Do Earnings Explain the January Effect?

Transcription

1 Do Earnings Explain the January Effect? Hai Lu * Leventhal School of Accounting Marshall School of Business University of Southern California Los Angeles, CA hailu@marshall.usc.edu Qingzhong Ma Department of Finance and Business Economics Marshall School of Business University of Southern California Los Angeles. CA qingzhom@marshall.usc.edu (First draft: 11/21/03) (This version: 01/21/04) (Preliminary; circulate for comments) Abstract This paper presents evidence on the correlation between stock returns in January and the earnings information released in the month. The annual earnings announced in January are predominantly positive, and the stock returns in late January are abnormally higher than in the remainder of the year. Both time-series and crosssectional analysis shows a strong relation between stock returns and the earnings information released in January, particularly in the second half of the month. The results suggest that the earnings are likely to be one important driving force of the January Effect. * We are grateful to Randy Beatty, Tyrone Callahan, Darin Clay, Larry Harris, Chris Jones, John Matsusaka, Lior Menzly, Kevin Murphy, Oguzhan Ozbas, Mark Weinstein, and FBE research lunch participants at the University of Southern California for helpful comments. Lu acknowledges the financial support and fellowship from the Leventhal School of Accounting, the Marshall School of Business at USC, and the SEC and Financial Reporting Institute. Ma acknowledges the fellowship support from the Marshall School of Business at USC. All errors are our own.

2 Do Earnings Explain the January Effect? One of the most puzzling empirical findings reported in finance is the significantly higher stock returns in January. After Rozeff and Kinney (1976) document the large January returns, numerous studies confirm the finding. Keim (1983) shows that much of the high return in January is attributed to small firms, and about half of the size premium occurs in the first five trading days. In addition, this anomalous January return pattern exists not only in the U.S. stock market, but also in foreign stock markets. This well-documented anomaly relating return seasonality and firm size is dubbed the January Effect (for example, Haugen and Lakonishok, 1988). More puzzling, although investors could invest in the anomalies (Jones and Pomorski, 2003), the January Effect is persistent (Schwert, 2002). Numerous hypotheses have been developed to explain the January Effect. One of the well known is tax-loss-selling (Roll, 1983; Reinganum, 1983). Since investors sell loser stocks in December to lower tax on net capital gains, the past loser stocks experience abnormal selling pressures that are relieved in January, resulting in large January returns. Yet, tax does not explain the entire January Effect (Reinganum, 1983; Reinganum and Shapiro, 1987). Other explanations include the window dressing hypothesis (Lakonishok, Shleifer, Thaler, and Vishny, 1991), the information hypothesis (Rozeff and Kinney, 1976), and the market microstructure issue (Reinganum, 1990; Ball, Kothari, and Shanken, 1995). 1 Surprisingly, most of the aforementioned studies focus on explaining the return patterns in the very beginning of January, although late January also witnesses abnormally higher stock returns than in the remainder of the year. 1 A brief summary of the hypotheses can be found in the Appendix I. 1

3 The purpose of this paper is to seek the potential driving forces that explain the abnormally high returns in late January. Two empirical facts are documented. First, there is a striking pattern in earnings announcements in January. A predominant number of the annual earnings announced in January are positive, and the daily percentage of positive earnings announced follows a clear declining trend from the beginning of January to the end of March. Second, the average daily market returns in the third and fourth weeks of January are significantly higher than in the rest of the year. The coincidence between the abnormally positive earnings announced during January and the abnormally high stock returns during the same period inspires us to put forth an earnings hypothesis, stating that at least part of the anomalous returns in January may be explained by the earnings released in January. The empirical evidence is consistent with the hypothesis. Within the sample period from 1972 to 2002, the market returns in January, particularly in the second half-month, appear to be partially explained by the earnings information released in the month. The earnings information also explains the January returns in cross-section. Specifically, we classify all of the stocks covered in the Center for Research in Security Prices (CRSP) at the University of Chicago into five portfolios based on their association with earnings. Firms announcing earnings with positive surprises are put in portfolio 1 and those with negative surprises in portfolio 5. Those firms that do not announce earnings during the month are placed in portfolios 2, 3, or 4 based on whether their industry peers, on average, announce earnings with positive, neutral, or negative surprises. The returns decrease monotonically from portfolio 1 to portfolio 5. Furthermore, the returns to large firms in late January are significantly higher than in any other period of the year, and there is weak evidence that earnings information mitigates the size effect in late January. Overall, the earnings hypothesis seems orthogonal to the stock returns in the first half of January but appears 2

