The Effect of Canadian and American Capital Gains Taxation on the Seasonality of Stock Prices. Devan Mescall

Transcription

1 The Effect of Canadian and American Capital Gains Taxation on the Seasonality of Stock Prices Devan Mescall I d like to thank Ken Klassen, Patricia O Brien and Alan Webb for their valuable comments. I would also like to thank Sean Speers and Colin Wallace for their computer support and instruction.

2 Abstract The two most prominent theories for explaining the January Effect are the tax-lossselling hypothesis and the window dressing hypothesis. Poterba and Weisbenner(2001) find evidence which distinguishes the tax-loss-selling hypothesis from the window dressing hypothesis by attributing changes in turn-of-the-year returns to changes in U.S. tax policy. However, they are unable to rule out alternative explanations caused by the market. I apply their model to a sample of Canadian-only-listed firms in order to rule out potential confounds. I find that in the same results can be found in the absence of U.S. tax policy. I then develop a model to test for changes in the turn-of-the-year return based on changes in the taxation of Canadian capital gains in 1987 and I find the expected changes in turn-of-the-year returns. However, upon further investigation these trends can be found in the U.S. data absent of Canadian policy as well. Unlike Poterba and Weisbenner, I conclude that, the tax-loss-selling hypothesis can not be solely responsible for the results in either study. 1

3 I. Introduction The January effect refers to the anomaly where firms experience abnormal returns in the first five days of January. The persistence of these returns stands in opposition to the efficient market hypothesis and as such has been a target of investigation for the past twenty years. A number of competing explanations have been developed to explain the anomaly including turn-of-the-year liquidity and the undefined small-stock risk factor. However, the two most prevalent are the window dressing and the tax-loss-selling hypotheses. Although these two hypotheses are competing explanations they are not mutually exclusive. Separating the effects of each hypothesis has rarely been accomplished. Poterba and Weisbenner (2001) claim to find clear evidence of the tax-loss-selling hypothesis by investigating tax changes in the U.S. However, the authors are unable to rule out other market forces as an alternative explanation to U.S. tax policy for their results. I apply their model to a sample of Canadian-only-listed companies as a control group to their study. The integration of the Canadian and U.S. markets provide an opportunity to test the model on a sample that is susceptible to the same market forces in the absence of U.S. tax policy. I find the same results as Poterba and Weisbenner in the Canadian sample with the absence of U.S. policy. Therefore, it is unlikely that the tax policy is responsible for the results found by the authors. I investigate further by developing a model to test the effect of Canadian changes to capital gains taxation on turn-of-the-year returns. I find the expected results showing evidence in favour of the tax-loss-selling hypothesis. However, further testing finds the Canadian results in the U.S. data where Canadian tax policy is absent. The 2

4 evidence shows that the tax-loss-selling hypothesis can not be solely responsible for the results in either study. The rest of the study is organized as follows: section II is a review of relevant literature, section III develops the hypotheses, section IV discusses the research design, section V discusses the results, and finally section VI concludes the study. II. Relevant Literature In 1983, Kiem discussed an apparent contradiction of the efficient markets theory where small firms can generate excess returns in January. Twenty years later researchers are still divided on the source of this phenomenon. Haugen and Lakonishok (1987) conjecture that the January effect is caused by institutional investors who sell off poorly performing shares before the year end in order to avoid the appearance of having been invested in poor stocks. Athanassakos (2002) supports Haugen and Lakonishok. By looking at large firms that would likely be unaffected by individual investors, Athanassakos finds evidence that firms with poor last quarter earnings had higher turn-of-the-year returns than those with profits. He attributed this to gamesmanship or window dressing by institutional investors. Alternatively, the tax-loss-selling hypothesis conjectures that the anomaly is caused by individual investors who choose to sell off shares in loss positions in order to recognize the potential tax benefits. This creates a unique setting for investigating the fundamental tax research questions framed by Shackleford and Shevlin (2001) Do taxes matter? This hypothesis, though widely investigated, has produced mixed results. For example Berges et al (1984) found evidence of the January effect in Canada before capital gains were even taxed. Therefore, it seems unlikely that this seasonal effect can be caused solely by taxation. 3

