SHORT TERM TRADING AROUND DIVIDEND DISTRIBUTIONS: AN EMPIRICAL APPLICATION TO THE LISBON STOCK MARKET

Transcription

1 SHORT TERM TRADING AROUND DIVIDEND DISTRIBUTIONS: AN EMPIRICAL APPLICATION TO THE LISBON STOCK MARKET Maria Rosa Borges Instituto Superior de Economia e Gestão UTL Rua Miguel Lupi, LISBOA Tel: mrborges@iseg.utl.pt Abstract Dividend distributions provide us with an adequate framework for detecting and analysing the activity of short term traders. If the price adjustment after the dividend distribution is not enough to compensate for the distributed dividend, this means that there is an opportunity to make additional gains through arbitrage, that will be explored by rational short term traders. To put it differently, the existence of a strong arbitrage activity will force the stock price to reach a new equilibrium position. If this short term trading hypothesis is true, then we should observe the following: (i) an abnormal positive transactions volume in the dividend period; (ii) a negative relation between the intensity of trading activity and transactions costs; (iii) a positive relation between the intensity of trading activity and the liquidity of stocks. Several studies, including Lakonishok e Vermaelen (986), tried to find evidence of this short term trading activity, looking at the transactions volume behavior in the dividend period. The purpose of the present study is, by applying event time and calendar time methods, to test the short term trading hypothesis for the portuguese stock market, during 990 to 999. Keywords: Dividend distributions; Short term trading; Transactions volume; Transaction costs. Jel Code:; G2; G4

2 SHORT TERM TRADING AROUND DIVIDEND DISTRIBUTIONS: AN EMPIRICAL APPLICATION TO THE LISBON STOCK MARKET. INTRODUCTION The distribution of dividends provides a good opportunity to study the activity of short term traders. When a firm pays dividends, the stock price should fall by an amount that makes investors indifferent between transacting before or after the distribution. In this context, the role of short term traders is fundamental, because their actions in the market will take the stock price to a new equilibrium. In different ways, several authors have addressed the role of short term traders, in the context of dividend distributions. Many of these works considered the existence of a tax effect linked to the dividend distribution. The underlying idea to these studies is that because dividends are taxed higher than capital gains, investors would demand a premium to receive their returns in the form of dividends, and that premium should be reflected in the stock price after the dividend distribution. In short, the dividend distribution would lead investors to transact on the stock, forcing the stock price down, but by an amount less than the dividend amount, because of the unfavorable tax difference on dividends. Selected works are those of Elton and Gruber (970), Kalay (982), Miller and Scholes (982), Elton, Gruber and Reutzler (984), Eades, Hess and Kim (984), Lakonishok and Vermaelen (983) and Poterba and Summers (984). It is also important to point out the influence of transaction costs, as a restriction to short term trading and to stock price adjustment. These aspects have been stated by Kalay (982) and Elton, Gruber and Reutzler (984). Until now, all works refered focus on the dividend day, in the assumption that the stock price adjustment should happen entirely on the first market session where the 2

3 stock is transacted without the right to the dividend. Another approach has focused instead in the fact that the stock price adjustment might not be completed on the dividend day, due to market imperfections, and studied the price behavior in a set of days around the dividend day, defined as the dividend period. From these, we point out Eades, Hess and Kim (984), Lakonishok and Vermaelen (986) and Michaely (99). These studies allow a different approach to the tax issue, but also focus on the activity of short term traders during the dividend period. Eades, Hess and Kim (984) found empirical evidence of the existence of excessive returns in a period around the dividend day, which undermines the fiscal effect interpretation, and concluded that short term trading affects prices in a set of days near the dividend day. Another study that detected short term trading activity related to dividend distributions, was that of Lakonishok and Vermaelen (986). These authors have analised the transactions volumes in the dividend period, and shown that it increased significantly during that period. They also found that the reduction of transaction costs, in the years covered by the study, had led the abnormal transaction volume to increase even more. After 986, the interest of investigators on these issues reduced strongly, as a consequence of the American tax reform of 986 and of the general tendency in western taxation regimes, to apply similar taxes on dividends and capital gains, thus making the fiscal effect on the adjustment of stock prices disappear. In other markets, like the Portuguese market, the persistence of different taxes on dividends and capital gains, makes this issues still very relevant. The present empirical study is framed by the work of Lakonishok and Vermaelen (986), which we apply to the Portuguese market, using data from 990 to 999. More specifically, through the analysis of the transactions volume behavior, we will test the existence of a positive abnormal volume during the dividend period, which 3

