The Effect of Earnings Surprises on Stock Returns:
Analysis of the Canadian Market
Bachelor Thesis BSc Economics and Business Economics Specialisation in Finance By Ronaldo Siemon Sprenger - 11850256 Supervisor: Dr. Sjoerd van den Hauwe AbstractThis thesis examines the effect of earnings surprises on stock returns in Canada. The stock returns have been monitored in a pre-event, event, and post-event window. During the event window, a positive relation has been found between the earnings surprise and the abnormal stock return. While most of the abnormal stock return following an earnings announcement happens on the day of, and the day after the announcement, evidence suggests a post-earnings-announcement drift for negative surprises is observable for 60 days after the announcement. The size of the forecast error is indicative of the magnitude of the abnormal stock return. Furthermore, an asymmetric effect dependent on the earnings surprise being positive or negative has not been found for stocks listed on the Toronto Stock Exchange (TSX).
Statement of Originality This document is written by Ronaldo Siemon Sprenger, who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Table of Contents
1. Introduction ... 4 2. Related Literature ... 6 2.1 Efficient Market Hypothesis ... 6 2.2 Earnings Surprises and their Effects ... 6 2.3 Asymmetric Effects ... 8 3. Hypotheses ... 9 4. Data Description ... 10 5. Methodology ... 13 5.1 Research Method ... 13 5.2 Statistical Models ... 15 5.2.1 Hypothesis 1 - Regression ... 15 5.2.2 Hypothesis 2 - Regression ... 15 5.2.3 Hypothesis 3 – T-test ... 16 6. Empirical Results and Analysis ... 17 6.1 Relation between Earnings Surprises and Abnormal Returns ... 17 6.2 Positive and Negative Earnings Surprises ... 18 6.3 Asymmetric effects ... 20 7. Conclusion ... 22 Bibliography ... 24 Appendix ... 251. Introduction
Quarterly earnings announcements are a primary indicator of a firm’s performance. Security analysts estimate the future quarterly earnings of a firm. These metrics are valuable as (institutional) investors rely on analyst’s forecast to price a security accurately. When a firm announces actual quarterly earnings that differ from the analyst’s consensus, the price of a security tends to change. Ball and Brown (1968) were among the first to find evidence of a post-earnings-announcement drift. This drift indicates the tendency of positive earnings surprises to result in positive abnormal returns and negative earnings surprises, leading to negative abnormal returns. These results suggest the underreaction of (institutional) investors to the earnings announcements, facilitating the continuation of abnormal returns.
Several studies have researched the effect of earnings surprises on stock returns and have found significant results proving the existence of post-earnings-announcements drifts. Prior research focused primarily on U.S. markets. This thesis will test whether the effect of earnings surprises on stock returns is also present in the Canadian market using data from firms in the TSX. Furthermore, this research will be expanded by testing whether an asymmetric effect exists depending on the surprise being positive or negative. Therefore, the following research question will be resolved in this thesis: To what extent do earnings surprises affect abnormal returns in Canada, and are these returns asymmetrically affected dependent on the surprise being positive or negative? The research performed in this thesis will cover the abovementioned question in three parts. First, the extent of the earnings surprise effect on abnormal returns is researched by performing an event study. This study will be done by gathering the quarterly earnings surprises and the (quarterly) abnormal returns on security prices after the earnings announcement from firms within the TSX. A regression will be performed to measure the effect of earnings surprises on abnormal returns. Second, the existence of a post-earnings-announcement drift is analysed by performing a regression analysis over 60 days following an earnings announcement. Third, the existence of an asymmetric effect that is dependent on the surprise being positive or negative will be determined. The
The results of the event study are expected to confirm that earnings surprises cause abnormal returns following an earnings announcement. Furthermore, the results of the regression analysis are expected to provide evidence of a post-earnings-announcement drift, as several prior studies have provided evidence of such an effect Bernard and Thomas (1989). Moreover, no evidence of an asymmetric effect dependent on whether the earnings surprises are positive or negative is expected to be found (Skinner & Sloan, 2002). The results of the performed research contribute to the literature by providing further evidence of the existence of post-earnings-announcement drifts outside of U.S. markets and offer additional insight into the earnings surprise effect in Canada. Additionally, the existence of an asymmetric effect has not been researched in the Canadian market thus far; this thesis will provide such research. This thesis is arranged in several sections, the introduction above being the first. In the second section, relevant literature and earlier findings related to this research will be presented. Hypotheses, based on the relevant literature, will be presented in the third section. In the fourth section, the data used for the event study and regression analysis will be presented. Thereafter, the methodology used for this research and the statistical models will be described in section five. In the sixth section, the empirical results of this research will be displayed and interpreted. In the final section, conclusions will be drawn from the results, the research question will be resolved, and ideas for further research will be suggested.