5 Despite some contrary findings, support for the tax-loss-selling hypothesis has continued to evolve. Johnston and Cox (2002) refined a previous study by Ligon (1997) which attempted to distinguish between all of the competing explanations in one study. Johnston and Cox found that the Ligon study had some important strengths as it looked at the anomaly using macroeconomic variables. Unlike Ligon, Johnston and Cox chose to develop a regression model looking only at variables which would explain the abnormal January returns using the tax-loss-selling hypothesis. The authors used a New York Stock Exchange (NYSE) portfolio to calculate the abnormal returns from and found all but two of the coefficients for their explanatory variables were significant and as predicted. They concluded that tax-loss-selling does contribute to the January effect. The use of changes in the taxation of capital gains legislation has produced some of the most supportive findings in favour of the tax-loss-selling hypothesis. Bhabra et al (1999) and Gibson et al. (2000) discuss similar findings using the Tax Reform Act of 1986 (TRA 1986) which gives rise to a November effect. The TRA 1986 made it mandatory for all mutual fund companies to have an October 31 year end. Tests performed by Bhadra et al (1999) found abnormal returns could be found in November after 1986 but not before. Gibson et al. (2000) found that these returns were eliminated after 1990, which was the year the TRA 1986 was implemented in its entirety. Like Reinganum and Shapiro (1987), evidence of the November effect is significant as it dismisses alternative explanations related directly to January. However, the study is not able to dismiss alternative explanations that market factors that may have lead to the legislative changes and not the legislative changes themselves are the cause of the November anomaly. 4

6 One potential solution is to look into the phenomenon internationally. Reinganum and Shapiro (1987) recognizing the benefits of testing the tax-loss-selling hypothesis in a non-u.s. tax environment, perform a study looking at the January effect in the United Kingdom. The U.K. provides a unique environment for a study of the seasonal effects of capital gains tax legislation as they have an April 6 year end for individuals and a December year end for corporations. Therefore, the authors were able to test for both an April and January effect. Also data was available pre-1965 when Britain did not tax capital gains at all. The authors were able to find evidence of January and April effects post-1965 and no evidence pre Although this was strongly in favour of the tax-loss-selling hypothesis, further tests consisting of creating portfolios of firms in loss positions were unsuccessful at finding the hypothesized results. The authors acknowledge throughout the paper that the data had several problems and propose future studies with international data of better quality and more U.S. like qualities. In this way Canadian data would be better suited to compliment these studies as quality data is available.. It is important to consider what research has already been performed in the Canadian setting concerning seasonality and the taxation of capital gains. Despite potential benefits of the Canadian setting as previously discussed, few studies have taken advantage. Berges et al (1984) looked at Canadian stock data from to test if a January effect could be observed in Canada and if evidence of the tax-loss-selling hypothesis could be seen when the first taxation of capital gains was introduced in The authors found evidence of the January effect in Canada. However, they were unable to find support for the tax selling hypothesis as the anomaly existed before 1973 and tests using a loss portfolio method were unsuccessful. Tinic et al (1987) attempted to expand on the findings of Berges et al (1984) 5

7 by looking at the impact of American investors. By comparing portfolios of cross-listed companies and Canadian-only listed companies, the authors were able to show evidence of a change in Despite some findings in favour of the tax-loss-selling hypothesis, both Canadian studies find evidence of the January effect that is not easily explained by tax. In 2004, Chen and Singal attempt to disentangle the potential sources of this anomaly by considering each explanation separately. By assessing each stocks tax-loss-selling potential, they find evidence in favour of the tax-loss-selling hypothesis. Unfortunately, the same factors that give a stock potential for tax-loss-selling also make it a favourable choice for institutional investors to sell for window dressing. The authors attempt to remove this confound by looking for a similar effect midyear when institutional investors may also have incentives to cleanup their portfolios. They do not find a similar effect in June or July and use this as evidence against the window dressing hypothesis. This conclusion relies on a number of assumptions about institutional investors and, like many studies, can not distinctly rule out institutional investors effect on the January observations. However, Poterba and Weisbenner (2001) [P&W] devise a method to isolate evidence directly related to the tax-loss-selling hypothesis. They use changes in tax policy that effect individuals to create periods when one would expect stronger and weaker January effects. They then use a regression model and create hypotheses of the directions of coefficients based on the tax-loss-selling hypothesis. The authors find support for all of their hypotheses. They conclude that the tax-loss-selling hypothesis contributes to the January effect as the tax changes would not have affected institutional investors or be contributed to other rival explanations. However, the authors recognize that other market factors at work during the time of legislative changes can not be ruled out as alternative explanations. 6