4 will constitute evidence of the presence of short term traders in the market, alert to the profit opportunities that come linked to the dividend distribution event. The paper is organized as follows. In this part, the framing in which this empirical study is included has been presented; in part 2, the testable hypothesis are stated; parts 3 and 4 describe the event time and calendar time methods, and the way in which they will be applied, to test the hypothesis; in part 5, the data is described; in part 6, the statistical results are presented; and finally, in part 6, conclusions are made. 2. TEST HYPOTHESIS The purpose of this paper is to investigate the presence of short term traders, during the dividend period, for stocks quoted in the Lisbon Stock Exchange, from 990 to 999. If the price adjustment on the dividend day is not the necessary amount to exactly compensate the value of the dividend paid, after allowing for tax effects and transaction costs, then there will be an arbitrage opportunity that rationally, should be explored by short term traders. Another way of putting this: it is the existence of a strong short term trading activity that will force the price adjustment to be an equilibrium, where no abnormal returns can be made through arbitrage. If this short term trading hypothesis is true, than we should observe the following: a positive abnormal volume of transactions, during the dividend period; a negative relationship between the intensity of short term trading activity and transaction costs; a positive relationship between the intensity of short term trading activity and the average of transaction volumes, because the risk of reverting unhedged positions on the stock is reduced, besides the fact that high volumes will probably allow the negociation of lower transaction costs. 4

5 3. METHODS: EVENT TIME AND CALENDAR TIME In this work, some concepts will be used, and they should be clarified from the start. So, by dividend day, or day 0 (zero), or even dividend distribution day we mean the first market session where the stock is transacted without the right to the dividend. Note that day 0 might not coincide with the day when the dividend is, in fact, paid to the stockholder. As a corollary, the last day where the stock is transacted with the right to the dividend will be generally defined as day. However, in the Lisbon stock market, until 997, the last session where the stock was transacted with the right to the dividend, occured five sessions before the dividend day, that is, day 5. We also define dividend period, as the group of eleven market sessions surrounding the dividend day, specifically, from day 5 to day +5. The purpose of defining a dividend period is to analyze transactions volume behavior, in the market sessions occurring in the neighbourhood of the dividend day. The empirical testing of the short term trading hypothesis, that is, the hypothesis that there will exist an abnormally excessive volume of transactions during the dividend period, can be done using two alternative methods: Event time method: for each event (dividend day, defined as day 0), the normal transactions volume was estimated as the average of the transactions on the stock during the 40 days ranging through days 64 to 25, relative to day 0. Then, for each of the days surrounding day 0, that is, from day 5 to day +5, we compute the abnormal transactions volume, calculated as the difference between the transactions volume on that day, and the estimated normal volume. Calendar time method: Each observation consists of the average of all the firms that paid dividends on a specific calendar date. The abnormal volume (for every day from -5 to +5) is calculated as the difference between the average of the volumes of the stocks that paid dividends on the same calendar day, and the estimated normal volumes of those firms. The normal volume of transactions of 5

6 the firms that paid dividends in a specific calendar date was determined as the average of the volumes of those stocks, in the period ranging from day -64 to day Every calendar day has the same weight, independently of the number of firms that are pooled on that observation. An advantage of the calendar time method is that it reduces the problem of the aggregation of the data, which will affect the quality of the statistical results, if the observations are not independent from each other. In fact, the hypothesis of independence between the observations is questionable, a priori, for the firms that paid dividends on the same calendar day, because all of them have been subject to the same common influences, namely, the global trend of the stock market. This potential shortcoming of the event time method does not affect the results obtained by the calendar time method. 4. APPLYING THE METHODS: THE SAV STATISTIC For each observation i the average transactions volume, V, and the standard deviation of transactions volume, σ(vi), were calculated for the period ranging from day -64 to day -25, relative to the dividend day (day 0). Thus, the average volume is defined as: i V () 25 i = V it 40 t= 64 where, i =,..., N, is the number of observations in the sample; V it = transactions volume of stock i on day t. The standard deviation of the transactions volume during [-64,-25], for each observation i, is given by: σ 25 V 39 V V t = 64 ( i) = ( it i) 2 2 (2) 6