2. Related Literature
This section will explore the literature and its statistical findings related to the research topic in this thesis. First, the efficient market hypothesis and its relation to earnings surprises will be presented. Second, literature that has previously established the evidence of earnings surprises and its effect will be discussed. Last, prior research of asymmetric effects to earnings surprises will be examined. 2.1 Efficient Market Hypothesis The efficient market hypothesis suggests that capital markets reflect all publicly available information in stock prices. Malkiel (2003) relates the efficient market hypothesis to the idea of a “random walk”. The concept of a random walk is based on the idea that as soon as new information appears, it is immediately reflected in a revised stock price. As a result, today’s price changes are reflected only by today’s news, and tomorrow’s price change will be unaffected by the news of today. The existence of a post-earnings-announcement drift challenges this idea, as this would indicate that tomorrow’s price change is affected by the news of today.
2.2 Earnings Surprises and their Effects
To determine whether post-earnings-announcement drifts exist, it first needs to be established whether post-earnings-announcements indeed contain new information that is not priced into the security prices. Beaver, Bates & Davidson (1968) provide such proof. In their paper, they find significant movements in price and volume as a result of reported earnings. These results suggest that investors beliefs about the value of a security are, at least in part, influenced by the content of the reported earnings.
Ball and Brown (1968) were one of the first to find evidence of a post-earnings-announcement drift. The results of their research demonstrate that when actual income is higher than expected income, the security prices move in that same direction and vice versa. Furthermore, Ball and Brown (1968) discovered that new information is not immediately reflected in the new security price. Their research found that when new information became available to the public, security prices continued to drift upwards or downwards depending on the nature of the information, for approximately one month.
Therefore, this research provides evidence for the underreaction to earnings announcements.
The underreaction to earnings surprises may indicate a failure of analysts to take the implications of current earnings into account when predicting future earnings. Bernard and Thomas (1990) test the hypothesis that prices fail to reflect the implications of current earnings in future earnings. Their research studied the three-day price reactions for four consecutive quarterly earnings announcements following an earnings surprise in quarter t. Following the earnings announcement in quarter t, the results show an underreaction during the earnings announcements for quarters t+1 through t+3, and a minor overreaction for quarter t+4. These results are consistent with the idea of a seasonal random walk of earnings expectations. Therefore, the hypothesis tested by Bernard and Thomas (1990) holds, the market expects future earnings to be equivalent to the earnings of the prior year.
In line with Bernard and Thomas (1990), Abarbanell and Bernard (1992) provide further evidence of analysts’ forecasts following a naïve earnings expectation or seasonal random walk. Their results show a positive correlation for the first three quarters, i.e. lags declining in magnitude following quarter t. Contrary to the seasonal random walk model, no negative correlation was found for quarter t+4. Furthermore, Abarbanell and Bernard (1992) find that the main result found by Bernard and Thomas (1990) holds, even in a much smaller sample. Thus, providing further evidence of post-earnings-announcement abnormal returns as a result of naïve earnings expectations in subsequent earnings announcements.
The existence of a post-earnings-announcement drift may prove a market inefficiency. Though, if the market is genuinely inefficient, this inefficiency should be exploitable (Malkiel, 2003). Whether the post-earnings-announcement drift is exploitable is researched by Bernard and Thomas (1989). In their research, a portfolio is constructed with a long position in the 10% most positive earnings surprises and a short position in the 10% most negative earnings surprises. These positions are held for 60 days after the earnings announcement. Bernard and Thomas (1989) found cumulative abnormal returns of 6.31% over these 60 days, which is approximately 25% when annualised. The
results of their research suggest that there is a positive relation between the size of the earnings surprise and the size of the post-earnings-announcement drift. Therefore, the results provide evidence that this market inefficiency is indeed exploitable. Logically, when the unexpected earnings are more positive, it is to be expected that the security price has a more substantial positive change and vice versa. Bernard and Thomas (1990) show significant evidence that the magnitude of the abnormal returns after an earnings announcement is strongly correlated with the magnitude of its earnings surprise.
2.3 Asymmetric Effects
There is no rational expectation that implies positive expected returns as a result of positive earnings surprises would be of greater magnitude than if they were negative. Skinner and Sloan (2002) research whether an asymmetric effect can be found between positive and negative earnings surprises. In their research, they attempt to explain the underperformance of growth stocks as a result of an asymmetrically substantial negative stock return following a negative earnings surprise. Growth stocks report, on average, an equal amount of positive and negative earnings surprises. Which suggests that if the stock returns following these announcements were equal, there should be no over or underperformance of the stock. The results found by Skinner and Sloan (2002) show that following a negative earnings surprise, the highest growth stocks in the sample show a -7.32% stock return opposed to a 6.32% stock return after a positive earnings surprise. Therefore, the underperformance of growth stocks can be explained by asymmetrically substantial negative stock returns. Additionally, no evidence for asymmetric effects has been found for value and growth stocks combined.