8 I attempt to remove these confounds by replicating the study, using firms in a like economy but with out the effects of the U.S legislation. Because of the integration of the Canadian and U.S. economies, a study using Canadian-only listed data can be used as a control group against these confounds to further investigate these findings. III. Hypothesis Development To test for evidence of the tax-loss selling hypothesis, P&W break up the U.S. data into three different tax regimes to reflect U.S. tax policy from 1963 to Although there are many facets to capital gains tax policy, P&W focus on the treatment of short term and long term capital losses over the period. As summarized by Poterba and Weisbenner (2001), the regimes are: Regime I. This includes years when the long term holding period was six months and long-term losses were just as valuable as short-term losses for tax reduction. This regime covers the month of January in years and Regime II. This includes years when the holding period was six months, and longterm losses were only half as valuable as short-term losses for reducing tax. It comprises the January returns from 1970 to 1976 and 1985 to Regime III. This includes the years with a 12 month holding period. It comprises the January returns from 1979 to 1984 as well as 1989 to The strong tax incentives of regime II to recognize a loss as a short term loss differentiate it from the other two regimes. This is the focus of P&W s study. The authors hypothesize and find four key results that are evidence of the tax-loss-selling hypothesis. 1) If the tax-loss-selling hypothesis holds, the turn-of-the-year returns in regime II will be more influenced by losses that occur in the last six months of the year than returns in other regimes. 7

9 2) The difference between the effect of losses in the first half of the year and the effect of losses in the second half of the year on turn-of-the-year returns will be greatest in regime II. 1 3) Losses will have a more statistically significant effect on the turn-of-the-year returns than gains 4) Losses will have a larger effect on turn-of-the-year returns than gains. As a result, they conclude tax-loss-selling has an effect on turn-of-the-year returns. This findings are significant as they show evidence of the tax-loss-selling hypothesis that is not confounded by the window dressing theory as institutional investors are not affected by these tax policies. However, they are unable to rule out potential confounds that may be occurring in the market over the same periods as the tax regimes. P&W claim that because the regimes are non-sequential a number of confounds can be ruled out. However, changes in government policy and especially tax policy are often the result of events in the economy. Therefore, an alternative explanation exists. One way to eliminate these confounds is to test the model on data which is susceptible to the same market forces but free of the policy implications. If the same results are found in the absence of these tax policies, then perhaps there is another explanation driving the results. However, if the results are not found in the absence of the policy then many confounds relating to the market can be eliminated and the findings of P&W will be stronger. Canadian data provides an opportunity for a control group to U.S. tax policy as the close economic relationship of the two countries insures many similarities with the exception of policy. Despite this fact, Canadian data is rarely used in this manner. 1 Due to reduced deductibility for long-term losses, a person would not want to hold a loss over the six month short term period. As a result, losses from the first half of the year would be expected to be sold off during the year, before December. 8