7 For each observation i, by deducting from the transactions volume of each day, the average volume of transactions, we obtain a residual, which is the abnormal volume on day t: AV it = V V (3) it i To correct possible heteroscedasticity problems associated to the fact that there are significative differences in the sizes of the transactions volumes of the different firms, the data has to be standardized. Dividing AV it by the standard deviation of transactions volume during [-64,-25], we obtain the standardized abnormal volume, SAV it : SAV it AV σˆ it = (4) ( V ) i The standardized abnormal volume can be used to test the hypothesis of short term trading in the dividend period. Consistently with this hypothesis, we should observe a significant increase of the transactions volume, in the dividend period. Thus, admitting that the volumes of the several firms (observations) are independent of each other, and normally distributed, the null hypothesis that the average of abnormal volumes is zero can be tested using the following t-statistic: T SAV t = SAV = (5) σˆ ( SAV ) where σˆ(sav SAV ) = N t SAV it N i=, is the average of abnormal transactions on day t, and is the standard deviation of the average of abnormal transactions, estimated from the standard deviations of all the observations included in the sample, in the period [-64, -25]: 7

8 N ˆ( ) SAV it σ SAV = 39 SAV it (6) N i= t= t= 2 Because SAV it is a standardized variable, with 0 mean and standard deviation of, in the period [-64;-25], we have ˆ( σ SAV ) =, and the T statistic can be simplified to: 2 N T SAV t = (7) N which has a t-student distribution, with 39 degrees of freedom. 5. THE DATA The construction of the database began by taking all the quoted firms in the Lisbon stock exchange, in the period from 990 to 999. Starting from this potential universe, the number of firms/observations was initially narrowed down, by excluding: () the firms that had several market sessions without being transacted, and (2) the firms that became quoted very closely to the dividend day, making it impossible to calculate the transactions volume for all the days in the [-64;-25] period. After this initial selection of the data, the sample included 393 observations spreading through 70 calendar dates, which means that the calendar time method uses only around 43% of the observations that are used by the event time method. A priori, the reduced coincidence of dividend distributions in the same calendar days (on average 2.3 firms paid dividends on the same calendar day), suggests that the problem of dependence between the observations might not be important, and so, the calendar time method should not lead to very different results, comparatively to the event time method. In any case, both methods were applied to the data. For the (40 days between -64 and +25)- = 39 degrees of freedom. 8

9 significancy tests, we took into account the work of Brown and Warner (980), who discussed the most appropriate tests for event studies. 6. RESULTS In this part, we look at the results of the tests that will confirm or invalidate the hypothesis that there are short term traders active in the market, during the dividend period. In part 6., we test the existence of an abnormal volume of transactions, in the dividend period. In part 6.2, we test the existence of a negative relationship between the intensity of short term trading activity and transaction costs. Finally, in part 6.3, we test the existence of a positive relationship between the intensitity of short term trading and the average volume of transactions. From these results, we will draw a conclusion on the validity of the short term trading activity, during the dividend period. 6.. DETECTING ABNORMAL VOLUMES OF TRANSACTIONS 6... FIRST RESULTS In the years 992 through 997, the firms that paid dividends had their stock unquoted during the four sessions preceding the dividend day. As a consequence, the volume of transactions on those days is null. If we included these observations in the calculation of the abnormal volume, this would seriously bias the results on days 4 through. Taking this into account, we split our original sample (sample A) into two subsamples, one for the period where the stock was unquoted during the four sessions prior to the dividend payment (sample B) and another including all observations where the stock remained quoted until the dividend day (sample C). Tables Ia and Ib show the statistical results for these three samples, applying the event time method and the calendar time method, respectively. For the reasons stated, no abnormal volumes where calculated on days 4 through, for samples A and B. There is still another possible bias on the volumes transacted on the other days of the dividend period, although not directly quantifiable, which is the fact that the transactions that 9

