Determinants Of The ReturnsEarnings Correlation

скачати

Determinants Of The Returns-Earnings Correlation Essay, Research Paper

Determinants of the Returns-Earnings CorrelationThis paper is based on a chapter of the author’s dissertation at Columbia University. The author thanks her committee members James Ohlson, Trevor Harris, Ashiq Ali, and Bruce Lehmann for their guidance. This paper has benefitted from the comments provided by the participants of the Ohio State University colloquium, the Third Conference on Financial Economics and Accounting, Georgetown University, and from the helpful suggestions of Stephen Penman, Ram Ramakrishnan, Stephen Ryan, Douglas Schroeder and anonymous reviewers. Financial support from the Ernst and Young Foundation is gratefully acknowledged. Abstract. The weak correlation between accounting earnings and security returns documented by numerous empirical studies is an issue of concern in current accounting research. Given that price is determined not solely by accounting earnings but also by other sources of information about future earnings, this paper focuses on the relation between earnings and other information to understand the returns-earnings association. The analysis indicates that current earnings exhibit high explanatory power for returns if they correlate with expected future earnings (or with other information which reflects expected future earnings). A high price-earnings (P/E) ratio coupled with a high return on equity (ROE) can ex ante indicate earnings growth, and the earnings of firms with these attributes are positively correlated with future earnings. The high growth subset obtains an impressive returns-earnings R2 of 31 percent and an earnings coefficient of 6.17, demonstrating that it is possible to identify firms whose earnings are strongly correlated with returns using a parsimonious set of firm characteristics. Accounting earnings are a key decision variable for analysts and investors, yet empirical researchers have found that earnings are only weakly correlated with contemporaneous security returns.1 Security price impounds information as it arrives, while the accounting process typically records it with a lag. Thus, price, which reflects future earnings expectations, is determined not solely by accounting earnings, but also by other sources of information about future earnings. Whereas for broad cross-sections, the returns-earnings association is weak, this paper shows that by focusing on the relation between earnings and other information it is possible to identify ex ante a subset of firms whose earnings are strongly correlated with returns. The analysis reveals that earnings exhibit a high explanatory power for returns if they are positively correlated with other value-relevant information. The intuition underlying this observation is as follows: Price reflects expected future earnings; investors use current accounting earnings and other sources of information to arrive at their expectations of future earnings. If current earnings also reflect expected future earnings, then investors would place greater weight on current earnings to determine price. Current earnings, therefore, exhibit high explanatory power for returns if they correlate with future earnings (or with other information which reflects future earnings). The above observation then raises a question: Is it possible to ex ante identify firm characteristics which indicate a positive correlation between current and future earnings? A heuristic argument indicates that the Price/Earnings (P/E) ratio and the return on equity (ROE) serve as indicators of future earnings expectations. A high P/E ratio indicates either that current earnings are normal but future earnings expectations are high, or that current earnings have a negative transitory component.2 A high P/E ratio coupled with low current profitability (ROE) indicates an increase in earnings of the subsequent period, which will be temporary in nature since it results from the reversal of the negative transitory component. On the other hand, a high P/E ratio with high current profitability indicates an earnings increase of a permanent nature in the next period, which will be a reflection of future earnings growth. The high P/E-high ROE criterion thus identifies ex ante a subset of the population of firms whose earnings are strongly correlated with future earnings expectations and hence with security returns.3 At the other extreme, the high P/E-low ROE criterion characterizes low growth firms whose earnings are weakly correlated with returns. For the high growth subset, the R2 from a regression of returns on earnings over the annual window is 31 percent and the earnings coefficient is 6.17, both considerably higher than what is normally observed for broad cross-sections. Given the apparent interest in the returns-earnings R2 in the literature, the R2 obtained by this small subset of firms is impressive. The contribution of this paper, however, lies in the process of identification of these firms with the aid of a parsimonious set of firm characteristics. The high P/E-high ROE criterion selects cases where current earnings, for a