almost all the mean TA and DCA values from years 1 to +1 remain significantly positive at the 10% level or better, with the highest mean for TA being 24.52% and the highest for DCA being 23.41%. The conservative quartile in Table 3 reveals quite different results for TA and DCA, with no clear patterns being discernible for either the mean TA or control-firm adjusted TA; there is also no obvious pattern for DCA, although surprisingly, the mean DCA and control-firm adjusted DCA are significantly negative at the 1% level in year 0. Our results are consistent with the effects of managers advancing their accruals to increase reported net income in the convertible issuance period, similar to the findings for IPOs and seasoned equity offers (SEOs) as documented by Rangan (1998) and Teoh et al. (1998a and 1998b). (ii) Prediction of Post-offer Underperformance with Discretionary Current Accruals Around the Offer The median operating performance for our sample firms over the 11-year period surrounding the convertible bond offers (years 5 to +5) are reported in Table 4, with operating performance measured by OIBD/assets reported in Panel A, and operating performance measured by ROA reported in Panel B. As the table shows, convertible bond issuers experience deterioration in their operating performance surrounding the convertible bond offers; for example, there is a decline in the median OIBD/assets, from 14.98% in year 5, to 13.24% in year 0, with a further decline to 12.93% in year +5. A similar pattern is also displayed by the median ROA around the issuing year. Barber and Lyon (1997) and Lie (2005) suggest the use of a matched sample of firms, by pre-event performance levels, as a means of controlling for the mean- reversion tendency in performance measures, particularly where firms report either abnormally good or bad pre-event performance. As compared to control firms matched by industry and performance, both the median OIBD/assets and ROA for our sample firms are better in the pre-issue periods and worse in the post-issue periods, similar to the findings for SEOs and convertible bond offers reported by Loughran and Ritter (1997) and McLaughlin et al. (1998). However, the differences in performance between our sample firms and the matched firms are statistically significant in some years only for ROA. There are, however, differences between the aggressive and conservative quartiles in the patterns of matched-firm adjusted operating performance. In the aggressive quartile, OIBD/assets reaches its peak of 15.67% in year 2, and then declines to 11.78% in year +5. In years 2, +2 and +3, the differences in OIBD/assets between the sample firms and the matched firms are statistically significant at the 10% level or better. In contrast, in the conservative quartile, as compared to the control firms, no significant differences are found in OIBD/assets surrounding the offer year. This suggests that managers may be engaging in the upward manipulation of earnings prior to their convertible bond offers. A similar pattern of operating performance is found from our examination of ROA. In the aggressive quartile, ROA again reaches its peak of 7.27% in year 2, and then declines to 2.42% in year +5. In years 2 and +5, the differences in ROA between the