All of the research covered so far have used samples of predominantly U.S. firms. Providing further evidence of post-earnings-announcements drifts in other markets would indicate that this effect is not merely an anomaly. Therefore, it can be valuable to research whether the same results can be found using a sample of Canadian firms. Furthermore, the research performed in this thesis will determine whether an
3. Hypotheses
In this section, hypotheses will be drawn from the related literature. These hypotheses will provide provisional answers to the research question that is solved in this thesis. The literature discussed is predominantly from U.S. samples. As the Canadian market has a similar financial system, the same results are expected to hold here. Hypothesis 1: The abnormal returns following positive earnings surprises are positive, and that of negative earnings surprises are negative. Prior studies have shown that a positive earnings surprise leads to positive abnormal returns and that negative earnings surprises lead to negative abnormal returns, which has been proven empirically by Bernard and Thomas (1989). Furthermore, the tendency of analysts’ forecast to underreact rather than overreact further provides the continuation of abnormal returns (Bernard & Thomas, 1990).
Hypothesis 2: A post-earnings-announcement drift is observable in the first 60 days following an earnings announcement.
The research of Ball and Brown (1968) found that when new information became available to the public, security prices continued to drift upward or downward, depending on the nature of the information. Skinner and Sloan (2002) find that most of these post-earnings-announcement drifts remain observable in the first 60 days. Hypothesis 3: There is no asymmetric effect in abnormal returns, dependent on whether the earnings surprise was positive or negative.
Skinner and Sloan (2002) show that there is no asymmetric effect present when combining value and growth stocks in the same sample. Assuming the TSX consists of similar composition of value and growth stocks, no evidence for an asymmetric effect will be found.
4. Data Description
To research the earnings surprise effect in the Canadian market, data from the Toronto Stock Exchange (TSX) is obtained. Data from the TSX is used as this is the benchmark index of Canada. The data required to perform the regressions consist of the index return, constituents, actual EPS, forecasted EPS, and stock returns. The data will comprise a relatively broad sample of 20 years; this will ensure that the results are not influenced by specific events, such as an economic boom or a recession. Therefore, the starting date chosen for the data is January 1st, 2000.
The index constituents of the TSX are obtained from Compustat. Compustat is an extensive database covering fundamental data from financial markets and is published by Standard and Poors. Over the 20-year period of January 1st 2000 to December 31st 2019, 879 unique firms have been, or still are, a part of the TSX. The data obtained includes ticker symbols and CUSIP code per firm, which will be used to collect data from other databases. FactSet is a provider of financial data and analytics. Using their database, the daily market return of the TSX is obtained. The data of the market return ranges from December 30th 1999 to April 9th 2020. An illustration of the market return is found below. Figure 2: Return of the TSX. To obtain the quarterly earnings surprises, data from the Institutional Broker Estimate System (IBES) database is used. The data is abstracted from this database based on ticker codes. Quarterly earnings data through the period of January 1st 2000 to February 29th 2020 of 879 securities are collected. A number of 11,309 observations are obtained, containing quarterly earnings data of 242 securities. Note that not all firms report
quarterly earnings data; therefore, some firms may be lost. There were 187 observations without data on quarterly earnings, which have been removed. The data consists of the earnings announcement date, the actual EPS and the forecasted EPS. Using the actual and forecasted EPS, the forecast error is calculated with the following formula. 𝐹𝑜𝑟𝑒𝑐𝑎𝑠𝑡 𝑒𝑟𝑟𝑜𝑟 (𝐹𝐸) =𝐴𝑐𝑡𝑢𝑎𝑙 𝐸𝑃𝑆 − 𝐹𝑜𝑟𝑒𝑐𝑎𝑠𝑡𝑒𝑑 𝐸𝑃𝑆 𝐹𝑜𝑟𝑒𝑐𝑎𝑠𝑡𝑒𝑑 𝐸𝑃𝑆 (1) With the earnings data obtained from the I/B/E/S database, certain variables from the research done by Skinner and Sloan (2002) are replicated. In their study, three variables for earnings surprises are created. These variables are labelled; SURPRISE, GOOD, and BAD. MacKinlay (1997) designated an earnings announcement as good news if the actual EPS exceeded the forecasted EPS by 2.5%, designated an earnings announcement as bad news if the actual EPS was 2.5%. The same 2.5% threshold is used in the research performed in this thesis. Therefore, the variable GOOD assumes the value of 1 if the forecast error exceeds 2.5%, and 0 in all other cases. BAD takes on the value of 1 if the forecast error is less than -2.5% and 0 in all other cases. SURPRISE assumes the value of 1 if the variable GOOD has the value of 1, -1 when BAD has the value of 1 and 0 in all other cases. Furthermore, the continuous variable labelled FE captures the sign and magnitude of the forecast error. To mitigate outlier problems, the sample will be winsorised using the actual EPS, forecasted EPS, and FE variable to exclude the top and bottom 1%.
Furthermore, the earnings data from the I/B/E/S database contains information regarding the magnitude of the earnings surprise. The standardised unanticipated earnings (SUE) score measures the number of standard deviations the mean forecasted EPS and the actual EPS. The SUE score is calculated with the following formula. 𝑆𝑈𝐸 𝑆𝑐𝑜𝑟𝑒 =𝐴𝑐𝑡𝑢𝑎𝑙 𝐸𝑃𝑆 − 𝐹𝑜𝑟𝑒𝑐𝑎𝑠𝑡𝑒𝑑 𝐸𝑃𝑆 𝑆𝑡𝑎𝑛𝑑𝑎𝑟𝑑 𝐷𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛 (2) Using the SUE score, deciles are created. These deciles are used to analyse the relation between forecast error size and abnormal returns.