10 One unique aspect of Canadian data is that the market is made of companies that are cross-listed in the U.S. and those that are listed only in Canada. I hypothesize that the effects of U.S. legislation will have a greater effect on the cross-listed companies than the Canadianonly-listed companies. Likewise, the Canadian-only-listed companies should show little effect of the U.S. legislation and clearer effects of Canadian legislation than the cross-listed companies. Therefore, the Canadian-listed-only companies should be the best control group for the U.S. companies. This leads to the first hypothesis. H1: Applying the P&W model and regimes to Canadian-only-listed companies should not yield the same results found by P&W Over the past 40 years Canadian policy concerning the taxation of capital gains and losses has evolved. However, Canada makes no distinction between long and short term capital gains so the findings of P&W based on the tax implications of short term and long term gains in regime II, should not be found in the Canadian data. The timing of changes in Canadian policy have differed from the United States. A comparison can be found in Figure 1. Up until 1973, Canada had no tax on capital gains and losses. In 1973, Canada introduced a tax on capital gains where 50% of gains were included in taxable income. Tax losses were eligible to be deducted only against capital gains and could be carried back for a maximum of three years. This prevailed until 1987 when the rate of inclusion was increased to 75% beginning in 1990 with a two year phase-in period of 66 2/3%. Therefore, Canadians in 1988 and 1989 had tax incentives to not recognize capital losses until 1990 when the rate increased in order to save tax on 75% of their loss rather than only 66 2/3%. This leads to our second hypothesis. 9

11 H2: If the tax-loss-selling hypothesis holds, losses should have a much smaller effect on turn-of-the-year returns in 1988 and 1989 than in other years. The inclusion rate remained at a high of 75% for ten years. If the tax-loss-selling hypothesis holds, I would expect losses by the Canadian-only-listed firms to have the largest effect during this period. In 2000, rates were reduced twice; once in February to 66 2/3% and then in October to a 50% rate again. This meant that Canadians could recognize losses at the 75% value for a maximum of three more years due to carryback rules. This policy leads to hypothesis three. H3: Losses for the year 2000, 2001 and 2002 should have a much larger effect on turn-ofthe-year returns than other years By finding these effects of Canadian tax policy on turn-of-the-year returns, more evidence can be gained for the tax-loss selling hypothesis. These effects are unique to noninstitutional investors in the Canadian environment. Therefore their existence can provide further evidence of tax-loss-selling. Like the results of P&W, it is important that these results only hold in the environment where the tax policies apply. Therefore it is beneficial to look for these same results in the U.S. data to rule out the existence of alternative explanations. H4: The effects of Canadian tax policy outlined above will not be found in U.S. firms. IV. Research design I investigate the effect of changes in tax policy on turn-of-the-year returns in four steps. The first step is a replication of the model and data used by Poterba and Weisbenner [2001] (P&W) to investigate U.S. policy using U.S data. The second step is applying the Poterba and Weisbenner U.S. model using Canadian data as a control group to test for 10

12 potential confounds to the previous study. Thirdly, the U.S. model is modified to reflect the Canadian environment and is used to test Canadian tax regimes over the past 40 years using Canadian data. Lastly, the Canadian model is tested using U.S data as a control for changes in Canadian tax policy. Data Two primary data sets are used in this study. The first is data from the CRSP NYSE and AMEX daily files based on the data used by Poterba and Weisbenner. This will be referred to as the U.S data though it does contain some Canadian cross-listed firms. However, the effect of Canadian investors is assumed to be insignificant. The data ranges from 1963 to As the daily returns began in CRSP in July of 1962, this was the largest sample that could be attained. As in P&W, NASDAQ firms are not included as their data is only available back to 1973 and therefore does not include some of the tax policy changes analyzed. 2 The second data set used was TSE data from the CFMRC daily and monthly database. The data spanned from 1950 to Daily data was used back to 1973 which was the limit of the database. In order to conduct the study it was important to have data that matched the U.S. sample back to Therefore, monthly returns were used from 1963 to In order to perform tests 2 and 3, I had to divide the Canadian data into cross-listed companies and Canadian-only-listed companies. As this information was not available from the CFMRC database, the firms were identified by matching them by ticker and name for the 2 My sample is slightly larger but comparable to P&W. In the study, they divide the sample into 3 regimes resulting in samples of 14,932, 20,645, and 32,701 and my sample was 16,002 21,116 and 54,665. Despite the sample differences the results were comparable. 11