4 to explain the abnormally high returns in the second half of the month. The results suggest that there are possibly two distinct forces driving the higher stock returns in January one related to the first half of the month and the other to the earnings announced in the second half. The remainder of this paper proceeds as follows. Section I describes the data, and Section II analyzes the patterns of the earnings announcements and stock returns in January. In section III, we establish the relationship between the earnings and returns through both time-series and cross-sectional analysis. In Section IV, the link between earnings and size premium is explored. Section VI presents our conclusions. I. Data Two data sources are used in this study. The earnings information from January 1972 and December 2002 is drawn from Standard and Poor s COMPUSTAT datasets. 2 All firms in Industrial Quarterly, Full Coverage Quarterly, and Research files with valid earnings announcement dates are included in the sample. The data representing the whole stock market are from the CRSP. We do not exclude any particular securities in the return analysis, but if a stock s prior month market capitalization is not available from CRSP it is excluded from the analysis related to firm size. II. Patterns of Earnings Announcements and Stock Returns A. Patterns of Earnings Announcements This subsection presents the patterns of earnings announcements in January. First, a large percentage of firms end their fiscal years in December and start their annual earnings announcements in January; second, the majority of the earning announcements in January are 3

5 positive, and the daily percentage of positive earnings follows a clear declining trend from the beginning of January to the end of March. There are a total 583,767 earnings announcements reported in COMPUSTAT from January 1972 to December These consist of 438,279 quarterly and 145,488 annual announcements. 3 Table I shows the distribution of the quarterly and annual earnings announcements as a percentage of total earnings announced in the year for each of the twelve calendar months. The monthly percentage of the announcements is calculated first for each year and then averaged over the 31 years. The table shows that 60% of annual announcements cluster in the first three months. The quarterly announcements exhibit a different pattern most of the announcements take place in April and May, July and August, and October and November. The two patterns are consistent with the fact that a majority of the firms have fiscal year-end in December. Unreported analysis shows that about 60% of the firms have fiscal year-end in December. The fiscal year-end for the remaining firms spreads over the other eleven months. As many as 5% of all firms have fiscal year-end in June, and as few as 1% have fiscal year-end in July. More importantly, these 60% firms with fiscal year-end in December account for about 80% of the market capitalization of all firms that have earnings announcement dates recorded in COMPUSTAT. Market capitalization is defined as the market value of the firms at the prior year-end. Since the firms covered in COMPUSTAT represent a large percentage of the market capitalization of all CRSP firms (nearly 79%), the firms with fiscal year-end in December actually account for about 62% of total market value. Furthermore, 72% of the firms in Standard and Poor s Industrial Index end their fiscal years in December. The discussion above suggests 2 In COMPUSTAT, earnings information is available from

6 that the events associated with the firms having fiscal year-end in December may have an important impact on market returns around the turn of the year. [Insert Table I about here] We further analyze the annual earnings announcements by the firms with fiscal year-end in December and find that %, 43.36%, and 21.35% of the announcements come in January, February, and March, respectively. A more striking finding is that the daily percentage of positive earnings (i.e., profit) announcements follows a clear declining trend from January 2 to March 31. Panel A of Figure 1 shows the 31-year average daily percentage of positive earnings announced and the cumulative percentage of earnings announcements from January to March. The figure shows the declining trend of the percentage of positive earnings from the beginning of January to the end of March. The daily percentage of positive earnings over this period is about 95% during the first half of January, but it declines to nearly 50% by the end of March. The highest daily percentage of positive earnings is 98% on January 13, and the lowest is 53% on March 31. Another noticeable feature in Panel A of Figure 1 is that only 0.69% of the firms announce earnings in the first week of January. The cumulative percentage increases quickly. More than 50% of these firms have announced earnings by mid-february, and most of the earnings are positive. By the end of March, 92% of the firms with fiscal year-end in December have announced earnings. 4 In order to see the intensity of earnings announcements activities in different periods, Panel B of Figure 1 shows the daily percentage of earnings announcements for the first three months of the year. The daily percentage of earnings announcements first 3 The fourth-quarter earnings announcements are treated as annual earnings announcements, while announcements in the other three quarters are treated as quarterly earnings announcements. 4 The SEC requires that firms file their annual financial statements (10-K) within 90 days after the fiscal year ends, so March 31 is presumably the due date for filing the 10-Ks. 5