To understand how securities, respond to the earnings surprises, stock returns are obtained from the Center for Research in Security Prices (CRSP). The stock returns can be obtained in the sampling intervals of per month and day. MacKinlay (1997) finds the smaller the sampling interval, the higher the statistical power. Therefore, daily stock data is used. The stock return is obtained based on the official ticker code. The data collected consists of 651,598 daily stock return observations, which span the period of January 1st, 2000 to April 13th, 2020. There are 291 unique securities in this sample, some of the 879 securities may be omitted as not all securities report daily earnings data. Using the daily market return of the TSX, the abnormal return is calculated. Observations for which the abnormal return could not be calculated are removed, leaving 621,697 observations. Merging the stock return data with the quarterly earnings surprise data results in 462,968 observations for 182 unique firms. Matches have been found for 7327 earnings announcements. The earnings surprises are matched with daily stock data starting 12 days before the earnings announcement until 60 days after the earnings announcement. The merged sample consists of 3731 positive earnings surprises, 2813 negative earnings surprises and 783 neutral earnings surprises. In the table below the summary statistics of the actual and forecasted EPS are presented. Table 1: Summary Statistics Actual and Forecasted EPS Actual EPS Forecast EPS Forecast Error Mean 0.331 0.35 0.068 Std. Dev. 0.676 0.576 2.00 Min. -2.7 -1.667 -9.75 Max. 2.73 2.658 10.398 Observations 462,968 462,968 462,968 The average actual EPS of the sample is approximately 0.33, while the average forecasted EPS is approximately 0.35. This means that on average, analysts overestimate earnings by 6.77%. Despite the suggestion that analysts tend to overestimate earnings, analysts are more frequently positively surprised rather than negatively in this sample.
5. Methodology
In this section, the research methodology that is used to answer the research question will be examined. Subsequently, the data needed to perform said research will be described. Thereafter, the research models will be formulated.
5.1 Research Method
The methodology used in this thesis is the event study methodology, as defined by MacKinlay (1997). In his research, he outlines the procedure for an event study which this thesis will carefully follow.
The first steps to conduct an event study are to define the event of interest and to identify the event window. The event of interest for this research is the quarterly earnings announcement. The event window covers the date of the quarterly earnings announcement. In practice, the event window is often expanded to at least one day before and after the earnings announcement, which is done for this research as well. After the event window is determined, the selection criteria for the firms in the sample must be defined. This thesis seeks to find evidence of an earnings surprise effect on stock returns in Canada. Therefore, firms are selected from the Canadian benchmark index, the TSX. When the foundations of the event study are there, the impact of the event needs to be assessed. In this thesis, the impact consists of the abnormal stock returns resulting from an earnings surprise. The abnormal return is determined by subtracting the normal expected return from the actual return following the event. The following formula can illustrate this.
𝐴𝑅= > = 𝑅= >− 𝐸(𝑅= >|𝑋>) (3)
Where 𝐴𝑅= > and 𝑅= > are the abnormal and actual return respectively for a given firm i and period t. 𝐸(𝑅= >|𝑋>) is the normal expected return, to determine its value, the normal return 𝑋> is needed. The normal return 𝑋> can be modelled in two approaches, the
constant mean return model, or the market model where 𝑋> is the market return. There are several benefits to the market model. MacKinlay (1997) found the market model to be more stable as the variance of the abnormal return is reduced relative to the constant
mean return model. Therefore, the market model is used in this thesis. The market model associates the return of a security to the return of the market portfolio. The following formula can illustrate the statistical model. 𝑅= >∗ = 𝛼=+ 𝛽=𝑅F > + 𝜀= > (4)
Here, 𝑅= >∗ is the normal expected return for firm i in period t, and 𝑅F > is the market return. The parameters of the market model are 𝛼=, 𝛽=, and 𝜀= >. As firms from the TSX are
selected, the index return of the TSX will be the proxy used for the market return. The formula measuring abnormal returns can be adjusted to the following equation. 𝐴𝑅= > = 𝑅= >− 𝑅= >∗ (5) The abnormal returns can be aggregated through time periods, the formula below defines the cumulative abnormal returns as the sum of abnormal returns from period t1 to t2. 𝐶𝐴𝑅= (𝑡K, 𝑡M) = N 𝐴𝑅= > >O >P>Q (6) The timeframe used to measure the abnormal returns includes the estimation window, the already defined event window, and the post-event window. The effect of an earnings announcement can already be seen before the announcement date; this can be the result of pre-announcements or information leakages. Skinner and Sloan (2002) find that most of the effect of such pre-announcements can be captured 12 trading days before the announcement date. Bernard and Thomas (1989) researched the longevity of the announcement drift. Their research shows that most of the post-earnings-announcement drift takes place during the first 60 trading days following an earnings announcement. Therefore, the estimation window encompasses [-12,-2], the event window includes [-1,1], and the post-event window consists of [2,60]. The timeline is illustrated below. -12 -1 1 Event window (earnings announcement) 60 Post-event window (post-earnings-announcement drift) Estimation window (pre-announcements)
5.2 Statistical Models
The hypotheses presented in this thesis will be tested using OLS regressions and a t-test. Regression models will be used to measure the effect of earnings announcements on abnormal stock returns. For these models, the distributional assumption is made. This means the stock returns are assumed to be normal, independently, and identically distributed. Furthermore, a t-test will be used to compare the means of abnormal stock returns.