10 could not be done on days 4 through, tended to cumulate on the days before the suspension (until day 5) or after the suspension (day 0). 0

11 Table Ia. Abnormal Volume of Transactions in the Dividend Period (Event Time Method) Day Sample A (Total) Number of Observations: 393 Average Volume Transact. (000 EUR) 849 (c) (c) (c) (c) Average Abnormal Vol.(000 EUR) (a) -69 (c) (c) (c) (c) Weighted Average Abnormal. Vol. (b) 45.23% (c) (c) (c) (c) 6.06% 39.94% 23.66% 93.2% 6.2% 40.27% SAV (T statistic) 5.003** (c) (c) (c) (c) 3.047** 4.33** 5.224** 7.564** 6.945** 2.606** Sample B (unquoted on days -4 through -) - Number of Observations: 87 Average Volume Transact. (000 EUR) 97 (c) (c) (c) (c) Average Abnormal Vol.(000 EUR) (a) -45 (c) (c) (c) (c) Weighted Average Abnormal. Vol. (b) 47.5% (c) (c) (c) (c) 59.58% 62.60% 38.04% 60.5% 43.33% -8.26% SAV (T statistic) 3.93** (c) (c) (c) (c) 7.6** 4.799** 3.963** 5.5** 5.48** Sample C (quoted on days -4 through -) - Number of Observations: 206 Average Volume Transact. (000 EUR) Average Abnormal Vol.(000 EUR) (a) Weighted Average Abnormal. Vol. (b) 43.49% 24.56% 2.23% 86.29% 78.07% 62.40% 9.37% % 22.73% 77.26% 84.32% SAV (T statistic) 3.82** 2.355* ** 9.025** 0.769** ** 5.54** 4.687** 3.628** N (a) average abnormal volume on day t, obtained from AV = AV it, where t=-5,...,+5 and i=,...,n. N i= (b) the values presented in the table result from: first, for each observation, the ratio between the abnormal volume and the normal volume was calculated; second, we N AVit determined the average of all the observations as. N i= Vi (c) results are not displayed for these days, because these samples include observations where the stock was unquoted in the 4 sessions before the dividend day. (*) significant at the 5% level. (**) significant at the % level.

12 Table Ib. Abnormal Volume of Transactions in the Dividend Period (Calendar Time Method) Day Sample A (Total) - Number of Observations: 70 Average Volume Transact. (000 EUR) 992 (c) (c) (c) (c) Average Abnormal Vol.(000 EUR) (a) -98 (c) (c) (c) (c) Weighted Average Abnormal. Vol. (b) 2.54% (c) (c) (c) (c) 63.3% 37.86% % 68.22% 7.75% 68.83% SAV (T statistic) (c) (c) (c) (c) 2.47** 2.283* 8.95** 6.822** 5.746** 3.30** Sample B (unquoted on days -4 through -) - Number of Observations: 7 Average Volume Transact. (000 EUR) 68 (c) (c) (c) (c) Average Abnormal Vol.(000 EUR) (a) -85 (c) (c) (c) (c) (c) (c) (c) (c) 27.30% 54.82% 6.47% 39.24% 59.5% % (c) (c) (c) (c) 2.338** 2.335** * 5.023** Sample C (quoted on days -4 through -) - Number of Observations: 99 Average Volume Transact. (000 EUR) Weighted Average Abnormal. Vol. (b) -8.37% SAV (T statistic) Average Abnormal Vol.(000 EUR) (a) -07 Weighted Average Abnormal. Vol. (b) 42.99% SAV (T statistic) % -.88% 49.73% 54.63% 89.5% 25.70% % 260.7% 80.53% 35.8% ** 6.597** 3.938** ** 7.462** 3.275** 4.79** N (a) average abnormal volume on day t, obtained from AV = AV it, where t=-5,...,+5 and i=,...,n. N i= (b) the values presented in the table result from: first, for each observation, the ratio between the abnormal volume and the normal volume was calculated; second, we N AVit determined the average of all the observations as. N i= Vi (c) results are not displayed for these days, because these samples include observations where the stock was unquoted in the 4 sessions before the dividend day. (*) significant at the 5% level. (**) significant at the % level.