7 increases, and then decreases. The peak is on January 26, when 2.12% of the firms announce annual earnings. Apparently, the earnings announcements activity during the second half of January is more intensive than in other periods. Since most of the subsequent analysis is based on weekly subperiods, the related information from Figure 1 is summarized in Table II, where the information on the percentage of positive earnings and the percentage of earnings announcements in January and February is presented on a weekly basis. The percentage of positive earnings generally declines from one week to the next, except for the first week in January. Only 2.18% of the firms announce earnings during the first two weeks of January, whereas over 25% of the firms announce earnings during the third and fourth weeks. The evidence indicates that firms tend to announce good news early and to delay announcing bad news. [Insert Figure 1 about here] [Insert Table II about here] The finding that firms tend to announce their earnings more promptly when they have good news to report and tend to delay reporting when they have bad news is not new (e.g., Chambers and Penman, 1984; Kross and Schroeder, 1984; and Penman, 1984). This phenomenon occurs in both quarterly and annual earnings announcements. However, such a high concentration of positive earnings announcements in January is first documented here. 5 Keeping in mind the high January returns, one might ask whether this pattern of earnings announcements may affect the market returns in January. Furthermore, given the fact that the earnings released in February are still highly positive, one might wonder why there is not a February effect 5 In analyzing the earnings information in January, we exclude the quarterly earnings as well as the annual earnings announced by firms with fiscal year-end other than December. Quarterly earnings are mostly announced during the month after the fiscal quarter; only a small percentage of annual earnings are released by firms with fiscal year-end other than December, as shown in Table I. 6

8 documented in the literature. These puzzling questions lead us to examine closely the stock return patterns in January and February. B. Stock Returns in January and February The January Effect is also termed the turn-of-the-year effect because a large portion of the abnormal returns in January occurs during the first two weeks (Schwert, 2002). Nevertheless, the returns from the second half of January are not negligible compared with those in other periods of the year. In this section, the return patterns in January and February are carefully examined. Table III shows the average daily equal-weighted market returns in each of the first eight weeks of the year. The data are from the CRSP daily index file. The first, second, and third weeks are defined as the first, second, and third five-day trading windows, respectively, and the remaining days are defined as the fourth week. Panel A shows the summary statistics of the average daily market returns. Similar to the findings in previous studies, the mean daily return in the first week of January is the highest, at 0.60%. It gradually declines to 0.31% and 0.22% in the second and third weeks, but it increases to 0.28% in the fourth week. The average daily returns during each of the first two weeks in February are 0.19% and 0.15%, respectively. Comparatively, the average daily return from March to December is only 0.05%. Panel B shows the return differential the difference between the returns in each of the first eight weeks of the year and that during the remainder of the year and the t-statistics from testing the null hypothesis that the difference is zero. The first to fourth weeks account for 42%, 20%, 13%, and 25% of the total return differential in January, respectively. More than half (62%) of the return differential takes place during the first two weeks, but that from the second half of the month (38%) is not trivial. Interestingly, the return differential during the first two weeks of February is 7