86 CHOU, WANG, CHEN AND TSAI Table 4 Operating Performance of Convertible Bond Issuers Fiscal Year Relative to the Issue Year 5 4 3 2 1 0 1 2 3 4 5 Panel A: OIBD/assets Whole Sample Medians (%) 14.98 14.69 14.69 15.36 14.49 13.24 13.06 12.63 13.03 13.15 12.93 sample Benchmark Medians (%) 13.83 14.08 14.08 14.28 14.42 13.31 12.95 12.88 14.48 13.12 12.84 Z-Stat. (Sample Benchmark) (1.37) (1.17) (1.17) (1.64) (0.02) ( 0.22) ( 0.44) ( 0.43) ( 1.11) ( 0.14) ( 0.58) N 134 219 219 283 302 279 247 212 176 152 137 Aggressive Sample Medians (%) 15.65 14.69 14.70 15.67 14.67 12.16 11.07 8.79 11.39 11.68 11.78 quartile Benchmark Medians (%) 11.59 14.08 14.08 13.61 14.57 13.17 11.84 13.69 14.51 16.40 16.24 Z-Stat. (Sample Benchmark) (1.16) (1.17) (0.42) (1.93) c (0.11) ( 0.87) ( 1.00) ( 2.69) a ( 1.91) c ( 1.58) ( 1.44) N 26 49 49 71 76 67 59 49 39 32 31 Conservative Sample Medians (%) 13.40 14.70 14.00 14.55 14.06 13.25 13.15 12.73 13.18 12.51 12.56 quartile Benchmark Medians (%) 11.77 14.08 12.42 14.19 14.06 13.06 14.08 11.95 13.58 11.52 12.14 Z-Stat. (Sample Benchmark) (0.47) (0.42) (0.57) ( 0.33) ( 0.05) (0.05) ( 0.03) ( 0.27) ( 0.07) ( 0.19) ( 0.58) N 29 49 49 71 75 68 59 50 42 39 33

EARNINGS MANAGEMENT AND LONG-RUN UNDERPERFORMANCE 87 Table 4 (Continued) Fiscal Year Relative to the Issue Year 5 4 3 2 1 0 1 2 3 4 5 Panel B: ROA Whole Sample Medians (%) 5.25 5.14 5.14 5.32 5.50 4.51 3.74 3.23 3.49 3.81 3.17 sample Benchmark Medians (%) 4.74 5.07 5.07 4.70 4.89 3.77 3.62 3.55 4.19 3.76 4.00 Z-Stat. (Sample Benchmark) (0.94) (1.03) (1.03) (1.96) b (1.89) c (1.43) ( 0.61) ( 0.73) ( 1.58) (0.07) ( 1.37) N 134 219 219 283 302 279 247 212 176 152 137 Aggressive Sample Medians (%) 5.87 5.14 6.64 7.27 6.36 5.09 3.35 2.08 3.49 3.84 2.42 quartile Benchmark Medians (%) 2.80 5.07 4.99 4.58 5.91 4.29 3.37 3.48 3.12 4.60 4.52 Z-Stat. (Sample Benchmark) (1.71) c (1.03) (1.25) (3.18) a (1.18) (0.68) (0.52) ( 1.61) (0.34) ( 0.37) ( 2.27) b N 26 49 49 71 76 67 59 49 39 32 31 Conservative Sample Medians (%) 5.34 6.64 3.68 4.72 5.11 4.25 3.75 4.01 3.74 3.60 2.54 quartile Benchmark Medians (%) 5.63 4.99 2.62 3.97 4.16 3.58 4.06 4.16 4.54 3.75 3.25 Z-Stat. (Sample Benchmark) ( 0.23) (1.25) (0.68) (0.12) (0.98) (0.15) ( 0.74) ( 0.84) ( 1.10) ( 0.23) ( 0.06) N 29 49 49 71 75 68 59 50 42 39 33 Notes: This table presents median operating performance of 312 convertible bond issuers between 1981 to 1998. Panels A and B report OIBD/assets and ROA for the whole sample, the aggressive quartile, and the conservative quartile, respectively. OIBD/assets is operating income before depreciation plus interest income deflated by fiscal year-end total assets. ROA is net income divided by fiscal year-end total assets. The number of observations (N ) varies because of data unavailability. The aggressive and conservative quartiles contain firms with the highest and smallest discretionary current accruals (DCA) at year 0, where year 0 is the fiscal year of the convertible offering. T-statistics are reported in parentheses. a, b, and c represent 1%, 5% and 10% significance levels, respectively.