5.2.1 Hypothesis 1 - Regression
To test the first hypothesis, the variable SURPRISE is regressed on AR. This will show the relation of the earnings surprise with the abnormal returns and its sensitivity. The regression for hypothesis 1 can be estimated in the following equation: 𝐴𝑅= > = 𝛼= + 𝛽K𝑆𝑈𝑅𝑃𝑅𝐼𝑆𝐸= > + 𝜀= > (7) Where 𝐴𝑅= > = Abnormal stock returns for a given firm i in period t; 𝑆𝑈𝑅𝑃𝑅𝐼𝑆𝐸= > = Sign of the earnings surprise for a given firm i in period t. The intercept 𝛼= provides the abnormal returns for earnings surprises not covered by the SURPRISE variable. The coefficient 𝛽K will represent the sign and sensitivity of the SURPRISE variable to AR. A positive relation between the two variables is expected, which would result in a positive coefficient 𝛽K. Furthermore, the error term 𝜀= > is assumed to have a population mean of zero.
5.2.2 Hypothesis 2 - Regression
To test the second hypothesis, the variables GOOD and BAD are regressed on AR. The inclusion of two separate surprise variables allows the regression to provide a differential effect between positive and negative earnings surprises. As a result, the effect of a positive and negative surprise and its magnitude on abnormal stock returns are presented. The regression for hypothesis 2 can be estimated in the following equation:
Where 𝐴𝑅= > = Abnormal stock returns for a given firm i in period t; 𝐺𝑂𝑂𝐷= > = Takes on the value of 1 when the forecast error exceeds 2.5% for a given firm i in period t, in all other cases 0; 𝐵𝐴𝐷= > = Takes on the value of 1 when forecast error is less than -2.5% for a given firm i in period t, in all other cases 0. The intercept 𝛼= provides the abnormal returns for earnings surprises that are covered by neither the GOOD nor BAD surprise variable. The coefficients 𝛽K and 𝛽M will represent
the sign and sensitivity of the GOOD and BAD variable to AR, respectively. As a positive relation between earnings surprises and abnormal stock returns is expected in hypothesis 1, the coefficient 𝛽K for good surprises is expected to be positive, while the coefficient 𝛽M for bad surprises is expected to be negative. Furthermore, the error term 𝜀= > is assumed to have a population mean of zero. 5.2.3 Hypothesis 3 – T-test An independent group t-test is used to test hypothesis 3. Through this t-test, the mean abnormal returns of positive and negative earnings surprises are compared. The t-test is two-tailed, and the group is based on the SURPRISE variable. The hypotheses of this t-test can be rewritten as: 𝐻Z: 𝜇]K= 𝜇K
6. Empirical Results and Analysis
In this section, the empirical results are presented and analysed. Based on the results, the three hypotheses can be accepted or rejected. The first hypothesis is tested by analysing the abnormal stock return and its magnitude after an earnings surprise. Where after the second hypothesis is tested by analysing the effect of positive and negative earnings during different timeframes. Lastly, hypothesis 3 is tested by comparing the means of abnormal returns. 6.1 Relation between Earnings Surprises and Abnormal Returns To test the first hypothesis, the abnormal return following an earnings announcement is analysed. This is done by regressing the surprise variable to the abnormal stock returns on the day of the earnings announcement. Table 2: Regression abnormal returns on surprise variable during the event window Abnormal Return Surprise 0.404%*** (14.52) Constant -0.058%* (-2.25) Observations 26,588 t statistic in parentheses * p<0.05, ** p<0.01, *** p<0.001 Based on this regression, a positive and significant coefficient of approximately 0.004 is found. This means, during the event window, abnormal returns increase by 0.404% if the earnings surprise is positive, and decrease by 0.404 if the earnings surprise is negative. This is in line with the results observed by Bernard and Thomas (1989). Therefore, the results of this regression provide significant evidence for a positive relation between earnings surprises and abnormal returns.