13 The results show evidence of a positive abnormal volume in the dividend period, for both methods, in samples A and C, but stronger in this last sample. In sample B, the results show that the transactions volume on days 5 and [0;+5] is smaller than the normal volume, determined from days [-64;-25]. However, the standardized abnormal volume is positive in most of the days within the dividend period, and the SAV statistic allows rejecting the null hypothesis of no abnormal volume in the dividend period, at very low significancy levels. These apparently contradictory results could be explained if the reduction in the transactions volume is more common in firms with higher transaction volumes, which would affect strongly the unstandardised abnormal volume and if, at the same time, most of the other firms increased their transactions volume during the dividend period, which would result in an increasing SAV statistic. The results obtained from sample C show evidence of positive abnormal volumes in most days within the defined dividend period [-5;+5]. The SAV statistic is significantly different from zero, in most of the dividend period. The null hypothesis is rejected at lower significancy levels, when applying the event time method. From this results, we can draw two main conclusions: first, for all three samples, there is a statiscally significant abnormal volume in the dividend period; second, the results change little, with the methods applied, event time or calendar time. We will proceed our analysis presenting only the results from the event time method, but we will also refer the cases where the calendar time method would lead to different results REFINING THE SAMPLE The detection of a positive abnormal volume, in most of the days within the dividend period, is consistent with evidence that there are short term traders active in the market, exploring the profit opportunities linked to dividend distributions. However, a closer inspection of the data revealed the presence of observations where firms recorded excessively high abnormal volumes during the dividend period, 3

14 possibly arising from contemporaneous events, for example, a public offering or a hostile take-over. This few observations could have a strong impact on our statistical tests. In table II, we present the most remarkable cases. Table II. Firms with excessively high volumes of transactions during the dividend period Firm Year Average Volume Transactions Period [-64,-25] Period [-5,+5] Standard deviation Day Volume of Transactions on that day Cires Reditus Banco Fomento Exterior Inapa Inapa The impact of the observations presented in Table II is shown on Graphs Ia and Ib. Graph Ia : Weighted Abnormal Volume 400% 350% 300% Weighted AV 250% 200% 50% 00% Sample A Sample B Sample C 50% 0% -5-50% Day The peak of abnormal volume on day +2, for sample C, is completely explained by the volume transacted by Banco Fomento e Exterior, on that day, which was 690 times higher than the normal volume for that stock in the period [-64;-25]. In sample C, the cumulated weighted abnormal volume has increased more expressively on days [+;+5], that is, on the days following the distribution. 4

15 Graph Ib : Cumulated Weighted Abnormal Volume 200% Cumulated Weighted AV 000% 800% 600% 400% 200% Sample A Sample B Sample C 0% Day With the purpose of controlling the impact on the statistical results of this bias, a total of 24 observations where excluded from the original sample. The observations excluded were those which, in at least one of the days within the dividend period, the abnormal volume exceeded more than ten times the estimated standard deviation for the period [-64;-25]. If these transactions resulted from short term trading activity, we would expect to find a positive abnormal volume before and after the dividend day. In general, this condition of symmetry is not observed in the 24 outlier observations As these abnormal volumes can hardly be attributed to short term trading activity, and reflect, instead changes in the stockholder structure of the firm, we tested the robustness of our initial results to the exclusion of these outliers. We can look at the results in Table III and Graphs IIa and IIb. It is clear that for any of the samples, that now exclude outliers, A-24, B-24, and C- 24, the positive abnormal volume became lower. At the same time, as it was expected, the peak of abnormal volume present is sample C, disappeared in sample C-24. The statistical results also change substantially. In the new samples resulting from the exclusion of outlier observations, the abnormal volume is only statistically significant on days 0 and +4 in sample A-24, day 0 in sample B-24, and days -2, - and +4, in sample C-24. 5

