A Re-Examination of Residuals around the Ex-Dividend Date
Berry, Michael A., Gallinger, George W., Akron Business and Economic Review
A Re-Examination Of Residuals Around The Ex-Dividend Date
Much controversy exists about whether share prices are affected by a dividend-related tax effect on the day the stock goes ex-dividend. Campbell and Beranek  and Elton and Gruber  were early proponents of the tax effect phenomenon. Several researchers [e.g., 1, 2, 5, 16, 17, 18] addressed the issue of whether the market requires higher returns, and hence lower current prices, on high dividend yielding stocks to compensate for the tax disadvantage of dividend income. (1,2) Unfortunately, resolution of the issue is elusive.
Much of the debate in recent years is traced to studies by Litzenberger and Ramaswamy  and Miller and Scholes , who focus on the interpretation of an observed positive correlation between dividend yields and stock returns. Litzenberger and Ramaswamy contend that the correlation signifies a tax effect, whereas Miller and Scholes claim that information effects are responsible for the correlation. Eades, Hess, and Kim  and Lakonishok and Vermaelen  find positive excess returns on cash dividend ex-dates, which supports a tax-effect hypothesis. However, they also find excess returns on ex-dates for non-taxable stock dividends and stock splits, a phenomenon that they are unable to explain.
The purpose of this study is to re-examine excess returns around the ex-dividend date for cash dividends, with particular emphasis on the ex-day. We use daily returns and allow beta to be a random coefficient. We find no excess returns on ex-dividend days, contrary to the studies by Litzenberger and Ramaswamy , Eades, Hess, and Kim , and Lakonishok and Vermaelen . Litzenberger and Ramaswamy used monthly returns and, thus, were forced to interpolate ex-dividend event dates. Eades, Hess, and Kim, using daily data, assumed beta to be stationary in the single-index market model, which is contrary to much evidence [e.g., 8, 11]. Lakonishok and Vermaelen compute excess returns by subtracting the equally weighted market return for day t from the equally weighted portfolio of all stocks that go ex-dividend on day t. They assume the "ex-date" portfolio beta to equal one. In a footnote (number 5), they admit that such an assumption may induce a bias into the excess returns, although they feel it would be insignificant. (3)
DATA AND METHODOLOGY
The data base was created by randomly selecting 225 dividend paying firms that trade on the New York Stock Exchange. Reference was then made to Standard & Poor's Dividend Record to randomly select one dividend announcement date for each firm for each of the sample years 1971, 1976, and 1981. These years were chosen to permit representation of a broad cross-section of market conditions. The daily CRSP data base was then accessed to provide daily returns data around each of the selected ex-dividend dates. Securities with missing data were eliminated from each sample. This resulted in samples consisting of 165 securities in 1971, 167 securities in 1976, and 210 securities in 1981. Four portfolios were constructed for analysis: one portfolio for each of the years and a portfolio that pooled the three years. Analysis of individual years was to permit the isolation of a potential year effect if it existed.
Data analyses were conducted as follows. Ninety observations were selected around the posting date for each security. This consisted of ten days of data after the posting data and 80 days prior to the posting date. This ensured that a sufficient number of observations were available around the ex-date for inclusion in the single-index market model (SIMM). Since the ex-dates varied cross-sectionally in each sample period, cross correlation of residuals was minimized. As will be apparent later in this paper, any confounding of dividend announcement information with ex-dividend date information was minimized.
Next, an average residual methodology was implemented using Engle and Watson's  state-space varying parameter regression (VPR) model. …