given firm at a given time, provide a good indication of investors’ expectations of future earnings. It is worthwhile to note that this correlation between current and future earnings need not relate to the time-series properties of past earnings. In fact, firms do not remain in the same P/E-ROE subsets for a sustained length of time, which suggests that the time-series “persistence” in past earnings does not translate into the future. The results thus reinforce the argument made by Penman (1992b) that, as information changes over time, so do pricing multipliers, thereby limiting the usefulness of estimating earnings response coefficients for an individual firm over time. The rest of the paper is organized as follows. The next section discusses the model and the determinants of the returns-earnings correlation. The following section analyzes the role of the P/E ratio and the ROE as indicators of growth. The fourth section describes the data, sample selection, and research design. Section five reports the results and is followed by concluding remarks in the final section. The returns-earnings correlation component analysisThis section analyzes the components of the returns-earnings correlation metric by exploiting one representation of the returns-earnings model, where returns equal earnings, scaled by beginning price, plus the (scaled) change in unrecorded (economic) goodwill. This characterization allows for a divergence between market and book values (termed unrecorded goodwill) which results from leads or lags in accounting recognition of events. Assuming the clean surplus relation (change in book value equals earnings minus dividends), we can expressyt = xt + g (1a)whereyt = (Pt + dt – Pt-1)/Pt-1 = market returns over the period (t-1,t) including dividendsxt = Xt/Pt-1 = accounting earnings for the period t-1 to t (Xt), normalized by the beginning market value (Pt-1)g = )gt/ Pt-1 =(gt – gt-1)/ Pt-1 = [(Pt - BVt) - (Pt-1 - BVt-1)]/ Pt-1 = the change in goodwill from period t-1 to t, normalized by the beginning market valuedt = dividends paid at date tBVt = book value at date tFrom the returns-earnings expression (1a), one observes that, as the difference between market value and book value changes during the year (g), the perfect correlation between returns and earnings becomes distorted. Applying equation (1a) to a regression framework, we can express the regression of returns on (scaled) earnings asyit = ” + $xit + ,it (1b)where i denotes firm i, ” and $ are the regression parameters, and ,it is a random error term.4 Given the clean surplus relation and that Cov[xt,yt] = Cov[xt,g] + Var(xt) [from the definition of g in (1a)], the returns-earnings correlation metric is divided into components: The first term on the right hand side is the covariance component which captures the ability of current earnings (levels) to explain the change in goodwill. Cov[xt,g] > 0 implies > 1 and results in a higher returns-earnings correlation. The change in goodwill captures value-relevant information reflecting future earnings, and to the extent earnings are positively correlated with that information, both the R2 and are higher. On the other hand, if = 1 (Cov[xt,g]=0), then the correlation depends on the ratio of earnings variance to the variance of returns — the variance component. The higher the variance of earnings relative to the variance of returns, the higher is the R2. Intuitively, if earnings are less predictable (high variance) the greater will be the effect of realized earnings on returns.5Equation (1b) uses earnings levels as the explanatory variable for returns. Given the interest in the specification of the earnings variable in the literature, the effect of an alternative earnings specification on the returns-earnings R2 is considered next. Ohlson and Shroff (1992) demonstrate that the basic earnings (levels) specification can be improved by replacing earnings levels by the residual (ut) from the regression of xt = Xt/Pt-1 on the beginning E/P ratio — (Xt-1/Pt-1).6 The returns-earnings regression with the improved specification can be expressed asyt = ” + $ut + ,t (2)Given the definition of g, and assuming that Cov[yt,Xt-1/Pt-1] = 0, we obtain(This follows because Cov[yt,ut] = Cov[yt,(xt-T0-T1Xt-1/Pt-1)] = Cov[yt,xt] = Cov[xt,g] + Var(xt), where T0 and T1 are the estimated parameters of the regression of Xt/Pt-1 on Xt-1/Pt-1.)Interestingly, this result shows that the numerators of the two components of Dy,x and Dy,u remain unchanged even with the improved earnings specification. Regardless of the earnings specification employed, the covariance between the change in goodwill and earnings levels (and not ut), and the variance of earnings levels (not ut) still remain the factors determining the R2. Thus, if Cov[yt,Xt-1/Pt-1]=0, then (Dy,u-Dy,x) measures the effect of specification improvement. Hence, it follows that the weak returns-earnings correlation for broad cross-sections of firms could result from (i) low earnings variance relative to the variance of returns; and/or (ii) negative covariance between earnings and the change in goodwill (that is, The historical (scaled) earnings variance is expected to