Chapter 5. U.K data and variables
55 See Appendix 5.3, Table A5.3.4 for examples o f the relationship between X4 and the published profit figure.
5.4. Descriptive Statistics
Table 5.6, Panel A shows the descriptive statistics of the main raw variables in per-share form - 4 earnings measures, stock price and book value. The distribution of all these variables is long-tailed, which means that accounting variables and stock price tend to have high density around their median. Conversely, there are potentially influential outliers in the data set. Therefore, the trimming or winsorising criteria should be applied in the data analysis in order to avoid the effect of extreme outliers. In this U.K. study, I delete the 1% most extreme outliers for the estimation of RI and 01 persistence parameters. However, I retain all data for the test of reliability of value estimates.
On the other hand, mean values of each variable are much higher than median values, which means that the distribution of all variables tends to be right-skewed.69 Right skewness of stock price is because some firms have extremely large stock prices that dominate the stock market. Meanwhile, right-skewness of earnings variables and book value is because the accounting system is conservative. That is, the accounting system tends to postpone the recognition of revenues/gains and accelerate the recognition of expenses/losses (Myers, 1999a). Together with Table 5.6, Panel B, we can see that firms' stock price, earnings and book value have increased over time.
Table 5.6, Panel A also presents the descriptive statistics of RI. The negative median
68 90% o f each variable is centrally dense within 0.2 - 0.5% o f the corresponding range.
69 Even though mean value o f X3 is also much higher than median value o f it, its skewness is negative. The negative skewness o f X3 seems to be caused by large negative extraordinary items.
Chapter 5. U.K. data and variables
(mean) residual income regardless of earnings measures is consistent with previous research and is because ROE is smaller than the discount rate, on average.70 There are two possible reasons why the median (mean) values of RI in Panel B are higher than those in Panel A. One is that the discount rate has fallen recently so that many ROE values exceed the discount rate. Based on 8,346 observations from 1989 to 1998, the median (mean) discount rate is just 13.3% (13.6%), while the median (mean) ROE is about 14% (16-17%). The other reason is that firms' earnings seem to have increased more than firms' book value, on average, over time. Compared to median values for the periods 1969-1998, median values of earnings for the periods 1989-1998 have increased more than 60%, while median values of book value have increased about 40%.
Table 5.6, Panel A also shows the relationship between alternative earnings measures. The mean of XI is larger than the mean of X3 or X4, because the exceptional {EXC) and extraordinary {EXT) items, especially large items, tend to occur as losses rather than profits, in general, so that their mean values are negative, on average. We can see that AEX (all abnormal items), EXC and EXT tend to have large negative numbers. As described in Appendix 5.6, XI is approximately X3 minus A E X(0.102 + 0.038), or X4 minus EXC (0.132 + 0.008). In addition, XI is X2 plus SA (full tax adjustments after SSAP 15) (0.126 + 0.014), and X3 is X4 plus EXT (0.132 - 0.030).
On the other hand, Table 5.6, Panel B shows statistics of analysts' earnings forecasts and 'other information'. First, median (mean) value of one-year ahead analysts' earnings
70 Long-run median (mean) ROEs are less than 13.4% (14.8%) regardless of earnings measures. Thus, these ROEs o f U.K. industrial companies are less than the cost o f capital (15-15.5%), on average.
forecasts are positive and greater than the median (mean) value of realized earnings, which means that analysts tend to forecast earnings optimistically. The optimistic behavior of analysts is consistent with evidence reported in prior studies (e.g., O'Brien, 1988; De Bondt and Thaler, 1990; Brown, 1997; Brown, 1998; Richardson et al., 1999; Easterwood and Nutt, 1999).
Second, median (mean) value of analysts' earnings forecasts is increasing in forecasting windows. That is, two-year ahead analysts' earnings forecasts are larger than one-year ahead analysts' earnings forecasts, and three-year ahead analysts' earnings forecasts are larger than two-year ahead analysts' earnings forecasts. This indicates that analysts' tend
71 to be more optimistic when they forecast eammgs over a longer time horizon. The tendency of the incremental optimism over forecasting windows is also consistent with previous research, and it may, at least partly, cause the superiority (in terms of bias and accuracy) of longer horizon EBO models reported in Sougiannis and Yaekura (2000).