Table 3: Regression abnormal returns on SUE score deciles during event window
SUE Score Abnormal Return t-statistic
1 -1.21% -10.7*** 2 -0.70% -6.28*** 3 -0.43% -3.79*** 4 -0.20% -1.8 5 -0.00% -0.02 6 0.27% 2.45* 7 0.36% 3.21** 8 0.62% 5.58*** 9 0.90% 8.05*** 10 1.19% 10.66*** * p<0.05, ** p<0.01, *** p<0.001
Furthermore, the table above provides additional evidence for a positive relation between earnings surprises and abnormal returns. The table presents the regression coefficients from the decile with the lowest SUE score, up until the decile with the highest SUE score. The results show that the lower the SUE score, the lower the abnormal returns are. Likewise, the higher the SUE score, the higher the abnormal returns. This is in line with research performed by Bernard and Thomas (1989), whose results show the same relation. Therefore, significant evidence suggests a positive relation between the size of the forecast error and abnormal returns. As a result, the first hypothesis is accepted. 6.2 Positive and Negative Earnings Surprises
In the table below, the abnormal returns are regressed using the good and bad news variable across different time windows. In the estimation window, a positive abnormal return of 0.025% and a negative abnormal return of -0.015% are found for good and bad news, respectively. Though these results are not significant. During the event window, a positive abnormal return of 0.405% and a negative abnormal return of -0.404% are found for good and bad news, respectively. Both of these results are highly significant at the 0.10% level. Furthermore, during the post-event window, an abnormal negative return of -0.01% and -0.043% are found for good and bad news, respectively. Only the coefficient
Table 4: Regression abnormal returns on good and bad news Abnormal Return Estimation Window Abnormal Return Event Window Abnormal Return Post-Event Window Good News 0.025% (0.97) 0.405%*** (6.71) -0.010% (-0.77) Bad News -0.015% (-0.51) -0.404%*** (-6.04) -0.043%** (-3.06) Constant 0.019% (0.86) -0.059% (-1.22) 0.057%*** (4.88) Observations 73,937 26,588 379,558 t statistic in parentheses * p<0.05, ** p<0.01, *** p<0.001 The significant results found during the event window present a clear positive relation between the sign of the earnings surprise and the abnormal return. Thus, providing further evidence for hypothesis 1. This is in line with the related literature. Furthermore, the significant negative abnormal return for the bad news variable during the post-event window provide significant evidence of a post-earnings announcement drift for negative earnings surprises. This indicates that a portfolio, which is constructed of short positions in stocks with negative earnings surprises, that are bought on the earnings announcement date and sold 60 days later, are able to achieve positive abnormal returns. Consequently, further evidence is provided that challenges the efficient market hypothesis.
The results found in this thesis stand in contrast to the post-earnings-announcement drifts observed by Bernard and Thomas (1989), who find significant evidence for both positive and negative post-earning-announcement drifts in the 60 days following an earnings announcement. As the results in this thesis provide such evidence only for negative surprises, the second hypothesis is partially accepted. The second hypothesis does not hold for positive surprises.
In the table below, the abnormal returns on good and bad news are regressed per day. Where day 0 is the announcement date. For positive earnings surprises, significant abnormal returns are found on day 0, day 1, and day 4. Likewise, for negative earnings surprises, significant results are found on day 0, day 1, and day 6. Thus, during the event window, significant results are found on the announcement date and the day thereafter. The results suggest the majority of abnormal returns are concentrated close to the announcement date, namely on the day of, and the day after the earnings announcement. The table for the full timeframe can be found in the Appendix.
Table 5: Daily abnormal returns for positive and negative earnings surprises
Positive Earnings Surprise Negative Earnings Surprise
Day Abnormal Return t-statistic Abnormal Return t-statistic
-1 0.04% 0.62 0.00% (-0.03) 0 0.57% 5.1*** -0.50% (-4.09)*** 1 0.59% 4.78*** -0.70% (-5.16)*** 2 0.01% 0.14 -0.04% (-0.31) 3 0.00% 0.03 -0.04% (-0.32) 4 0.27% 2.08* 0.20% 1.36 5 0.02% 0.29 -0.12% (-1.40) 6 -0.10% (-1.12) -0.24% (-2.51)* 7 0.05% 0.67 -0.08% (-1.11) 8 0.01% 0.14 0.08% 1.09 9 0.00% 0.04 -0.05% (-0.47) 10 -0.01% (-0.05) -0.09% (-0.70) * p<0.05, ** p<0.01, *** p<0.001 6.3 Asymmetric effects The abnormal returns as a result of good and bad news found in the regressions in section 6.2 are of a relatively similar magnitude. Though, a definitive answer as to whether an asymmetric effect exists is not provided. To determine whether the abnormal return of positive news significantly differs from the abnormal return of negative news, a t-test is used. For the purpose of this t-test, the sign of the abnormal returns for negative surprises
Table 6: t-test Comparison Positive and Negative Surprises on Abnormal Return
Observations Mean Standard Deviation
Positive Surprise 13,086 0.35% 0.042 Negative Surprise 9,613 0.46% 0.045 diff 0.116 Ha: diff ≠ 0 t-statistic = 1.997 p = 0.046 df = 22,697 In the table above, the mean of abnormal returns after a positive and negative surprise during the event window is compared. Subtracting the two means from each other results in a difference of 0.116. The p-value is 0.046. Using an alpha of 0.01, the difference is not significantly different from 0. This is in contrast to Skinner and Sloan (2002), who find significant evidence of an asymmetric effect in growth stocks. This can be explained, at least in part, by the selection of the constituents used in the sample. As the TSX consists of mature firms, it is not expected that the results found by Skinner and Sloan (2002) would hold in this sample. Therefore, this t-test provides significant evidence against the existence of an asymmetric effect depending on the earnings surprise being positive or negative. As a result, the third hypothesis is accepted.
