### Table 3: Comparison of pre- and post-event ISD for good news

### Table 3: 60 Day Post Event CAR organized by event period CAR and size classifications.

"... In PAGE 7: ... However, we do not replicate the Foster Olson and Shevlin (1984) result. We find that a drift effect is still present using a standardized CAR proxy for unexpected earnings ( Table3 ). It is weaker than the drift based on SUE, but still statistically significant.... In PAGE 17: ... Although we are aware of no model that predicts these phenomena, the extreme decile behavior is consistent with the potential survival effects discussed earlier. In Table3 we analyze the post-announcement abnormal return by independent size and 3-day announcement period standardized abnormal return classifications. The standardized abnormal return over days -2 to 0 is used as an alternative proxy for unexpected earnings.... ..."

### Table 6: 60 Day Post Event CAR organized by independent SUE and Ohlson probability of bankruptcy classifications: Measures of skewness, kurtosis and correlation with post-event period standard deviation

"... In PAGE 21: ...2.2 Bankruptcy risk model sub-sample results In Table6 we provide evidence that the SUE effect is associated with return distribution differences and risk differences across portfolios. We classify firms by independent SUE and bankruptcy probability measures, basing the bankruptcy probability measure on the model in Ohlson (1980).... In PAGE 21: ...56% for the relatively low risk firms. Although the results in Table6 are consistent with the survival explanation, their main importance is not in establishing a link between drift and bankruptcy probability but in demonstrating clear differences in the sample distributions of abnormal returns across SUE portfolios, and in showing how distributional differences are associated with bankruptcy risk. The skewness of portfolio abnormal return is particularly noteworthy.... In PAGE 21: ... In all cases the skewness of abnormal returns is positive. The results in Table6... In PAGE 22: ... However, we note that the correlation is also consistent with positive skewness in the distribution of returns. In Table 7 we conduct a similar analysis to that in Table6 , but this time with respect to abnormal returns in the next subsequent three-day earnings announcement period, approximately three months after the announcement of SUE . Previous research has found strong positive t correlation between the abnormal return over this subsequent three-day period and SUE, measured at time t.... ..."

### Table 7: CAR on next quarterly announcement classified by independent SUE and Ohlson probability of bankruptcy classifications: Measures of skewness, kurtosis and correlation with post-event period standard deviation

"... In PAGE 22: ... However, we note that the correlation is also consistent with positive skewness in the distribution of returns. In Table7 we conduct a similar analysis to that in Table 6, but this time with respect to abnormal returns in the next subsequent three-day earnings announcement period, approximately three months after the announcement of SUE . Previous research has found strong positive t correlation between the abnormal return over this subsequent three-day period and SUE, measured at time t.... In PAGE 22: ... Previous research has found strong positive t correlation between the abnormal return over this subsequent three-day period and SUE, measured at time t. Table7 indicates that in our sample the difference in the three-day abnormal return between low-SUE and high-SUE portfolios is 1.05%.... In PAGE 23: ...isk firms. This portfolio separately displays abnormal returns of 1.07% over the three-day period. Table7 also reveals dramatic differences in returns skewness across different partitions. Although the statistics are not directly comparable across Tables 6 and 7 because of the length of the different event periods (sixty days versus three days), the two tables contain qualitatively similar results.... ..."

### Table 1: Comparison of pre- and post-event ISD

### Table 6: Comparison of pre- and post-event ISD for good news

### Table VII.a Mean Calendar-Time Portfolio Abnormal Returns (CTARs) for Acquirers (7/61 - 12/93) CTARs are calculated each month as the difference between the event portfolio return and the expected return on the portfolio, standardized by the portfolio residual standard deviation. Each month we form equal and value-weight event portfolios containing all sample firms that have completed the event within the previous three years. The event portfolio is rebalanced monthly to drop all companies that reach the end of their three-year period and add all companies that have just executed a transaction. The portfolio expected returns are proxied by both 25 value-weight portfolios formed on size and book-to-market equity based on NYSE breakpoints (25 Size-BE/ME), and the Fama and French three-factor model (FF 3-Factor), which amounts to estimating individual firm factor loadings over a five-year post-event estimation period (requiring at least 36 months of valid returns), and then averaging these to form the monthly portfolio factor loadings. We calculate event portfolio residual variances using 60 months of residuals. Residuals are calculated from portfolio regressions on the FF three-factor model and as monthly differences of event portfolio returns and size-BE/ME portfolio returns. Mean CTARs and standard errors are calculated from the time-series of monthly CTARs. The number of monthly observations are reported in square brackets, and t-statistics are in parenthesis.

2000

Cited by 17

### Table II. Specifically, we extended the post-event window for cumulative excess returns through

2000

Cited by 5

### Table 3. Performance of Neural Network Recognizers for Various Domain Groups

"... In PAGE 2: ... 3) and used pre-selected simi- larity regions (strategy b) for the training process. The perfor- mance of 40 such networks is shown in Table3 A. The number of false positives and false negatives is extremely low in most cases.... In PAGE 2: ... This was neces- sary as a preliminary analysis (not shown) re- vealed that many of the COG groups are too small for training ANNs, and in addition, BLAST found too few significantly similar neighbors for many of the larger groups. From the resulting data set we chose a number of groups, listed in Table3 B, and identified the members on the ba- sis of their annotations. Then, the PpNSDH11642,PpAVSH11642, Pnfp NSDH11642,andPnfp AVSH11642 functions were determined from a database versus database comparison, and the neural networks trained for the selected groups as described in Methods.... In PAGE 2: ... Then, the PpNSDH11642,PpAVSH11642, Pnfp NSDH11642,andPnfp AVSH11642 functions were determined from a database versus database comparison, and the neural networks trained for the selected groups as described in Methods. The predictive performance of the ANNs trained for these func- tional groups is only slightly inferior to those ob- tained with the domain groups ( Table3 B). In some cases, there is a conspicuously high number of false predictions, for example, of 428 permease sequences, 29 false positives (6.... In PAGE 3: ...listed in Table3 B, 21 and 13 of them, respectively, coincide with conflicting annotations between various sequence data- bases such as SWISS-PROT, PIR, and COG. However, it has to be pointed out that this comparison was carried out on an experimental data set, therefore, the results cannot be used to draw conclusions on the quality of the underlying databases.... In PAGE 7: ...1 complete genomes (Tatusov et al. 2000). First, a nonredun- dant data set was created from the COG database and from those sequences of the SWISS-PROT and PIR databases that contained functional annotations in their feature tables. A few sufficiently large ( gt;200 members) groups were chosen at random as examples for ANN analysis ( Table3 B) and their members identified on the basis of their annotations as given in the various databases. The sequence annotations were also checked by visual inspection.... In PAGE 7: ...7% of the positive and 94.9% of the negative decisions listed in Table3 Awere reached by a 5 : 0 vote. ACKNOWLEDGMENTS This work was supported in part by EMBnet, the European Molecular Biology Network, in the framework of EU grant no.... ..."