Data and sample Directory UMM :Data Elmu:jurnal:J-a:Journal of Empirical Finance (New):Vol7.Issue1-2.2000:

Ž . al. 1992 , issue announcement returns are negatively related to the number of days between the information releases and subsequent issue announcements.

4. Data and sample

4.1. Equity issue data We extracted 3932 US seasoned equity offerings from the Securities Data Corporation database, for the period 1980–1994. CRSP data are available for 2651 of the issuing firms at the time of the offers. Using Lexis rNexis, we identified announcement dates for 1648 of these offerings. 5 From this sample, 117 observa- Ž . tions were deleted for one or more of the following four reasons. 1 On 64 of the Ž . announcement dates, the firm also announced its intention to issue debt. 2 Thirty-two of the announcements were for shelf offerings, in spite of our efforts to Ž . eliminate shelf offerings when the data were extracted from the SDC file. 3 The wording of 19 of the announcements implied that the issuing firms had previously announced their intentions to issue equity, even though we could not find the Ž . earlier announcements. 4 For 14 of the announcements, it was unclear whether the announcement pertained to the same class of common stock as the CRSP data. After deleting these observations, we were left with a sample of 1531 announce- ment dates. Enough CRSP data were available to compute 3-day market adjusted returns for 1509 of these observations. Previous studies of issue announcement effects have used announcement dates Ž . from the Wall Street Journal WSJ Index. The disadvantage of this source is that it reports few issue announcements after 1983. To test the suitability of our Lexis rNexis data, we obtained a sample of 389 WSJ announcements from Assem Safieddine. Safieddine’s sample consists of equity issues reported by Investment Dealers’ Digest from 1980 to 1983, and includes 99 issues from our Lexis rNexis sample. Out of the 99 issues for which we have both WSJ and Lexis rNexis announcement dates, there are only three issues for which the two sources agree. There are 67 issues, however, for which the WSJ date is only one trading day after the Lexis rNexis date. This is sensible, because announcements reported on the news wires are not published in the WSJ until the following day. There are additional five observations for which the WSJ reported the announcement 2 trading days after Lexis rNexis, and 16 observations for which the WSJ announce- ment followed Lexis rNexis by more than 2 days. On the other hand, the WSJ announcement preceded the Lexis rNexis announcement for eight issues, each 5 Some announcements were deleted at this point because other major announcements were made on the same days. time by at least 3 days. Summarizing these results, Lexis rNexis provided earlier announcements for 88 of the 99 issues, while the WSJ index had earlier announce- ments for only eight of the issues. Because we are interested in capturing the earliest announcement, the Lexis rNexis announcement dates are superior to the dates from the WSJ index. 4.2. Stock split data For our sample of stock splits, we extracted all 7650 stock splits from the CRSP NYSE, AMEX, and NASDAQ files from 1979 to 1994. We begin in 1979 because we are interested in the effects of split declarations that occur before issue announcements, and our equity issue data begins in 1980. From this initial sample, we deleted three duplicates, 132 reverse splits, and 28 splits with split factors less than 20 6 , giving us a final sample of 7487 splits. 4.3. Earnings releases and diÕidend announcements Ž . Korajczyk et al. 1991 present evidence that earnings releases reduce asym- metric information before equity issues. To examine the effects of pre-issue earnings releases on the issue announcement abnormal returns, we obtain firms’ earnings release dates from Compustat quarterly files and the WSJ Index. Earnings release dates are available for 1413 observations out of the 1531 issue announce- ments in our sample. Ž . Loderer and Mauer 1992 investigate whether dividend announcements affect subsequent issue announcement returns. To control for the effects of dividend announcements, we collect the dividend declaration dates from the CRSP NYSE, AMEX and NASDAQ files. We divide firms from our issue announcement sample Ž . Ž into firms with dividend increases 194 observations and decreases 25 observa- . Ž . tions before the issue announcements, no dividend change 539 observations and Ž . no cash dividend paid 773 observations during the 250 trading days preceding the issue announcement. 4.4. Abnormal returns around eÕent dates Using CRSP data, we compute the market-adjusted abnormal returns over days y1 to q1 around the split declaration and issue announcement dates, where day 0 Ž . is the event day. Including the day before the event day y1 captures the market’s response when the news is released a day before it is reported by the Ž . press. We include the day after the event day q1 because many firms in our 6 A two-for-one split has a split factor 100. sample are small and may experience light trading on the event date; consequently, some of the price adjustment may not be observed until the following day. For firm i on day t, the market-adjusted abnormal returns are computed by subtracting the AmarketB returns R from the returns of the individual stock R , m t i t i.e., AR s R y R . 12 Ž . i t i t m t For NYSE rAMEX stocks, we use the return to the CRSP NYSErAMEX equal-weighted portfolio as our proxy for the market return. For NASDAQ stocks, our proxy for the market return is the return to the CRSP NASDAQ equal-weighted portfolio. We deleted the December 8, 1992 issue announcement by TSS. This observation had an abnormal return of 259, and had substantial influence on some of the test statistics. We use market-adjusted returns rather than market model prediction errors in most of our tests. This is because price run-ups in the periods before splits and equity issues can bias estimates of the market model parameters. According to Ž . Brown and Warner 1985 , event studies with market-adjusted returns have comparable power to procedures that use the market model. To insure that our conclusions about asymmetric information are not driven by our measure of abnormal returns, we confirm our major findings using market model prediction errors.

5. Conditional event-study analysis of issue anticipation and issue announce- ment returns