16 Table III. Abnormal Volume of Transactions in the Dividend Period, Excluding Outliers (Event Time Method) Day Sample A-24(Total; no outliers) - Number of Observations: 369 Average Volume Transact. (000 EUR) 877 (c) (c) (c) (c) Average Abnormal Vol.(000 EUR) (a) -9 (c) (c) (c) (c) Weighted Average Abnormal. Vol. (b) -9.42% (c) (c) (c) (c) 3.36% -6.29% -3.64% -0.5% 6.59% % SAV (T statistic) (c) (c) (c) (c).726* * Sample B-24 (unquoted on days -4 through ; no outliers) - Number of Observations: 78 Average Volume Transact. (000 EUR) 83 (c) (c) (c) (c) Average Abnormal Vol.(000 EUR) (a) -69 (c) (c) (c) (c) Weighted Average Abnormal. Vol. (b).3% (c) (c) (c) (c) 3.06% 2.64% 6.34% 3.39% -2.05% -0.52% SAV (T statistic) (c) (c) (c) (c) 2.706** Sample C-24 (quoted on days -4 through ; no outliers) - Number of Observations: 9 Average Volume Transact. (000 EUR) Average Abnormal Vol.(000 EUR) (a) Weighted Average Abnormal. Vol. (b) -9.42% % -5.99% 6.32% 9.8% % -4.6% -2.94% -3.47% 4.64% % SAV (T statistic) ** 4.483** * N (a) average abnormal volume on day t, obtained from AV = AV it, where t=-5,...,+5 and i=,...,n. N i= (b) the values presented in the table result from: first, for each observation, the ratio between the abnormal volume and the normal volume was calculated; second, we N AVit determined the average of all the observations as. N i= Vi (c) results are not displayed for these days, because these samples include observations where the stock was unquoted in the 4 sessions before the dividend day. (*) significant at the 5% level. (**) significant at the % level.

17 Graph IIa : Weighted Abnormal Volume 40% 30% 20% Weighted AV 0% 0% -5-0% % Sample A-24 Sample B-24 Sample C-24-30% -40% -50% Day Graph IIb : Cumulated Weighted Abnormal Volume 80% 60% Cumulated Weighted AV 40% 20% 0% -5-20% % -60% -80% -00% -20% Day Sample A-24 Sample B-24 Sample C-24 Graphs IIa and IIb also show clearly the decreased significancy of the weighted abnormal volume, which was more evident in samples A and C. On the whole, the evidence of abnormal volumes in the dividend period is weaker. Nevertheless, in sample C-24, the positive abnormal volume is still significant on days 2, - and +4. 7

18 Because the evidence on abnormal volumes became weaker, with the refined sample, we tried to identify a criteria for the definition of subsamples, where the short term trading activity might be more evident. This analysis is carried out in parts 6.2 and THE ABNORMAL VOLUME AND TRANSACTIONS COSTS The first adjustment consisted of excluding the observations of year 990 from sample C, which will now include only observations from the year 997 on (sample D). One theoretical reason to do this is that, in 995, some components of the transaction costs were liberalised, reducing the overall transaction costs. This reduction should have stimulated short term trading, making more probable the detection of a positive abnormal transactions volume in the dividend period. The second reason for using this subsample, is that it now only includes the observations were the stock remained quoted in days 4 through, thus avoiding the potential bias that the suspension might cause. Table IV shows the results obtained for sample D, that includes observations from 996 and onwards, after the 24 outlier observations were removed. These results confirm the existence of a positive abnormal volume, on days -2, - and +4, now more significant that in the previous samples. However, for the calendar time method, the abnormal volume on day +4 is not significant. In the remaining days of the dividend period, the abnormal volume is always negative, although not statistically significant (except for day +5) THE ABNORMAL VOLUME AND THE LIQUIDITY OF STOCK The presence of short term traders in the dividend period can be better clarified, by checking if the abnormal volume is stronger in a subsample including only the stocks with higher liquidity, where we should expect to find higher trader activity. There are two reasons for this: first, high volumes facilitate quick round-trip transactions; second, higher volumes will normally allow lower transaction costs. 8