9 also positive and significantly different from zero, while those in the second half of February are insignificant. The return differential in the third week of February is even negative. The average daily return in February overall is insignificantly different from the other ten months. In order to see the size effect on the market returns, we repeat the exercise in Table III but replace the equal-weighted market returns with the value-weighted ones. The results are shown in Table IV. Panel A shows the summary statistics of the average daily returns over the eight weeks. Interestingly, the mean average daily return is the highest in the last week of January. Panel B lists that this is the only period in which the value-weighted market return is significantly different from that in the other ten months. What is worth noting is the magnitude: the mean value-weighted daily market return in the last week of January is 0.21%, which is comparable to the same period equal-weighted counterpart 0.28%. Comparatively, the valueweighted market return in the first week of January does not exhibit any significant difference from that in the other ten months, confirming that the abnormally high stock return in the beginning of the year is really a small firm phenomenon. One might expect that the size effect in late January would not be as pronounced as that in the beginning of the month. More detailed exploration of this point is discussed in section IV. [Insert Table III about here] [Insert Table IV about here] In summary, the earnings announced in late January are abnormally favorable, and the market returns in late January are abnormally high. This coincidence motivates us to posit an earnings hypothesis as follows: the anomalous returns in late January may be partially explained by the earnings information. The next section provides both time-series and cross-sectional evidence testing the hypothesis. 8

10 III. Stock Returns and Earnings Information in January This section explores the relationship between stock returns and the earnings information released in January. While our focus is on the month of January, in some cases the results in February and even March are provided for comparison. In the first subsection, a time-series regression analysis shows that the market returns in January over the last 31 years are highly correlated with the earnings information. The second subsection presents evidence that earnings information explains stock returns in cross-section. Five portfolios are formed on the basis of the stocks association with earnings information. The earnings hypothesis predicts that the portfolios associated with favorable earnings earn higher returns than those associated with unfavorable earnings. A. Time-series Analysis This subsection presents regression evidence that higher market returns in January, especially in the second half-month are associated with more favorable earnings. Specifically, the following time-series regression model is estimated: R w,, m, y = a + b * OddsRatiom y + ε, (1) where R w m, y, is the average daily raw market return over week w in month m of year y, and OddsRatio, is the proxy for the earnings information released in month m of year y. m y We construct a proxy for the earnings information released during the month on the basis of stock price reactions to earnings announcements. For each of the earnings announcements, the cumulative abnormal return (CAR) for the announcing firm in excess of its previous year average daily return over the three-day window from one day before to one day after the announcement date is examined. If the CAR is positive (non-positive), the earnings 9

11 announcement is defined as one with positive (negative) surprise 6 ; for each month-year, the earnings surprise odds ratio, OddsRatio, m y, is the number of positive earnings surprises divided by the total number of earnings announcements during the month. The mean value of OddsRatio is 53.45%, 50.42%, and 50.88% with a standard deviation of 9.50%, 7.23%, and 7.52% for January, February, and March, respectively. Table V presents the regression results of the average daily market return in each week of the month on OddsRatio. Weeks 1, 2, and 3 refer to the first, second, and third five-trading-day window of the month, and week 4 refers to the remaining trading days of the month. The average daily return over a trading window is defined as the average daily equal-weighted market return over the window. All of the t-statistics are based on standard errors robust to heteroskedasticity. Although each of the regressions has only 31 data points, the results suggest that earnings odds ratio explains the aggregate stock returns in time-series. For the twelve regressions, all the coefficients on odds ratio are positive, and most of them are statistically significant; R 2 s range from 5% in the second week of January to 71% in the last week of January. A higher earnings odds ratio is significantly correlated with higher market returns in the last two weeks of January and all weeks in February and March. It is not surprising that the odds ratio does not appear to have strong explaining power over the returns in the first two weeks of January when only a very small number of earnings announcements occur. The evidence in Table V is consistent with the earnings hypothesis. [Insert Table V here] B. Cross-Sectional Analysis 6 The method of classifying earnings is a modified version of Penman (1987). 10