predict the earnings variance in the next period and is therefore used as one of the partitioning attributes. Of special interest to this paper is the development of simple criteria which can indicate ex ante the sign of the covariance between earnings and the change in goodwill. Since the change in goodwill captures expectations of future earnings (via other value-relevant information), if current earnings reflect expected future earnings, a positive covariance between earnings and the change in goodwill results. The next section develops heuristic arguments which indicate that the P/E ratio and the ROE at the beginning of the year jointly serve as indicators of future earnings growth. P/E ratio and ROE as indicators of growthA simplistic view of pricing would suggest that a firm sells for a multiple of earnings. Hence, the higher the P/E ratio, the larger is the value attached by the market to a dollar of earnings, and the higher is the expectation of future earnings. Beaver and Morse (1978), however, suggest that a high P/E ratio identifies earnings which are temporarily depressed but which will increase in the future. While both views are plausible, they need not be true in all cases.7 If current profitability (ROE) is below normal, then a high P/E indicates a negative transitory component in current earnings which will reverse in subsequent years and result in a temporary earnings increase. But if current profitability is high, then a high P/E indicates future growth in earnings of a permanent nature.8To evaluate how combinations of beginning P/E ratio and ROE can differentiate the explanatory power of earnings for returns, the discussion that follows relies on some prior empirical and theoretical results. First, Penman (1991) provides empirical evidence that high (low) ROE firms with low (high) book value/price (B/P) ratios have considerably higher (lower) ROE in subsequent years than firms with high (low) ROE and high (low) B/P ratios. Second, Ohlson and Shroff (1992) show theoretically that a sufficiently positive serial correlation in earnings changes results in a higher returns-earnings slope coefficient (implying a higher correlation between earnings and the change in goodwill) than a negative serial correlation in earnings changes.9 Using these results, the sign of the covariance between earnings and the change in goodwill is predicted for combinations of P/E and ROE individually. High P/E-High ROEThis sample screen includes firms with low B/P ratios. Given high current ROE, a high P/E ratio indicates subsequent earnings increase of a permanent nature, which will lead to a positive change in goodwill.10 Since firms with high ROE and low B/P ratio are observed to have high subsequent ROE, current earnings changes of these firms are expected to have high positive correlation with future earnings changes, which suggests that current earnings will be positively correlated with the change in goodwill. High P/E-Low ROEThis represents one of the two low growth screens with medium B/P ratios. Since earnings are temporarily depressed, an increase in earnings is expected next period. Because of the negative transitory component in current earnings, next-period earnings will revert back to normal, and hence the correlation between current and future earnings changes is likely to be negative, suggesting a negative correlation between earnings and the change in goodwill. Low P/E and Low ROEFirms with an anticipated decline in earnings and high B/P ratios form this screen. In the next period, earnings are expected to decrease, and the change in goodwill will be negative. Since these low ROE firms are observed to have low ROEs in subsequent years also, current earnings changes will be positively correlated with future earnings changes, suggesting a positive correlation between earnings and the change in goodwill. Low P/E and High ROEThis screen includes low growth firms with medium B/P ratios. Since their current earnings are abnormally high, a decrease in earnings will be experienced in the next period. Because of the presence of high transitory earnings which will revert back to normal in subsequent years, the correlation between current and future earnings changes is likely to be negative, leading to a negative correlation between earnings and the change in goodwill. Panel A of Table 1 summarizes the predicted characteristics of the P/E-ROE combinations in a 2×2 matrix. Panels B and C report some informative summary statistics for the P/E-ROE combinations based on the top and the bottom third of firms ranked on the respective attribute. For the most part, the summary numbers for the four cells are consistent with the predictions in Panel A and the detailed descriptions in this section. Insert Table 1 here. For the purpose of this study firms with high P/E-high ROE and high P/E-low ROE are selected as two extreme attribute screens which are expected to differentiate negative from positive covariance between earnings and the change in goodwill. Hence, earnings of the high P/E-high ROE screen are expected to exhibit substantially higher explanatory power for returns compared to earnings of the high P/E-low ROE