88 CHOU, WANG, CHEN AND TSAI sample firms and the matched firms are statistically significant at the 5% level or better. This pattern stands in sharp contrast to the performance of the conservative quartile, where, as compared to the control firms, no significant differences in ROA are found. (i) Abnormal Stock Returns 4. PREDICTION OF POST-OFFER STOCK RETURNS The abnormal post-issue stock return performance levels for the whole sample, as well as those for the aggressive and conservative quartiles, are reported in Table 5 for various holding periods. These are measured by buy-and-hold abnormal returns (BHAR) and cumulative abnormal returns (CAR) using size and book-to-market matched firms as the benchmarks. The long-run mean buy-and-hold abnormal returns on security i over the holding period T are defined as: [ ] BHAR T = 1 N T T (1 + R i,t ) (1 + R benchmark,t ), (5) N i=1 t=0 where R i,t is the monthly raw return for firm i in month t, and R benchmark,t is the monthly raw return for the benchmark in month t. Holding-period stock returns are calculated for different post-offer period lengths, ranging between one and five years, with our computation of the returns being based upon the CRSP monthly returns file. CAR T, the cumulative mean abnormal return at month T, is defined as the sum of monthly abnormal returns over T months. The cumulative mean abnormal returns (CAR) are defined as: [ ] T 1 N CAR T = (R i,t R benchmark,t ), (6) N t=0 i=1 where R i,t and R benchmark,t are as previously defined. As Table 5 shows, abnormal returns reveal a strong pattern of underperformance for the aggressive quartile vis-à-vis the conservative quartile. Measured by BHAR, this underperformance ranges between 8% ( 7.17% vs. 1.01%) for the one-year holding period and 44% ( 27.21% vs. 16.91%) for the five-year holding period. Measured by CAR, the underperformance similarly ranges between 8% for the oneyear holding period and 42% for the five-year holding period. The differences between the aggressive and conservative quartiles in the five-year holding period returns of both BHAR and CAR are statistically significant at the 1% level. 9 Our results suggest that more aggressive earnings management in the issue year is an effective predictor of inferior post-issue stock return performance. The OLS regressions on post-issue stock return performance on DCA in the issue year are presented in Table 6, for various holding periods. The model is stated as: AR i,t = β 0 + β 1 DCA i,t + β 2 ln(mve) i,t + β 3 ln(bm) i,t + β 4 CP i + ε i,t, (7) t=0 9 For simplicity, these results are not shown in Table 5.

EARNINGS MANAGEMENT AND LONG-RUN UNDERPERFORMANCE 89 Table 5 Long-Run Mean Abnormal Stock Returns Subsequent to Convertible Bond Offers Buy-and-Hold (BHAR) Cumulative Abnormal (CAR) Holding Whole Aggressive Conservative Whole Aggressive Conservative Period Sample Quartile Quartile Sample Quartile Quartile +1 Mean Abnormal Return 3.10 7.17 1.01 2.62 6.91 0.75 Cross-sectional t-stat 1.68 c 1.70 c 0.23 1.42 1.64 0.19 Skewness-adjusted t-stat 1.63 1.63 0.24 1.43 1.65 c 0.19 +2 Mean Abnormal Return 0.86 13.44 6.64 0.42 13.45 7.49 Cross-sectional t-stat 0.30 2.18 b 1.01 0.15 2.15 b 1.24 Skewness-adjusted t-stat 0.30 1.80 c 1.10 0.15 2.20 b 1.22 +3 Mean Abnormal Return 0.68 13.49 11.02 0.02 16.31 8.03 Cross-sectional t-stat 0.18 1.71 c 1.26 0.01 1.83 c 0.98 Skewness-adjusted t-stat 0.18 1.53 1.37 0.01 1.84 c 0.97 +4 Mean Abnormal Return 2.91 22.21 11.15 1.41 13.58 11.43 Cross-sectional t-stat 0.69 2.71 a 1.23 0.33 2.36 b 1.34 Skewness-adjusted t-stat 0.67 2.21 b 1.34 0.33 2.44 b 1.34 +5 Mean Abnormal Return 2.94 27.21 16.91 0.10 27.26 15.18 Cross-sectional t-stat 0.63 2.60 a 1.73 c 0.02 2.58 a 1.48 Skewness-adjusted t-stat 0.62 2.12 b 1.85 c 0.02 2.53 b 1.44 Notes: This