7. Conclusion
This aim of this thesis is to identify the effect of earnings surprises on the Toronto Stock Exchange (TSX). Through testing three hypotheses, this research will determine to what extent earnings surprises affect stock returns, and whether an asymmetric effect is present in the Canadian stock market. The acceptation or rejection of these hypotheses provides evidence that will help answer the research question.
The sample used in this research contained 7327 earnings announcements of TSX constituents during the period of January 2000 to December 2019. The earnings announcements have been paired with associated stock returns of 12 days prior to the announcement up to 60 days after the announcement. The earnings announcements were separated into positive and negative earnings surprises based on the size of the forecast error. An earnings announcement is determined to be a positive surprise if the forecast error is more than 2.5% and is considered to be a negative surprise if the forecast error is less than -2.5%. These designations are in line with the event study performed by MacKinlay (1997). However, the assignment of the two different groups and its cut-off points may affect the results found in this thesis. The abnormal stock returns following either a positive and negative earnings surprise are analysed through an event study. The first hypothesis presented in this thesis suggests that positive earnings surprises lead to positive abnormal returns, and negative earnings surprises lead to negative abnormal returns. The results found in this research find significant evidence of a positive relation between forecast error and abnormal returns. Furthermore, a significant positive relation is found between the size of the forecast error and abnormal returns. This is in line with the results observed by Bernard and Thomas (1989). Therefore, the first hypothesis is accepted.
The second hypothesis posed in this thesis implies a post-earnings-announcement drift is observable for 60 days after an earnings announcement. Through a series of regressions, the effect of positive and negative earnings surprises in each time window is tested. Significant results have been found during the event window and the post-event window. The coefficients indicate a 0.405% positive abnormal return for positive
surprises during the event window. During the post-event window, a significant negative return of 0.043% has been found for negative earnings surprises. Therefore, some evidence is provided for a post-earnings-announcement drift following a negative earnings surprise. Thus, only partial evidence is found supporting the hypothesis.
The third hypothesis tests whether an asymmetric effect in abnormal returns is observable depending on the surprise being positive or negative. Through comparing the means by use of a t-test, the difference is measured and tested. The results show that the difference between the two means is not substantially different from 0. Therefore, there is no significant evidence of an asymmetric effect. Thus, the third hypothesis is accepted. While Skinner and Sloan (2002) find significant evidence of an asymmetric effect in growth stocks, no evidence is found in the research performed in this thesis. This can be explained by the selection of the constituents used in the sample, as the TSX consists of mature firms rather than high-growth firms.
The research performed in this thesis uses a different time period, different market, different selection criteria of firms and different determinants of what constitutes as a positive and negative surprise than prior research. Each of which may have a significant impact on the results that have been found. In particular, the selection criteria of firms may have had an impact. As only firms that are listed on the TSX are selected, a moderate fraction of the Canadian market is not captured, such as the small and medium-sized businesses. Furthermore, the assignment of positive and negative surprises based on forecast error size may have an impact on the results, as earnings surprises close to 0 are not accounted for. Using the evidence that is found in this thesis, the research question is answered. There is a positive relation found between the forecast error and the abnormal return in the TSX. The extent of this effect is correlated with the size of the forecast error. The results provide further evidence that challenges the efficient market hypothesis. Furthermore, no evidence has been found to indicate an asymmetric effect depending on the surprise being positive or negative. Future research could repeat this study in different markets and time periods to see if the same results hold elsewhere. Moreover, other variables could be controlled for, such as analyst coverage.
Bibliography
Abarbanell, J. S., & Bernard, V. L. (1992). Test of Analysts' Overreaction/Underreaction to Earnings Information as an Explanation for Anomalous Stock Price Behavior. The
Journal of Finance, 47, 1181-1207.
Ball, R., & Brown, P. (1968). An Empirical Evaluation of Accounting Income Numbers.
Journal of Accounting Research, 159-178. Beaver, W., Bates, R., & Davidson, H. (1968). The Information Content of Annual Earnings Announcements. Journal of Accounting Research, 6, 67-92. Bernard, V. L., & Thomas, J. K. (1989). Post-Earnings-Announcement Drift: Delayed Price Response or Risk Premium? Journal of Accounting Research, 27, 1-36. Bernard, V. L., & Thomas, J. K. (1990). Evidence that stock prices do not fully reflect the implications of current earnings for future earnings. Journal of Accounting and
Economics, 13, 305-340. MacKinlay, A. C. (1997). Event Studies in Economics and Finance. Journal of Economic Literature, 35, 13-39. Malkiel, B. G. (2003). The efficient market hypothesis and its critics. Journal of Economic Perspectives, 17, 59-82. Skinner, D. J., & Sloan, R. G. (2002). Earnings Surprises, Growth Expectations, and Stock Returns or Don't Let an Earnings Torpedo Sink Your Portfolio. Review of
Accounting Studies, 7, 289-312.