12 The previous subsection shows time-series evidence that the earnings surprise odds ratio is positively related to the market returns in January. This subsection examines whether earnings explain the stock returns in cross-section. Specifically, stocks with different associations with earnings surprises have different returns. The earnings hypothesis predicts that announcing firms with positive surprises earn higher returns than those announcing with negative surprises. For those firms that do not announce earnings during the month, we explore the information transfer effect. Non-announcing firms whose industry peers on average announce earnings with positive earnings surprises earn higher returns than non-announcing firms in industries associated with neutral earnings surprises, which in turn are higher than those associated with negative surprises. An industry (sharing the 4-digit Standard Industry Code) is defined as associated with positive, neutral, and negative earnings surprises on average if the earnings surprise odds ratio for the industry in the month is greater than 55%, between 55% and 45%, and below 45%, respectively. 7 The industry odds ratio is defined as the number of announcing firms in the industry with positive surprises, divided by the total number of announcing firms in the industry. If an industry has no earnings announcement at all, it is classified as one with neutral surprises. For every month-year, five portfolios are constructed from all of the stocks covered in the CRSP on the basis of their association with earnings surprises. Specifically, each stock is classified into one of the following five categories: announcing, positive, non-announcing, positive industry, non-announcing, neutral industry, non-announcing, negative industry, and announcing, negative, where announcing, positive means the firm itself announces earnings with positive surprise and non-announcing, positive industry means the firm itself 7 The cutoff numbers are essentially arbitrary, but not critical. The results are robust when we use or as cutoff points. An industry with no announcing firms is treated as associated with neutral earnings surprises. Using 2- digit SIC codes to define an industry yields the same main results. 11

13 does not announce but its industry is associated with positive earnings surprises. The rest of the terms are defined accordingly. Such a classification is based on the prior findings that the earnings announcement of one firm may contain the information about the value of another firm. This information transfer effect is especially salient when the two firms are in the same industry (Foster, 1981; Baginski, 1987; Schipper, 1990). Table VI exhibits the average daily returns of the five portfolios in each of the six halfmonths. 8 Panels A, B, and C show their returns in January, February, and March, respectively. The portfolios are rebalanced for each month, and the reported returns are the averages over the 31 years from 1972 to Several patterns emerge from Table VI. First, earnings information has little impact on the average daily returns in the first half of January, when all the portfolios returns are positive, ranging from 0.21% to 0.37%, and significantly different from zero. In general, the earnings information does not appear to explain the cross-sectional variation for the portfolio returns in the first week of January. Second, the daily return pattern for the five portfolios in the second half of January is consistent with the earnings hypothesis. Portfolio 1 has the highest positive daily return of 0.43% while portfolio 5 has the lowest and negative daily return of 0.14%. The magnitude of the returns decreases from portfolio 1 to 5. The differences in returns between portfolio 1 and 5, 1 and 4, and 2 and 5 are significant at 1% level. The difference between returns to portfolios 2 and 4 is marginally significant (p-value is 0.17). Surprisingly, the return for portfolio 4, whose member stocks are associated with negative earnings surprises, is also positive and significant. Third, in each of the four half-months in February and March, the returns 12

14 decrease monotonically from portfolio 1 to 5, a pattern that is consistent with the earnings hypothesis. Overall, the results in Table VI demonstrate that the earnings surprise odds ratio explains well the stock returns in cross-section during the second half of January and the four half-months of February and March, but appears orthogonal to the returns in early January. In summary, both the regression analysis in time-series and the portfolio analysis in cross-section reveal a strong positive relationship between the earnings information and the stock returns in the second half of January, supporting the earnings hypothesis. [Insert Table VI about here] IV. Earnings Information and Size Premium The January Effect is highly interrelated with the size effect (Keim, 1983; Reinganum, 1983). This section presents some exploratory evidence regarding the link between the earnings information and the size premium in January. All stocks in the CRSP are first formed into deciles on the basis of prior month-end market capitalization. Stocks in deciles one to three, four to seven, and eight to ten are then put into three size groups Small, Medium, and Large. The size premium is defined as the return to the Small group minus that of the Large group. Table VII lists the daily average returns for the three size groups. Panel A reconfirms that small firms perform exceptionally well in the first week of January compared with the other two size groups. The daily return to the Small group on average is 0.86% during the first week. It then decreases to 0.35% in the second week and is flat at 0.24% and 0.23% during the last two weeks, respectively. On the other hand, the returns to the Large group exhibit a V shape in 8 When calculating average daily returns for a stock over a time period, we first calculate the buy-and-hold return for the whole period and then average out over the days within the period. This is done to avoid the potential upward 13