table presents both matched-firm-adjusted buy-and-hold returns (BHAR) and cumulative abnormal returns (CAR) following convertible bond offers for various holding periods, where matching firms are chosen on the basis of size and the book-to-market ratio. The long-run mean buy-and-hold abnormal returns on security i over the holding period T are defined as: ] [ T N T (1 + Ri,t ) (1 + Rbenchmark,t) BHART = 1 N i=1 t=0 t=0 where Ri,t is the monthly raw return for firm i in month t, and Rbenchmark,t is the monthly raw return for the benchmark in month t. Holding-period stock returns are calculated for different lengths of post-offering periods. The cumulative abnormal return through month T, CART, is defined as the sum of monthly abnormal returns over T months. Cumulative mean abnormal returns (CAR) are defined as: [ ] T 1 N CART = (Ri,t Rbenchmark,t), N t=0 i=1 where Ri,t and Rbenchmark,t are as defined previously. The sample consists of 312 convertible bond offers during the period from 1981 to 1998. We use crosssectional t-statistics and skewness-adjusted t-statistics to test the significance of the mean values of BHAR and CAR. Aggressive issuers are the quartile of issuers with the highest DCA and conservative issuers are the quartile of issuers with the lowest DCA. a, b, and c represent 1%, 5% and 10% significance levels, respectively.,

90 CHOU, WANG, CHEN AND TSAI Table 6 Ordinary Least Squares Regressions of Abnormal Stock Returns Subsequent to Convertible Bond Offers Independent Variables Holding Period Intercept DCA ln(mve) ln(bm) CP Adj-R 2 F-value Panel A: Matched-Firm-Adjusted BHAR 1-year 0.09 0.23 0.01 0.01 0.14 0.01 1.67 ( 0.59) ( 2.21) b ( 0.70) ( 0.39) (1.10) 2-year 0.01 0.23 0.01 0.03 0.06 0.00 1.00 (0.05) ( 1.54) ( 0.32) (0.67) (0.31) 3-year 0.47 0.32 0.01 0.07 0.41 0.01 1.70 (1.52) ( 1.60) ( 0.09) (1.27) ( 1.65) 4-year 0.27 0.54 0.01 0.07 0.22 0.02 2.21 (0.80) ( 2.47) b ( 0.05) (1.19) ( 0.81) 5-year 0.49 0.67 0.01 0.10 0.46 0.04 3.57 (1.36) ( 2.89) a (0.27) (1.64) ( 1.63) Panel B: Matched-Firm-Adjusted CAR 1-year 0.05 0.19 0.01 0.01 0.11 0.00 1.25 ( 0.34) ( 1.89) c ( 0.73) ( 0.22) (0.85) 2-year 0.13 0.25 0.01 0.04 0.06 0.00 1.14 (0.53) ( 1.63) ( 0.04) (1.08) ( 0.31) 3-year 0.58 0.42 0.02 0.03 0.40 0.02 1.98 (1.92) c ( 2.12) b ( 0.77) (0.69) ( 1.68) c 4-year 0.63 0.63 0.02 0.04 0.46 0.03 2.85 (1.83) c ( 2.84) a ( 0.63) (0.68) ( 1.70) c 5-year 0.90 0.67 0.02 0.05 0.71 0.04 3.49 (2.41) b ( 2.78) a ( 0.54) (0.86) ( 2.43) b Notes: This table presents ordinary least squares regressions of post-issue stock return performance on discretionary current accruals (DCA) over various holding periods. The sample comprises of 312 convertible bond offers between 1981 to 1998. The regression model used is as follows: AR i,t = β 0 + β 1 DCA i,t + β 2 ln(mve) i,t + β 3 ln(bm) i,t + β 4 CP i + ε i,t, where AR is abnormal return as measured by BHAR or CAR, DCA is discretionary current accruals in the issue year, ln(mve) is issuer s logged market value of equity, ln(bm) is logged book-to-market ratio, CP is conversion probability in the issue year, and ε i,t is the error term. The dependent variable in Panel A is the matched-firm-adjusted buy-and-hold mean abnormal return (BHAR), where matching firms are chosen on the basis of size and the book-to-market ratio. The dependent variable in Panel B is the matched-firm-adjusted cumulative mean abnormal return (CAR). T-statistics are reported in parentheses. a, b, and c represent 1%, 5% and 10% significance levels, respectively. where AR represents abnormal returns as measured by BHAR or CAR; DCA are the discretionary current accruals in the issue year; ln(mve) is the issuer s logged market value of equity, ln(bm) is the logged book-to-market ratio, CP is the conversion probability in the issue year, and ε i,t is the error term. The dependent variable in Panel A is the matched-firm-adjusted BHAR. We include the issuer s logged equity market value, ln(mve), and logged book-to-market ratio,