13 appropriate year with the U.S. data. 3 Using the list of cross-listed companies from the December 2002 TSX Review the data was then hand checked and corrected to insure all of the firms listed were included. As a result there are 1490 cross-listed firms and Canadian-only-listed firms. 4 Test 1: Replication of Poterba and Weisbenner (2001) In order to further examine the results from the previous study it was important to replicate the study to ensure that the model was being applied properly to the appropriate data set. P&W break up the U.S. data into three different samples based on the tax regimes discussed previously. They then use a regression model to test for differences between the regimes. The model is: R Jani,t = β 1 GAIN July-Dec, i,t-1 +β 2 GAIN Jan-June, i, t-1 +β 3 LOSS July-Dec,i, t-1 +β 4 LOSS Jan- June,i,t- 1+β 5 (1/PRICE i,t ) + Ф i +λ+ε i,t (1) R Jani,t is the dividend inclusive return for the first five days in January. Roll(1983) includes the last trading day in December as well. However, P&W find no difference in their results so I remain consistent by only including the returns in January. The LOSS and GAIN variables are constructed as follows: LOSS July-Dec,i, t-1 =(P Dec30 /P July1-1) (2) LOSS Jan-June,i,t-1 =(P June30 /P Jan1-1) (3) GAIN July-Dec, i,t-1 =(P Dec30 /P July1-1) (4) GAIN Jan-June, i, t-1 =(P June30 /P Jan1-1) (5) 3 Matching solely on tickers was not effective as a review of the tickers used on U.S. exchanges for all firms cross-listed in December of 2002 revealed that only 24% had the same ticker on both exchanges. 4 The percentage of cross-listed firms in December of 2002 was 14.7% in my data the cross-listed makes up 9.31% over the entire sample. This amount appears reasonable when considered as an average from Although there are errors in this measure, due to the time demands of hand checking 15,963 observations the measure appeared reasonable as any error likely biases against my findings. 12

14 The LOSS variables are set equal to zero if the result is positive and the GAIN variables are set equal to zero if the result is negative. 5 The variable 1/PRICE is used as a proxy for trading costs. The terms Ф and λ are firm-specific and year-specific intercepts. Test 2 Application of the U.S. model to Canadian data To test P&W s assertion that their results are driven by U.S. policy, I apply their model to the Canadian data. I divide the Canadian-only-listed data into three samples based on the U.S. tax regimes used by P&W. Test 3- Development and testing of a Canadian model Like P&W, I divide the Canadian sample into three regimes based on the changing tax policy of the last 40 years. These regimes are outlined in Figure 2. In order to reflect the Canadian environment the P&W model was modified to the following: R Jani,t = β 1 GAIN Jan-Dec, i,t-1 +β 2 LOSS Jan-Dec,i, t-1 +β 3 LCAP Dec, i,t-1 +β 4 (1/PRICE i,t ) + Ф i +λ+ε i,t (6) GAIN Jan-Dec and LOSS Jan-Dec are calculated as above but for the entire year as there is no distinction between short and long term capital losses in the Canadian system. The remaining variables are calculated the same. 6 LCAP Dec is the log of market capital and is used as a proxy for size. This was an important control for comparing the cross-listed and Canadian-only-listed companies as the former are generally larger firms than the latter. This also provides a theoretical check as previous studies such as Roll (1983) and Berges et al. (1984) have found evidence linking the turn-of-the-year returns to smaller companies. In order to compare the Canadian-only-listed firms with the entire Canadian sample I use an interaction model as follows: 5 P&W use the price on December 24 to insure that none of the turn-of-the-year returns are reflected in the LOSS and GAIN variables. The use of December 30 biases against my findings. However, the effect is assumed to be minimal as the replication found results consistent with P&W. 6 R Jan for all firms from is the monthly return for January as daily return data was not available. Although this is a crude measure, it was used by both Berges et al and Tinic et al for their Canadian studies. 13