19 Table IV. Abnormal Volume of Transactions in the Dividend Period, and Costs of Transaction (Event Time Method) Day Sample D (Observations starting 997, no outliers) - Number of Observations: 28 Average Volume Transact. (000 EUR) Average Abnormal Vol.(000 EUR) (a) Weighted Average Abnormal. Vol. (b) -24.8% % -6.25% 2.89% 34.07% -6.62% % -2.30% -0.75% 26.37% % SAV (T statistic) ** 6.68** ** N (a) average abnormal volume on day t, obtained from AV = AV it, where t=-5,...,+5 and i=,...,n. N i= (b) the values presented in the table result from: first, for each observation, the ratio between the abnormal volume and the normal volume was calculated; second, we N AVit determined the average of all the observations as. N i= Vi (c) results are not displayed for these days, because these samples include observations where the stock was unquoted in the 4 sessions before the dividend day. (*) significant at the 5% level. (**) significant at the % level.

20 Table V. Abnormal Volumes of Transactions in the Dividend Period, By Quartiles (Event Time Method) Day Sample D - Number of Observations: 32 Average Volume Transact. (000 EUR) Average Abnormal Vol.(000 EUR) (a) Weighted Average Abnormal. Vol. (b) -46.2% -34.0% % 24.86% 5.89% -60.9% % -28.2% 20.76% 99.9% % SAV (T statistic) ** Sample D2 - Number of Observations: 32 Average Volume Transact. (000 EUR) Average Abnormal Vol.(000 EUR) (a) Weighted Average Abnormal. Vol. (b) -28.3% % % -42.5% -8.3% -5.4% % -5.42% -27.9% 8.20% % SAV (T statistic) Sample D3 - Number of Observations: 32 Average Volume Transact. (000 EUR) Average Abnormal Vol.(000 EUR) (a) Weighted Average Abnormal. Vol. (b) -2.9% 4.9% 28.9% 70.75% 3.4% 4.36% -0.53% 9.40% -7.58% -2.60% -7.99% SAV (T statistic) ** 4.369** Sample D4 - Number of Observations:32 Average Volume Transact. (000 EUR) Average Abnormal Vol.(000 EUR) (a) Weighted Average Abnormal. Vol. (b) -.89% -7.59% 2.8% 34.% 5.58% 3.2% -3.73% 25.02% 2.72% 9.99% -.0% SAV (T statistic) ** 9.976** 3.464** * 2.074* N (a) average abnormal volume on day t, obtained from AV = AV it, where t=-5,...,+5 and i=,...,n. N i= (b) the values presented in the table result from: first, for each observation, the ratio between the abnormal volume and the normal volume was calculated; second, we N AVit determined the average of all the observations as. N i= Vi (c) results are not displayed for these days, because these samples include observations where the stock was unquoted in the 4 sessions before the dividend day. (*) significant at the 5% level. (**) significant at the % level.

21 Sample D was divided by quartiles, based in the criteria of the average transactions volume in the period [-64;-25]. Thus, the first quartile (D) includes observations with an average transactions volume of 20 thousand EUR, the second quartile (D2) an average of 33 thousand EUR, the third quartil (D3) an average of 68 thousand EUR, and finally, the fourth quartile (D4) of 9399 thousand EUR. The results obtained for the quartiles of sample D can be read in table V and Graph III. Graph III : Cumulated Weighted Abnormal Volume 300% 200% Cumulated Weighted AV 00% 0% % -200% -300% Sample D Sample D2 Sample D3 Sample D4-400% Day For the first two quartiles, D and D2, the abnormal volume was negative in most days of the dividend period. In quartile D2, we have a negative abnormal volume in all days of the dividend period, except on day +4, where it is marginally positive, in terms of weighted abnormal volume. The two quartile that include the bigger firms, that is, the ones that were more transactioned in the stock market, D3 and D4, the cumulated weighted abnormal volume during the dividend period is positive, particularly in D4. In this quartile, the null hypothesis of no positive abnormal volume is rejected at a significancy level of %, for days 2, -, and 0, and it is also rejected on days +2 and +3, at a significancy level of 3%. 2