screen. Although both high P/E-high ROE and low P/E-low ROE screens predict positive covariance between earnings and the change in goodwill, the former category is selected because it includes firms which are likely to use conservative accounting practices (indicated by low B/P ratios) due to which the covariance is expected to be higher. For firms which use conservative accounting techniques, accounting recognition would have been slow in the past, but eventually value must surface through recognition of higher future earnings with which current earnings would be positively correlated. A similar argument applies for preferring the high P/E-low ROE category as the extreme subset with very low explanatory power of earnings for returns.11 Data, sample selection and research designThe sample, covering the years 1970-87, consists of all NYSE-AMEX firms which appear on the 1989 Compustat Annual Industrial file, with December 31 as the fiscal year end. Earnings per share (eps), dividends per share (dps), book value per share (bvps), and factors to adjust for stock splits and stock dividends are obtained from the Compustat file. Annual security returns are calculated by compounding daily returns, obtained from the 1989 CRSP Daily Returns file, for the twelve months ending three months after the fiscal year end. Prices at the beginning of the period of accumulation of returns are obtained from the 1989 CRSP Daily Master file. Of the sample of 725 Compustat-CRSP firms selected, firms which have missing data in any year of the study period are deleted for that year only. The following procedure is adopted to obtain the high growth (HG) and the low growth (LG) screens for testing their differential effect on the returns-earnings association. Firms are ranked into three groups on the basis of their E/P ratios at the beginning of the year, [Xt-1/Pt-1].12 Next, they are ranked (independently) into three groups based on their ROE, [Xt-1/BVt-1], at the beginning of the year. The ranking and group formation are repeated at the beginning of each test year so that the composition of the E/P and ROE groups changes annually. Firms are placed in four earnings variance groups at the beginning of each test period by ranking them on the variance of scaled earnings, [Xt/BVt-1] – that is, eps divided by bvps at the beginning of the period, calculated over five prior years. Observations with beginning bvps less than 0.1 are deleted. The requirement of five prior years of data to measure the variance results in the elimination of the years 1970-1974 from the test sample. The extreme groups, EV1, representing firms with the lowest (ex ante) earnings variance, and EV4, representing firms with the highest (ex ante) earnings variance, are the focus of the analysis. Firms belonging, in the same year, to (i) the lowest two E/P groups; (ii) the lowest ROE group; and (iii) EV1 form the LG screen.13 Firms belonging, in the same year, to (i) the lowest E/P group; (ii) the highest ROE group; and (iii) EV4 form the HG screen. Empirical resultsRanking on P/E and ROETable 2 reports results of pooled returns-earnings regressions for extreme P/E and ROE groups. As expected, grouping firms on extreme P/E ratios or extreme ROE alone does not substantially differentiate the explanatory power of earnings for returns. Panel A shows that the R2s of high, medium, and low P/E groups are 5, 6, and 5 percent, respectively, and each group has Insert Table 2 here. As predicted in the third section, when the high P/E group is also matched with extreme ROEs, the regression results for the high P/E-high ROE and the high P/E-low ROE groups differ substantially. Panel C shows that the high P/E-high ROE group achieves an R2 of 20 percent in contrast to 4 percent obtained by the high P/E-low ROE group. Restricting the high P/E-high ROE subset to firms with high prior earnings variance and the high-P/E-low ROE subset to firms with low prior earnings variance further magnifies the difference in the explanatory power of earnings for returns. Since prior earnings variance is high for the high growth firms, their earnings are less predictable and are expected to have a higher effect on the R2. Ranking on P/E, ROE, and Prior Earnings VarianceSome descriptive statistics relating to the overall sample and the high and low growth screens are reported in Panel A of Table 3. The LG screen, representing firms with low to medium E/P, low ROE and low earnings variance, obtains a high mean (median) B/P ratio of 1.223 (1.176), and mean (median) change in goodwill of 0.051 (0.016). The HG screen, including firms with low E/P, high ROE and high earnings variance, obtains a low mean (median) B/P ratio of 0.345 (0.326), and mean (median) change in goodwill of 0.123 (0.049). Insert Table 3 here. As expected, Panel B shows that the serial correlation in earnings changes is negative and equals -.023 (Spearman -.075) for the LG screen and is a positive .051 (Spearman .217) for the HG screen. Similarly, the slope coefficient obtained from a regression of )Xt/Pt-1 on )Xt-1/Pt-2 (though insignificant) is negative and equals -0.066 for the LG screen and is a positive 0.014 for the HG screen.