15 R Jani,t = β 1 GAIN Jan-Dec, i,t-1 +β 2 LOSS Jan-Dec,i, t-1 +β 3 LCAP Dec, i,t-1 +β 4 (1/PRICE i,t )+ β 5 CO +β 6 CO*GAIN Jan-Dec, i,t-1 +β 7 CO*LOSS Jan-Dec,i, t-1 + Ф i +λ+ε i,t (7) CO is a dummy variable which is 1 if the firm is listed only in Canada and 0 if it is crosslisted. I also use interaction models to test for evidence of policy changes in and R Jani,t = β 1 GAIN Jan-Dec, i,t-1 +β 2 LOSS Jan-Dec,i, t-1 +β 3 LCAP Dec, i,t-1 +β 4 (1/PRICE i,t )+ β 5 INCR +β 6 INCR*GAIN Jan-Dec, i,t-1 +β 7 INCR*LOSS Jan-Dec,i, t-1 + Ф i +λ+ε i,t (8) R Jani,t = β 1 GAIN Jan-Dec, i,t-1 +β 2 LOSS Jan-Dec,i, t-1 +β 3 LCAP Dec, i,t-1 +β 4 (1/PRICE i,t )+ β 5 *DECR +β 6 * DECR*GAIN Jan-Dec, i,t-1 +β 7 *DECR*LOSS Jan-Dec,i, t-1 + Ф i +λ+ε i,t (9) A dummy variable INCR was created for all firms in 1988 and 1989 when taxpayers are expecting an increase in inclusion rates. Since its tax advantageous to wait to recognize a loss, I expect β7 from equation (8) to be smaller than other years of comparable rates. DECR is a dummy variable which is equal to one for all firm-year observations from 2000 and Since taxpayers losses will be worth a third less when they are no longer able to carry them back to the 75% inclusion rates, I expect them to try to recognize the losses as soon as they can. Therefore β7 in equation (9) should be larger than other years of comparable rates. 7 As LOSS is negative this implies a large negative coefficient. Equation (8) and (9) are tested using a sample of all years when inclusion rates were between 50% and 66 2/3%. Test 4 Applying the Canadian model to the U.S. data Although Canadian data can be used as a control for testing U.S. policy, the reverse can also be true. In the fourth test, I apply the Canadian regimes and model to the U.S. data. As in test 2, if the Canadian tax policies are the cause of the findings then I would not expect the same results from the U.S. data. 7 14

16 V. Results The results of the first test were consistent with P&W. All four of the studies major findings were supported by the replication. Table 1 shows that the coefficient for LOSS July- Dec for regime II is significant and is the largest value of β3 for any regime. The LOSS coefficients are all significant and are all negative. This implies that an increased loss results in a larger return at the turn-of-the-year. The LOSS coefficients are all more significant and larger than the GAIN coefficients.. Lastly as predicted the difference between the LOSS coefficients is greatest for regime II. P&W explain this as a result of the negative tax implications of holding a short term capital loss over the maximum 6 month period. However, the results from test 2 do not support this conclusion. The same results are found in the Canadian-only listed companies which are not subject to U.S. tax policy. Table 1 shows that the coefficient for LOSS July-Dec is the largest and most significant in regime II for the Canadian data as well. This is unexpected as the policy change used by P&W to justify this result does not apply in Canada. Regime II also has the largest difference between the LOSS coefficients. This is also surprising as Canada does not make a distinction between long-term and short-term capital gains and losses. Although not all of the LOSS coefficients are significant, they are generally larger and more significant than the GAIN variables. This seems to agree with general theory that the turn-of-the-year returns are related to firms with losses; however, it does not directly relate to the tax-loss-selling hypothesis. Evidence of the tax-loss-selling hypothesis can be found when the Canadian model is applied to the Canadian data. The tests performed to investigate the markets reaction to the rate increases in 1987 and decreases in 2000 both show evidence in favour of the tax-loss- 15

17 selling hypothesis. In 1988 and 1989 when taxpayers knew a rate increase was imminent, I expected the coefficient on the interaction between the tax increase years (INCR) and the loss variable to be positive. This implies that for those years, a loss has less effect on turn-of-theyear returns. The coefficient is and is significant. In 2000 and 2001 the expected result is the opposite. As taxpayers face a decreasing value for their capital losses, they are expected to recognize them. By doing so the coefficient for the interaction between the variable for these years(decr) and the LOSS variable is expected to be negative. This implies that the LOSS variable for these years is larger and has a stronger effect on turn-of-the-year returns. Evidence of this is found as the coefficient is and significant. Further evidence can be found from the interaction between the Canadian-listed-only dummy variable and LOSS in regime III which is negative and significant. This shows that the LOSS variable for the Canadian-listed-only firms had a stronger effect than the crosslisted firms in the regime when Canadians faced the highest taxes. Also, the coefficient for size was significant and negative which supports prior studies by Roll and Berges which found that the turn-of-the-year returns were strongest for small companies. The coefficients from the Canadian data also seem to show evidence of findings by Gu (2003), who found that the January effect is declining over time. This may be why the coefficients on LOSS, which would be expected to increase over the tax regimes as taxes increase, actually decrease. Although this doesn t preclude the tax-loss-selling hypothesis it appears to show evidence of an omitted variable. This conclusion is further supported by the results of the fourth test. 16