15 the four weeks of January. In the first two weeks, the Large group earns average daily returns of 0.12% and 0.11%, respectively. The number decreases to nearly zero in the third week and rises to 0.14% in the last week. While the returns to the Large group are all insignificantly different from zero in the first three weeks, the return in the last week is 0.14%, significant at the 5% confidence level. Overall, the size premium decreases monotonically from the first to the last week in January. In the last week, the size premium is only marginally significant. No apparent size premium is found in February. To explore the relationship between earnings information and size premium in January, the following time-series regression analysis is estimated: SP w, m, y = a + b * OddsRatiom, y + ε w, m, y (2) where SP w, m, y is the size premium for the w th week in month m of year y, and OddsRatio m, y is defined the same as in regression model (1). Table VIII provides the regression results for each of the four weeks in January (Panel A) and February (Panel B). Each row represents a regression. All of the t-statistics are based on standard errors robust to heteroskedasticity. Two patterns are noteworthy. First, in the first week of January, the size premium increases with earnings surprise odds ratio. One possible explanation is that higher earnings surprise in January may be associated with lower returns in prior year, triggering higher level of tax-loss-selling in January and increasing the size premium in the beginning of the month. Second, the earnings odds ratio appears to have a weak negative effect on size premium during the last two weeks of January, suggesting that earnings information mitigates the size effect in late January. bias caused by bid/ask spread (Blume and Stambaugh, 1983). 14

16 [Insert Table VII about here] [Insert Table VIII about here] V. Conclusions This paper documents two empirical findings the predominantly favorable earnings information released in January and the abnormally high stock returns in January, and proposes an earnings hypothesis: the predominant positive earnings announced in January contribute to the abnormally high returns in the second half-month of January. Evidence from both time-series and cross-sectional analysis is consistent with the earnings hypothesis. While the January Effect is highly interrelated with the size effect, the results in this paper shed new light on the differences between them. In particular, weak evidence is found that the size premium in the second half-month of January decreases with the earnings surprise odds ratio in the month. The January Effect is persistent (Schwert, 2002). The results in this paper indicate that if the January Effect is in part driven by the earnings news about the fundamental value of the firms, one would not expect the effect to disappear unless the practice of disclosure changes. The implications of this study may reach beyond providing one more hypothesis explaining the January Effect. Penman (1987) discovers the coincidence of the arrival of good quarterly earnings news and the high returns in the first half of the month following the fiscal quarter end and the coincidence of firms tendency to announce bad news on Mondays and the lower returns on Mondays. The results in this paper suggest the close relationship between annual earnings and the January Effect. Taken together, these findings raise the possibility that the stock return seasonalities may be attributed to the fundamental news events earnings announcements. 15

17 REFERENCES Baginski, Stephen, 1987, Intraindustry information transfers associated with management forecasts of earnings 25, Journal of Accounting Research, Ball, Ray, S. P. Kothari, and Jay Shanken, 1995, Problems in measuring portfolio performance: An application to contrarian investment strategies, Journal of Financial Economics 38, Barry, Christopher B., and Stephen Brown, 1984, Differential information and the small firm effect, Journal of Financial Economics 13, Blume, Marshall E., and Robert F. Stambaugh, 1983, Biased in computed returns, Journal of Financial Economics 12, Chambers, Anne E. and Stephen H. Penman, 1984, Timeliness of reporting and the stock price reaction to earnings announcements, Journal of Accounting Research 22, Foster, George, Intra-industry information transfer associated with earnings releases, Journal of Accounting Economics 3, Haugen, Robert A., and Josef Lakonishok, 1988, The incredible January Effect: The stock market s unsolved mystery, Dow-Jones-Irwin, Homewood, IL. Jones, Christopher S. and Pomorski, Lukasz, "Investing in Disappearing Anomalies" (November 2002). Keim, Donald, Size-related anomalies and stock return seasonality, Journal of Financial Economics 12, Kothari, S. P, 2001, Capital markets research in accounting, Journal of Accounting and Economics 31,