EARNINGS MANAGEMENT AND LONG-RUN UNDERPERFORMANCE 91 ln(bm), as control variables in the regression, since underperformance following convertible bond offers may be attributable to a particular subset of firms. Furthermore, the conversion probability (CP) is also included to test how the convertible bond conversion mechanism affects post-issue stock performance. The results in Panel A indicate that the coefficients of DCA are all negative, with significantly negative coefficients at the 5% level or better for the one-, four- and fiveyear holding periods. 10 Panel A also shows that the coefficients of CP are positive for the one- and two-year holding periods, but negative for the other three holding periods. Furthermore, CP has a statistically insignificant relationship with post-issue stock performance for all five holding periods. In other words, it is apparent that the market cannot clearly discern whether the economic substance of convertible bond issues is debt-like or equity-like around the issue date. The results in Panel A suggest that convertible issuers with aggressive DCA levels in the issue year have significantly inferior performance after the offer, consistent with the results reported in Table 5. We also use the matched-firm-adjusted CAR as the dependent variable in order to check the robustness of our results; our findings are reported in Panel B. Similar to the results in Panel A, post-issue stock return performance is significantly poorer for convertible issuers with more aggressive earnings management in the issue year. The results hold, irrespective of the length of the holding period; therefore, our findings are robust to alternative measures of computation of the abnormal returns. 11 (ii) Abnormal Returns Based on the Fama-French and Carhart Models We estimate abnormal stock returns for the aggressive and conservative quartile issuers based upon the three-factor model of Fama and French (1993); the results are reported in Table 7. We adopt the calendar-time portfolio approach, as suggested by Fama (1998) and Mitchell and Stafford (2000), using value-weighted monthly abnormal stock returns. 12 Specifically, monthly convertible bond offer portfolios are formed in calendar time using the following regression model: R p,t R ft = α p + β p (R mt R ft ) + s p SMB t + h p HML t + e p,t, (8) where R p,t is the return on portfolio p in month t; R ft is the return on one-month Treasury bills in month t; R mt is the return on a market index in month t; SMB t is the difference in the returns for a portfolio of small and big stocks in month t; HML t is the difference in returns for portfolios of high and low book-to-market stocks in month t; and ε p,t is the error term for portfolio p in month t. The estimation of the intercept coefficient (α p ) provides a test of the null hypothesis of zero average abnormal returns. The estimated alpha coefficients for the aggressive quartile issuers in Panel A are significantly negative at the 1% level for various holding periods, ranging between one and five years. The monthly abnormal returns range between 1.09% and 1.68%, thereby implying annualized abnormal returns of between 12.32% and 18.40%. 10 We also compute t-statistics with heteroskedasticity-consistent standard errors (White, 1980), from which we find that the results are similar. 11 The results are also similar for value-weighted and equally-weighted market-adjusted CAR; thus, these results are not reported here. 12 We also use equally-weighted monthly abnormal stock returns, and find that the conclusions of this study remain unchanged.