Results of the pooled cross-sectional, time-series regression (1b) covering the period 1975-87, are reported in Panel A of Table 4.14 A distinct difference in the performances of the two extremes is evident from their R2s — 1 percent for the LG vis-a-vis 31 percent for the HG screens.15An examination of the correlation components for the HG and LG screens reveals that:(i) The variance effect does not contribute to the high correlation achieved by the HG screen. In fact, the ratio of SD(xt) to SD(yt) is lower (0.09) compared with (0.25) that for the LG screen.16(ii) The high returns-earnings correlation is mostly due to the high equal to 6.17 obtained for the HG screen (0.42 for the LG screen). As expected, for the LG screen is less than one, and Cov(Corr)[xt,g] is negative at -0.0022 (-0.14). For the HG screen, is substantially greater than one, and Cov(Corr)[xt,g] is positive at 0.0121 (0.47). The difference between the covariances obtained for the two screens suggests that beginning ROE and E/P can jointly indicate the ability of earnings to explain the change in goodwill. Insert Table 4 here. (iii) Panel B reports the results of regression (1b), using an alternative earnings specification suggested by Ohlson and Shroff (1992). Earnings changes and levels are used as two independent variables explaining returns (as estimated by Easton and Harris [1991]) which is shown to be equivalent to using the minimum variance earnings variable (ut) explained in the second section. For the HG screen the R2 remains unchanged at 31 percent and increases to 2 percent for the LG screen. To summarize, the above findings indicate that the low overall R2 for the one-year window is due to the negative covariance between earnings and the change in goodwill. The negative covariance implies the low value relevance of earnings which are priced by the market at a rate less than one. A firm-attribute screen, which is expected to reflect the explanatory power of next-period earnings with respect to returns, is shown to achieve a maximum R2 of 31 percent for the one-year window, which is comparable to that obtained for the five-year horizon (as documented by Easton, Harris, and Ohlson [1992]). Further, using an alternative earnings specification with both earnings changes and levels does not result in a substantial increase in the R2 for either the extreme subsets or the overall sample. This suggests that although the earnings specification issue is important to the test of the returns-earnings relation, using the basic levels specification instead of a “better” alternative is not the major factor contributing to the observed low R2s. Concluding remarksThis paper demonstrates that current earnings, which are correlated with future earnings growth, exhibit a high explanatory power for returns. Extreme P/E ratio and ROE combinations serve as good indicators of future earnings growth and are able to differentiate the value relevance of earnings. The high market response to a dollar of earnings observed for the high P/E-high ROE screen could possibly imply market mispricing. A second explanation for this result could be traced to risk as perceived by the market.17 The current analysis does not attempt to verify these competing explanations; the results merely demonstrate the ability of beginning P/E and ROE to identify firms for which the market response to a dollar of earnings in the next year is likely to be high, thereby indicating that their earnings are highly relevant to valuation. Underlying this finding, nevertheless, is the implication for earning excess returns if the direction and magnitude of the change in earnings were predictable at the beginning of the year. This paper constructs one scenario where readily available financial numbers like P/E and ROE are used as indicators of future earnings growth. This does not eliminate the possibility of replicating the high R2 results by using other firm attributes such as “quality of earnings” or accounting techniques to identify the high earnings growth subset. Indeed, there may even be a link between these attributes and the high P/E-high ROE criterion. The latter criterion identifies firms with low B/P ratios which suggests that these firms may be employing conservative accounting practices, so that earnings which are thus computed may have high value relevance.18 Endnotes1. See Bernard (1989) and Lev (1989) for a review of studies on the returns-earnings relation. 2. In other words, the high P/E may be sustainable (or permanent) or it may be a transitory phenomenon. 