18 Just as the Canadian data showed the same results of P&W in the absence of U.S. policy, the U.S. data shows the Canadian trends in the absence of Canadian policy. The coefficient INCR*LOSS is positive in the U.S. data just like the Canadian sample. Similarly the coefficient for DECR*LOSS is negative in both the U.S. and Canadian data. VI. Conclusion This study raises further questions as to the cause of abnormal turn-of-the-year returns. The strong evidence in support of the tax-loss-selling hypothesis found in Poterba and Weisbenner (2001) appears to be susceptible to alternative explanations as the Canadianonly listed data showed similar results in the absence of the U.S. tax policy. Alternatively, evidence that was found in the Canadian data which seemed to be related to Canadian tax policy was also found in the U.S. sample. It is possible that the economic events that lead to the change in the tax policy are actually to blame for the results found and not the policies. This omitted variable may also help explain results of Berges and Tinic who each found evidence of the January effect in Canada in the absence of capital gains tax. Evidence from this study and prior literature appear to point to an omitted variable that is decreasing over time. However, it must be pointed out that there are a number of limitations of this study which could have factored in the results. As noted previously, the measure of Canadianlisted only and cross-listed was not exact. Therefore, the expected results may have been blurred and biased against finding the desired outcome. Also even if a perfect measure of cross-listed firms was attained this does not preclude U.S. ownership of the Canadian-listed only firms. Therefore, the U.S. policy could affect the results of these firms. 17

19 The measure of return for the Canadian data was also crude as the data used the month of January return and not just the first five days. Although this data had been used in other studies, it likely reduced the chances of finding clear results. The division of regimes could be questioned. Although division based on inclusion rate appears obvious, these could be skewed when combined with changing marginal tax rates over the period. Redefining regimes based on top marginal tax rates may be an alternative that could be looked at in future research. On this same stream, the Poterba and Weisbenner study may be improved by considering one of the many other facets of U.S. capital gains tax policy such as rates. Although the results appear to negate evidence of the effects of tax policy in both the Canadian and U.S. setting, evidence of the anomaly are present. Large significant coefficients on the LOSS variables relative to the GAIN variables appear to support the existence of turn-of-the-year returns. This can be viewed as evidence of a relation of firms with losses to abnormal turn-of-the-year returns in the absence of tax and therefore may point to other explanations such as the window dressing theory. It may be that the effect of institutional investors overshadows the effect of the tax policies that were investigated. A future study may look to investigate these same tax policy issues but use a control for institutional investing. This may lead to a clearer view of the effects of tax policy on turn-ofthe-year returns. In summary, these findings do not preclude the tax selling hypothesis from being a factor in the seasonality of stock prices. However, they do add to the growing number of studies that are finding support for a theory which implies a combination of factors contributing to this anomaly. 18

20 Figure 1 Comparison of Canadian and U.S. Tax Regimes U.S. Tax Regimes RI - Med RII - High RIII - Low RIII - Low RI - Low RII - Med RII - High Canadian Tax Regimes Figure 2- Canadian Tax Regimes Regime I (1963 to 1972) During this period Canada had no tax on capital gains. Regime II( ) In 1973 began to be taxed. However, only 50% of the gain had to be taken into income. In 1988 and 1989 there was a phase-in period of a 66 2/3% inclusion rate as rates increased to 75% in Therefore this regime reflects an inclusion rate of 50% to 66 2/3% Regime III(1990 to 2001) - This regime reflects the highest taxed at an inclusion rate of 75%. Rates were reduced twice in 2000 once in February to 66 2/3% and then in October to a 50% rate again. However, 2000 and 2001 have been included in this regime as loss carrybacks allowed taxpayers to reduce previous gains at the 75% inclusion rate. 19