22 In quartil D3, the evidence of a positive abnormal volume is also significant at the 2% level, for days 3 and 2. In short, we detected the presence of a statistically significant positive abnormal in the two quartiles containing the more transactioned firms. In the two quartiles containing the smaller firms, the null hypothesis could not be rejected. 7. CONCLUSIONS The transactions volume behavior in the dividend period as shown evidence of a positive abnormal volume, although not very strong. Specifically, in the sample excluding outliers, that is, observations in which at least in one of the days in the dividend period, the abnormal transacted volume exceeded more than ten times the standard deviation of the normal volume, we found statistical evidence of a positive abnormal volume in days -2, -, and +4. In the other days, the abnormal volume is not statistically significant. These facts suggest that dividend distributions generate profit opportunities, which lead short term traders to the market, during the dividend period. These results and interpretation are consistent with another study of the author, Borges (2002), where it is shown that the stock price adjustment on the dividend day falls short of the ajustment theoretically expected and explainable by the unfavorable taxation of dividends. In that study, the author finds evidence of unexplored profit opportunities, as these are not exhausted on the dividend day, which is consistent with the presence of short term traders in the dividend period, besides the dividend day. On the other hand, the empirical results of the present study are consistent with the hypothesis of a negative relationship between short term trading and transaction costs. In the sample including only observations starting from 996, year in which the transaction costs reduction began to be relevant, the evidence of a positive abnormal volume became stronger. 22

23 The partition of the sample in 4 quartiles, based in the average volume of transactions, showed that short term trading activity is stronger in the quartiles including the more transacted firms. This is justified by the fact that firms with higher volumes facilitate the buying and selling of the stock, by short term traders, when trying to seize the profit opportunities that emerge in the market. In other words, firms with higher volumes will be less risky for short term traders. At the same time, higher volumes will also contribute to a reduction of transaction costs. Accordingly, the very low liquidity of a subset of the stocks quoted in the Lisbon stock market, which in some cases do not transact during several days, very likely keep short term traders away, due to the higher risk, both in terms of price and in terms of the ability to complete the round-trip transaction. We can conclude that inspite of the effort done by the Portuguese authorities in improving the legal and operational framework, and of the rapid growth of the market during the nineties, the Lisbon stock exchange continued to show some symptoms of inefficiency. REFERENCES Borges, Maria R. (2002) Fiscal Effect in Dividend Distributions, Proceedings of XII Jornadas Luso- Espanholas de Gestão Científica, Beira Interior University, Vol. II Finance, Covilhã. Brown, Stephen J. and Warner, Jerold B. (980) Measuring Security Price Performance, Journal of Financial Economics, Nº8. Eades, Kenneth M.; Hess, Patrick J. and Kim, E. Han (984) On Interpreting Security Returns During the Ex-Dividend Period Journal of Financial Economics, Vol. 3, p Elton, Edwin J. and Gruber, Martin J. (970) Marginal Stockholder Tax Rates and The Clientele Effect The Review of Economics and Statistics, Vol. LII, nº, February, p Elton, Edwin J.; Gruber, Martin J. and Rentzler, Joel (984) The Ex-Dividend Day Behavior of Stock Prices: A Re-Examination of the Clientele Effect: A Comment The Journal of Finance Vol. 39, Nº2, June, p Kalay, Avner (982) The Ex-Dividend Day Behavior of Stock Prices: a Re-Examination of The Clientele Effect The Journal of Finance, Vol. 37, nº 4, September, p Lakonishok, Josef and Vermaelen, Theo (983) Tax Reform and Ex-Dividend Day Behavior The Journal of Finance Vol. XXXVIII, nº 4, September, p

24 Lakonishok, Josef and Vermaelen, Theo (986) Tax-induced Trading Around Ex-dividend Days Journal of Financial Economics Vol. 6, nº 3, July, p Lisbon Stock Exchange Statistics Michaely, Roni (99) Ex-Dividend Day Stock Price Behavior: The Case of the 986 Tax reform Act The Journal of Finance Vol. XLVI, nº 3, July, p Miller, Merton H. and Scholes, Myron S. (982) Dividend and Taxes: Some Empirical Evidence Journal of Political Economy, Vol. 90, nº 6, December, p.8-4. Poterba, James M. and Summers, Lawrence H. (984) New Evidence That Taxes Affect the Valuation of Dividends The Journal of Finance Vol. XXXIX, Nº5, December, p