3. Intuitively, the high P/E-high ROE criterion links a firm’s accounting with its current economic situation. The criterion identifies firms which have high market expectations, and which measure earnings and book values with conservative accounting methods (low book value/price ratios). For these high growth firms, since accounting recognition has been slow, one expects lower current earnings to generate higher future earnings with which they would be positively correlated.4. For notational simplicity, the firm subscript i is deleted for the rest of the paper. 5. Consistent with the standard information-theoretic result, if earnings explain returns with noise, then the higher the variance of the signal (earnings) relative to the variance of the noise, the higher is the effect of earnings on returns. 6. Compared to alternative earnings variables (such as earnings changes or levels), ut obtains the highest correlation with returns, provided that returns are uncorrelated with the beginning E/P ratio. 7. Penman (1992a) discusses the integration of these opposing views. 8. This argument becomes clear using the model in Ohlson (1995), where setting dt=Xt, the P/E ratio can be expressed as and the ratio of expected next-period earnings to current earnings as where (1, (2, (3, 81, and 82 are the respective coefficients, and Lt captures information other than earnings and book value which reflects investors’ expectations of future earnings. A high P/E ratio may reflect high BVt/Xt (low ROE) and/or high Lt/Xt. High P/E coupled with high ROE (low BVt/Xt) implies high expectation of future profitability (Lt), whereas low P/E and low ROE may imply negative Lt, or expected decline in future profitability. High P/E-low ROE and low P/E-high ROE both reflect low Lt. 9. Also, Shroff (1995) shows theoretically that the higher the covariance between current earnings and the component of future earnings incorporated in returns, the higher is the covariance between earnings and the change in goodwill (and hence the slope coefficient), and the higher the returns-earnings R2. 10. Intuitively, whereas a dollar increase in earnings of a permanent nature increases the book value by a dollar, it increases price by a factor equal to the price multiple, resulting in a positive change in goodwill. 11. Furthermore, for low P/E firms, the limited liability constraint (or lower bound on prices) would confound a clear prediction of results. 12. The E/P ratio is used as the partitioning variable instead of the P/E ratio to minimize the small denominator problem. Firms with eps less than 0.1 in the previous year are excluded from the primary grouping on P/E and ROE. Tables 1 and 2 report results for this group of firms separately. 13. The results for the LG screen, using only the lowest E/P group instead of the lowest two, are substantially the same as those reported, but the inclusion of the lowest two E/P groups obtains a final sample that is comparable in size to the HG screen. 14. The results discussed in this section hold, in most respects, for two sub-periods (1975-81 and 1982-87) also. Further, in a year-wise analysis, earnings of the high P/E-high ROE group exhibit higher explanatory power for returns compared to earnings of the high P/E-low ROE group in eleven out of thirteen years. (Year-wise and sub-period results are not reported.)15. Since the objective of this analysis is to obtain the “best” versus the “worst” possible performances, stringent partitioning schemes are applied, which result in small screen sizes. When firms are partitioned into three earnings variance groups (instead of four), while the sample sizes increase to 235 for the HG screen and 421 for the LG screen, the R2 reduces from 31 to 28 percent for the HG screen, but remains unchanged for the LG screen. As expected, three earnings variance groups do not provide as strong a discrimination as four groups and thus result in a slight reduction in the R2 difference between the HG and LG screens. 16. The performance of extreme (prior) earnings variance groups, EV1 and EV4, was also analyzed. Although a distinction can be observed with respect to the variance effect (0.22 for EV1 versus 0.36 for EV4), the covariance effect shows a difference between the two extremes in the reverse direction (0.05 for EV1 and -0.09 for EV4) thus obtaining equal R2s of 7 percent for both groups. 17. Fairfield and Harris (1990) investigate the mispricing and risk explanations for the returns to P/E and P/B strategies. 