21 Table 1 - Coefficients from regression analysis using U.S. tax regimes U.S. Regime I U.S. Regime II U.S. Regime III Coefficients P-Value Coefficients P-Value Coefficients P-Value U.S. Data Loss January-June Loss July-December Gain January-June Gain July-December Difference between β3 and β Canadian-listed only Loss January-June Loss July-December Gain January-June Gain July-December *Coefficients in bold are significant at a 10% level Table 2 - Coefficients from regression analysis using Canadian tax regimes Regime I Regime II Regime III Coefficients P- Value Coefficients P- Value Coefficients P- Value Canadian-listed only Loss January-December Gain January-December /Price Log Market Cap US Data Loss January-December Gain January-December /Price Regression with interaction terms -Total Canadian sample Loss January-December Gain January-December /Price Log Market Cap Canadian only (dummy) CO-Loss January-December CO- Gain January-December *Coefficients in bold are significant at a 10% level 20

22 Table 3- Interaction models for inclusion rate increases and decreases CDA US Loss January-December Gain January-December /Price Log Market Cap ** Increase (Dummy Variable) Increase-Loss January-December Increase- Gain January-December CDA U.S. Loss January-December Gain January-December /Price Log Market Cap ** Decrease (Dummy Variable) Decrease-Loss January-December Decrease- Gain January-December *Coefficients in bold are significant at a 10% level ** Log market was introduced as a control variable for the Canadian tests due to the size difference between cross-listed and Canadian-listed-only firms. It was not included in tests on U.S. data. 21

23 Bibliography Athanassakos, George The Scrutinized-firm Effect, Portfolio Rebalancing, Stock Return Seasonality and the Pervasiveness of the January Effect in Canada. Multinational Finance Journal. 6, 1-27 Berges, Angel, John J. McConnell, and Gary G. Schlarbaum The Turn-of-the-year in Canada. Journal of Finance 39, Bhabra, Harjeet J, Upinder S. Dhillon, and Gabriel G. Ramirez. 1999, A November Effect? Revisiting the Tax Loss Hypothesis. Financial Management; 28, 5-15 Chen, Honghui and Vijay Singal All Things Considered, Taxes Drive the January Effect. The Journal of Financial Research, 3, Gibson, Scott, Assem Safieddine and Sheridan Titman, Tax motivated trading and price pressure: An Analysis of Mutual Fund Holdings. Journal of Financial and Quantitative Analysis. 35, Gu, Anthony Yanxiang The Declining January Effect: Evidence from the U.S. Equity Markets. The Quarterly Review of Economics and Finance.43, Johnston, Ken, and Don R. Cox Market Index Returns, Macromedia Variables, and Tax Loss Selling. Journal of Economics and Finance. 26, Keim, Donald. 1983, Size related anomalies and stock return seasonality: Further empirical evidence Journal of Financial Economics Lakonishok, Josef, and Seymour Smidt, 1986, Volume for winners and losers: Taxation and other motives for stock trading, Journal of Finance 41, Ligon, J A Simultaneous Test of Competing Theories Regarding The January Effect. The Journal of Financial Research 20: Poterba, James M. and Scott J. Weisbenner Capital Gains Tax Rules, Tax-loss Trading, and Turn-of-the-year Returns, Journal of Finance 56, Reinganum, Marc R. and Alan C Shapiro Taxes and Stock Return Seasonality: Evidence from the London Stock Exchange. The Journal of Business. 60, Roll, R. 1983, Vas ist das? The turn-of-the-year effect and the return premia of small firms, Journal of Portfolio Management 9, Shackleford, Douglas A. and Terry Shevlin Empirical Tax Research in Accounting. Journal of Accounting and Economics. 21, Tinic, Seha M., Giovanni Barone-Adesi, and Richard R. West 1987 Seasonality in Canadian Stock Prices: A Test of the Tax loss selling Hypothesis, Journal of Financial and Quantitative Analysis, 22,