18 Kross, William and Douglas A. Schroeder, 1984, An empirical investigation of the effect of quarterly earnings announcement timing on stock return, Journal of Accounting Research 22, pp Lakonishok, Josef, Andrei Shleifer, Richard Thaler, and Robert Vishny, 1991, Window dressing by pension fund managers, American Economic Reviews, Papers and Proceedings 81, Penman, Stephen, 1987, The distribution of earnings news over time and seasonalities in aggregate stock returns, Journal of Financial Economics 18, Reinganum, Marc R., 1983, The anomalous stock market behavior of small firms in January, Journal of Financial Economics 12, Reinganum, Marc R., 1990, Market microstructure and asset pricing: An empirical investigation of NYSE and NASDAQ securities, Journal of Financial Economics 28, Reinganum, Marc R., Alan C. Shapiro, 1987, Taxes and stock return seasonality: evidence from the London Stock Exchange, Journal of Business 60, Roll, Richard, 1983, Vas ist das? The turn-of-the-year effect and the return premia of small firms, Journal of Portfolio Management 9, Rozeff, Michael S. and William R. Kinney, Jr., 1976, Capital market seasonality: The case of stock returns, Journal of Financial Economics 3, Schipper, Katherine, 1990, Commentary: Information Transfers, Accounting Horizon, December, Schwert, William G., 2002, Anomalies and market efficiency, NBER working paper series

19 Appendix: Summary of Some Earlier Studies on the January Effect Rozeff and Kinney (1976) find the seasonality of the monthly returns on NYSE firms from 1901 to The mean return in January is higher than in any other months. Keim (1983) shows that monthly return and size are always negatively correlated for the twelve months of the year but that the size premium is most pronounced in January. Furthermore, he finds that about half of the annual size effect can be attributed to January and that about half of the January Effect occurs during the first few trading days of the month. Reinganum (1983) interprets the early-january abnormal returns as consistent with the tax-loss-selling hypothesis while admitting that tax cannot fully explain the January Effect. Roll (1983) also partially attributes the January return seasonality to tax-loss-selling pressure at the end of December. Blume and Stambaugh (1983) suggest that the estimates of the size effect based on daily returns are upward biased. Based on buy-and-hold returns, they find that the full-year size effect is half as large as previously reported. Lakonishok, Shleifer, Thaler, and Vishny (1991) propose a window-dressing hypothesis. Institutional investors are evaluated based on their performance and the riskiness of their portfolios, so they will buy the risky small stocks at the beginning of the year and sell before the end of the year. The predictions of the window-dressing hypothesis are similar to those of the tax-loss-selling hypothesis. Market microstructure and transaction costs issues have also been considered in the literature related to size effect (e.g., Reinganum, 1990; Ball, Kothari, and Shaken, 1995). Schwert (2002) discusses various anomalies and concludes that the small-firm, turn-ofthe-year effect has become weaker but still exists. 18

20 Table I Distribution of Earnings Announced in Twelve Months ( ) This table lists the distribution of annual and quarterly earnings announced in each of the twelve calendar months. Each of the numbers is the 31-year average of the percentage calculated each year. Month Annual (%) Quarterly (%) January February March April May June July August September October November December Total

21 Table II Distribution of Earnings Announcements During the First Eight Weeks of the Year ( ) This table lists the distribution of the earnings announcements in the first eight weeks of the year for the firms with fiscal year-end in December. The third column shows the average percentage of positive earnings announced in each of the weeks. The percentage of positive earnings announcements is defined as the ratio of the number of positive earnings announced to the total number of earnings announcements. The fourth column lists the percentage of earnings announcements in each of the weeks. The percentage of earnings announced is the total number of earnings announcements in the week normalized by the total number of firms with fiscal year-end in December. The fifth column shows the cumulative percentage of earnings announcements. Weeks 1, 2, and 3 are defined as the first, second, and third seven-day calendar day period of the month, respectively, and week 4 refers to the remaining days of the month. Month Week Percentage of Positive Earnings Announcements Percentage of Earnings Announcements Cumulative Percentage of Earnings Announcements January February