Appendix
Table 1: Abnormal Returns for Positive Earnings Surprises Positive Earnings Surprise
Day AR T-statistic Day AR T-statistic
-12 0.30% 3.37*** 26 -0.09% (-1.00) -11 -0.13% (-1.24) 27 -0.01% (-0.16) -10 0.00% 0.01 28 -0.07% (-1.04) -9 -0.26% (-2.89)** 29 0.04% 0.52 -8 0.09% 0.96 30 -0.18% (-1.85) -7 0.04% 0.62 31 -0.18% (-1.65) -6 0.15% 2.16* 32 -0.16% (-1.53) -5 0.04% 0.46 33 0.08% 1.07 -4 -0.05% (-0.45) 34 0.17% 2.33* -3 -0.06% (-0.58) 35 0.04% 0.54 -2 -0.01% (-0.14) 36 0.16% 2.1* -1 0.04% 0.62 37 -0.20% (-2.28)* 0 0.57% 5.1*** 38 -0.21% (-1.84) 1 0.59% 4.78*** 39 -0.09% (-0.80) 2 0.01% 0.14 40 -0.06% (-0.72) 3 0.00% 0.03 41 -0.06% (-0.82) 4 0.27% 2.08* 42 0.10% 1.37 5 0.02% 0.29 43 -0.08% (-1.30) 6 -0.10% (-1.12) 44 -0.15% (-1.60) 7 0.05% 0.67 45 -0.07% (-0.58) 8 0.01% 0.14 46 -0.08% (-0.85) 9 0.00% 0.04 47 0.04% 0.54 10 -0.01% (-0.05) 48 -0.01% (-0.12) 11 -0.22% (-2.30)* 49 0.09% 1.41 12 0.02% 0.31 50 -0.07% (-0.89) 13 -0.07% (-0.96) 51 0.06% 0.55 14 -0.02% (-0.31) 52 0.02% 0.18 15 -0.04% (-0.62) 53 -0.07% (-0.76) 16 -0.05% (-0.60) 54 0.00% 0.04 17 -0.20% (-1.90) 55 0.14% 1.99* 18 0.20% 1.89 56 -0.08% (-1.13) 19 -0.08% (-0.99) 57 0.03% 0.48 20 -0.04% (-0.58) 58 0.00% 0.04 21 -0.09% (-1.09) 59 0.01% 0.09 22 0.10% 1.38 60 0.00% (-0.02) 23 -0.12% (-1.30) * p<0.05, ** p<0.01, *** p<0.001 24 0.12% 1.01 25 -1.09% (-1.57)
Table 2: Abnormal Returns for Negative Earnings Surprises Negative Earnings Surprise
Day AR T-statistic Day AR T-statistic
-12 0.14% 1.44 26 -0.06% (-0.64) -11 -0.09% (-0.77) 27 0.01% 0.17 -10 -0.01% (-0.10) 28 -0.11% (-1.44) -9 -0.18% (-1.91) 29 0.06% 0.65 -8 -0.06% (-0.61) 30 -0.08% (-0.70) -7 -0.02% (-0.26) 31 -0.17% (-1.39) -6 0.07% 0.86 32 -0.11% (-0.90) -5 0.08% 0.76 33 0.09% 1.06 -4 -0.02% (-0.19) 34 0.12% 1.53 -3 0.06% 0.62 35 -0.05% (-0.77) -2 -0.10% (-0.97) 36 0.12% 1.45 -1 0.00% (-0.03) 37 0.02% 0.19 0 -0.50% (-4.09)*** 38 -0.14% (-0.94) 1 -0.70% (-5.16)*** 39 -0.19% (-1.59) 2 -0.04% (-0.31) 40 -0.03% (-0.36) 3 -0.04% (-0.32) 41 -0.15% (-1.91) 4 0.20% 1.36 42 0.01% 0.12 5 -0.12% (-1.40) 43 0.04% 0.56 6 -0.24% (-2.51)* 44 0.04% 0.42 7 -0.08% (-1.11) 45 0.18% 1.45 8 0.08% 1.09 46 -0.06% (-0.62) 9 -0.05% (-0.47) 47 0.06% 0.68 10 -0.09% (-0.70) 48 -0.01% (-0.15) 11 -0.33% (-3.38)*** 49 0.09% 1.2 12 -0.07% (-0.81) 50 -0.04% (-0.44) 13 -0.23% (-2.94)** 51 0.07% 0.62 14 0.02% 0.29 52 0.12% 0.99 15 0.00% (-0.06) 53 0.04% 0.42 16 0.13% 1.26 54 -0.08% (-0.88) 17 -0.10% (-0.89) 55 0.09% 1.19 18 -0.02% (-0.26) 56 -0.03% (-0.36) 19 -0.13% (-1.41) 57 0.11% 1.51 20 -0.08% (-1.06) 58 -0.04% (-0.40) 21 -0.13% (-1.55) 59 0.22% 1.6 22 0.14% 1.97* 60 0.01% 0.06 23 -0.10% (-0.98) * p<0.05, ** p<0.01, *** p<0.001 24 0.23% 1.91 25 -0.96% (-1.38)