18. Bernard (1993) provides weak support for the relation between B/P ratios and the degree of accounting conservatism from an analysis of a limited number of dimensions of conservatism. ReferencesBeaver, W., and D. Morse. What Determines Price Earnings Ratios. Financial Analysts Journal (July/August 1978), 65-76. Bernard, V. Capital Market Research in Accounting during the 1980s: A Critical Review. Paper for the University of Illinois Accountancy Ph.D. Program Golden Jubilee Symposium, 1989. Bernard, V. Accounting-Based Valuation Methods, Determinants of Market-to-Book Ratios, and Implications for Financial Statements Analysis. Working paper. University of Michigan, 1993. Easton, P., and T. Harris. Earnings as an Explanatory Variable for Returns. Journal of Accounting Research (Spring 1991), 19-36. Easton, P., T. Harris, and J. Ohlson. Aggregate Accounting Earnings can Explain Most of Security Returns: The Case of Long Return Intervals. Journal of Accounting and Economics (June/September 1992), 119-142. Fairfield, P., and T. Harris. An Investigation of Mispricing and Risk as Explanations of the Returns to Price-to-Earnings and Price-to-Book Value Trading Strategies. Working paper. Columbia University, 1990. Lev, B. On the Usefulness of Earnings, and Earnings Research: Lessons and Directions from Two Decades of Empirical Research. Journal of Accounting Research (Supplement 1989), 153-192. Ohlson, J. Earnings, Book Values, and Dividends in Equity Valuation. Contemporary Accounting Research (forthcoming). Ohlson, J., and P. Shroff. Changes vs. Levels in Earnings as Explanatory Variables for Returns: Some Theoretical Considerations. Journal of Accounting Research (Autumn 1992), 210-226. Penman, S. An Evaluation of Accounting Rate-of-Return. Journal of Accounting, Auditing and Finance (Spring 1991), 233-255. Penman, S. Return to Fundamentals. Journal of Accounting, Auditing and Finance (Spring 1992a), 465-483. Penman, S. Financial Statement Information and the Pricing of Earnings Changes. The Accounting Review (July 1992b), 563-577. Shroff, P.K. The Relation between Accounting Earnings and Security Returns over Long Intervals. Working paper. The Ohio State University, 1995. TABLE 1Panel A: Predicted characteristics of firms in extreme P/E and ROE combinationsP/E RatioHighLowROEHighLow B/PHigh Lt (Growth)E(Xt)/Xt-1 > 1E(g) > 0 Cov()xt,)xt-1) > 0Medium B/PLow Lt (Growth)E(Xt)/Xt-1 LowMedium B/PLow Lt (Growth)E(Xt)/Xt-1 > 1E(g) > 0 Cov()xt,)xt-1) 0Panel B: Descriptive statistics of P/E-ROE combinations and negative eps firms over 1975-87___________________________________________________________________________________________ HL HH LL LH HL HH LL LH Negative eps Means Medians Means Medians ____________________________ ___________________________ _______________E/P Ratio 0.059 0.074 0.175 0.198 0.057 0.072 0.168 0.172 -0.211 -0.086ROE 0.064 0.208 0.095 0.205 0.064 0.197 0.099 0.182 -0.341 -0.084B/P Ratio 1.036 0.371 1.934 0.989 0.945 0.355 1.721 0.928 1.275 1.024Xt/Xt-1 1.163 1.146 0.709 0.841 1.308 1.152 1.033 0.939 -0.391 -0.096g 0.107 0.091 0.219 0.110 0.029 0.040 0.157 0.045 0.241 0.121 Panel C: Results of pooled regressions and serial correlation in earnings changes of P/E-ROE combinations and negative eps firms)Xt/Pt-1 = a0 + a1)Xt-1/Pt-2 + ut’___________________________________________________________________ HL HH LL LH Negative eps _____________________________________ ___________N 1063 885 492 855 745a0 0.003 0.010 -0.058 -0.024 0.100t(a0) (0.82) (11.49)** (-5.29)** (-3.15)** (5.79)**a1 -0.050 0.013 0.166 -0.222 -0.360t(a1) (-1.56) (1.55) (2.48)* (-7.96)** (-10.59)**Corr()Xt,)Xt-1): – Pearson -.009 .118 .028 -.335 -.362 – Spearman -.089 .327 .078 -.236 -.363___________________________________________________________________Firms are ranked by P/E ratios and ROE and formed into three groups (on each criterion) at the beginning of each year. HL (HH) consists of firms in the highest P/E and the lowest (highest) ROE groups. LL (LH) consists of firms in the lowest P/E and the lowest (highest) ROE groups. E/P ratio is measured by dividing previous year’s eps (Xt-1) by the beginning price (Pt-1). ROE is measured by dividing previous year’s eps by the beginning bvps (BVt-1). B/P ratio is measured by dividing the beginning bvps by the beginning price. g = )gt/Pt-1 = [(Pt-BVt) - (Pt-1-BVt-1)]/Pt-1 = Change in goodwill from (t-1,t) normalized by the beginning price. * Significant at the .05 level. ** Significant at the .01 level. TABLE 2Results of pooled returns-earnings regressions for P/E and ROE groups and P/E-ROE combinations estimated over the period 1975-87_________________________________________________________________yt = ” + $xt + ,t (1b)Panel A: High, medium, low, and negative P/E groups N R2 t() t() _________________________________________________________________High P/E 2361 .05 0.107 0.827 (11.04)** (10.62)**Medium P/E 2369 .06 0.089 0.916 (8.27)** (12.23)**Low