Third, 'other information' is positive, on average,72’73 which means that analysts-based forecasts of RI are higher than RI forecasts based on the univariate AR(1) RI generating equation. The mean of analyst-based RI forecasts for 1989-1998 is -3.3% (unreported), while the corresponding figure for the realized RIs for 1969-1998 is -11.8% to -15.9%.
71 In order to make median (mean) values of 1 to 3-year ahead earnings forecasts comparable, the number o f observations are reduced to 3,711, but the evidence o f the incremental optimism over forecasting windows does not change. In this case, the median (mean) values o f 1 to 2-year ahead earnings forecasts are respectively 0.151 (0.196) and 0.175 (0.226).
72 The negative mean value of OI based on XI is caused by one extremely large negative OI (-132). The numbers in parentheses show mean and standard deviation of OI when one extremely large negative OI is deleted.
73 The intercept o f AR(1) RI regression is incorporated into the calculation of OI. If the intercept parameter is ignored in the calculation o f OI, and one extremely large negative OI is deleted, the median (mean) value o f OI is still positive, but smaller than the corresponding figures in Table 5.6, Panel B.
Chapter 5. U.K. data and variables
Note that the sample for the periods 1969-1998 is used for the univariate AR(1) RI generating process.
Table 5.7 describes the properties of some main variables using ratios. Here, the earnings measure X4, which is conceptually similar to the earnings measure employed in Dechow et al.'s (1999) U.S. study, is used. Overall, magnitudes and signs of figures from the U.K. sample are very consistent with corresponding U.S. figures reported in Chapter 4, even though the sample period is different. Panel A is based on the sample from 1969 to 1998 and Panel B from 1989 to 1998. From these two panels, i) book-to- market ratio and eamings-to-price ratio have decreased over time, and ii) firms' profitability (eamings-to-lagged book ratio, i.e., ROE) and abnormal returns (Rl-to- lagged book ratio, i.e., the spread between ROE and cost of capital) have increased over time. Also, as U.S. study in Chapter 4, 1) the median and mean value o f scaled OI (2.9% and 15.6%) are positive, and 2) the median and mean of scaled analysts-based RI forecasts for 1989-1998 (2.2% and 11.1%) are higher than the corresponding figures of the realized RIs (-2.2% and -1%) for 1969-1998.
Table 5.1: Examples o f inconsistency in Datastream fs post-SSAP 15 adjustment
(£ ,000) Example 1 : SCAPA Group
91 92 93 94 95 96 97 98
Effect on P/L account tax charge o f SSAP 15 (per note to Financial Statement)
384 -50 527 400 2400 400 2000 -500
SSAP 15 adjustment o f Datastream 384 -50 0 0 0 0 2000 -500
Example 2 : Cadbury Schweppes
90 91 92 93 94 95 96 97
Effect on P/L account tax charge o f SSAP 15 (per note to Financial Statement)
6200 2200 1700 7900 3200 7000 -3000 -4000
SSAP 15 adjustment o f Datastream 6200 2200 1700 0 0 7000 0 0
Note : Relevant notes in Financial statement of SCAPA Group (1998) and Cadbury Schweppes (1992) are respectively "Had full provision for deferred taxation been made for the whole Group then there would have been an additional credit o f £0.5m (£2.0m charge in the case o f 1997)" and "The charge o f £94.2m has been reduced by £1.7m in respect o f tax at the current year's rate on timing differences for which deferred tax has not been provided".
Table 5.2: Four earnings definitions and Datastream items74
Pre-FRS3 Post-FRS3 General
XI DS 210 DS 210 DS 210 X2 DS 182 DS 182 DS 182 X3 DS 625 + DS 193 DS 6251} DS 625 + D S 193 X4 DS 6252) DS 625 - (DS 1083 - DS 1094 - D S 1097) DS 6 2 5 - ( D S 1083 - D S 1094 - D S 1097) Note:
1) DS 193 still exists after FRS 3, but its numbers are all zero.