P/E 2366 .05 0.177 0.582 (16.56)** (11.51)**Negative P/E 745 .02 0.199 0.189 (10.59)** (3.94)**Panel B: High, medium, and low ROE groups N R2 t() t()_________________________________________________________________High ROE 2361 .08 0.081 1.114 (6.61)** (14.01)**Medium ROE 2369 .09 0.057 1.230 (4.70)** (15.18)**Low ROE 2366 .04 0.155 0.491 (17.59)** (10.41)**Panel C: P/E-ROE combinations N R2 t() t() _________________________________________________________________High P/E-Low ROE 1063 .04 0.119 0.681 (8.55)** (6.79)**High P/E-High ROE 885 .20 -0.240 4.969 (-7.89)** (14.76)**Low P/E-Low ROE 492 .04 0.276 0.330 (13.72)** (4.37)**Low P/E-High ROE 855 .07 0.126 0.819 (5.92)** (8.13)**_________________________________________________________________P/E and ROE groups are formed at the beginning of each year. Variable definitions:xt = Eps for the period t-1 to t (Xt) normalized by beginning price (Pt-1). dt = Dividends paid at date t. yt = (Pt + dt – Pt-1)/ Pt-1 = Security returns over (t-1,t) including dividends. Other variables are defined in Table 1. TABLE 3Panel A: Descriptive statistics of the overall sample and LG and HG screens over 1975-87_________________________________________________________________ Means Medians overall LG HG overall LG HG _______ ______ ______ _______ ______ ______ E/P Ratio 0.120 0.118 0.075 0.112 0.117 0.072 ROE 0.139 0.098 0.230 0.134 0.101 0.218 B/P Ratio 0.973 1.223 0.345 0.897 1.176 0.326 Xt/Xt-1 0.980 1.071 1.169 1.077 1.098 1.189 g 0.107 0.051 0.123 0.060 0.016 0.049Panel B: Results of pooled regressions and serial correlation in earnings changes of the overall sample and LG and HG screens)Xt/Pt-1 = a0 + a1)Xt-1/Pt-2 + ut’__________________________________________________________________ Overall LG HG _______ ______ ______ N 7096 296 177 a0 -0.009 0.006 0.011 t(a0) (-5.99)** (1.65) (3.35)** a1 -0.061 -0.066 0.014 t(a1) (-7.11)** (-0.40) (0.68) Corr()Xt,)Xt-1): – Pearson -.119 -.023 .051 – Spearman -.044 -.075 .217 ________________________________________________________________The low growth (LG) screen consists of firms with low to medium beginning E/P, low beginning ROE, and low prior earnings variance (EV1). The high growth (HG) screen consists of firms with low beginning E/P, high beginning ROE, and high prior earnings variance (EV4). Other variables are defined in Tables 1 and 2. * Significant at the .05 level. ** Significant at the .01 level. TABLE 4Results of pooled returns-earnings regressions and correlation component analysis, overall and for LG and HG screens estimated over the period 1975-87_____________________________________________________________________________________________Panel A: yt = ” + $xt + ,t (1b)_____________________________________________________________________________________________ N R2 Cov[xt,g] Corr[xt,g] Cov Var t() t() Effect Effect_____________________________________________________________________________________________overall 7096 .06 0.126 0.737 -0.0036 -.08 -0.08 0.33 (21.56)** (20.75)**LG 296 .01 0.104 0.415 -0.0022 -.14 -0.14 0.25 (3.26)** (1.82)HG 177 .31 -0.312 6.169 0.0121 .47 0.44 0.09 (-4.54)** (8.99)**Rest 6623 .06 0.129 0.722 -0.0042 -.09 -0.09 0.33 (21.72)** (20.22)** Cov Effect = Cov [xt,g]/[SD(xt)SD(yt)] and Var Effect = SD(xt)/SD(yt). Panel B: Changes and levels model (Easton and Harris [1991])yt = 20 + 21)xt + 22xt + et__________________________________________________________________________________________ R2 20 21 22 t(20) t(21) t(22)__________________________________________________________________________________________Overall .09 0.024 -0.931 1.575 (2.75)** (-15.05)** (23.95)**LG .02 0.012 -1.015 1.205 (0.21) (-2.00)* (2.64)**HG .31 -0.428 -2.036 7.775 (-3.70)** (-1.24) (5.32)**Rest .09 0.026 -0.936 1.568 (2.84)** (-15.05)** (23.65)**__________________________________________________________________________________________LG and HG represent the high and the low growth screens respectively. “Rest” includes all firms in the sample excluding the firms in the HG and LG screens. Other variables are defined in Tables 1 and 2. * Significant at the .05 level. ** Significant at the .01 level.

Додати в блог або на сайт

Цей текст може містити помилки.

A Free essays | Essay
46кб. | download | скачати


Related works:
Correlation Between Tempest And The New World
The Correlation Between Contemporary Transit And T
Correlation Between Masculinity And Empathy
The Correlation Between Chinese History And Beliefs
Behind Closed Doors The Correlation Between Multiple
Louie Freeh (Director Of Fbi) And Correlation
Correlation Between Proper Nutrition That Enhances Performance
© Усі права захищені
написати до нас