Stock recommendations in Swedish business magazines

Transcription

1 STOCKHOLM SCHOOL OF ECONOMICS Master Thesis in Finance Spring 2012 Stock recommendations in Swedish business magazines Announcement effect and long-term performance Henrik Rinaldo 1 Abstract From the three main Swedish business magazines, Veckans Affärer, Affärsvärlden and Aktiespararen, 354 buy and sell recommendations made during the years were investigated. The purpose of this thesis was twofold; to investigate whether or not there was an announcement effect present in the stock recommendations made in Swedish business magazines, and if there was any real substance to the recommendations made. With regards to the presence of an announcement effect an event study methodology was used. The results show that, when looking at all three magazines as a whole, there is an announcement effect. One possible reason why an effect is observed is that sources with recommendations about Swedish stocks are not in abundance, and that any news presented in the Swedish business magazines therefore is considered important. When limiting the sample to the individual magazines I find that they all have an announcement effect. Again, the reason here might be that since there are so few sources of information, that they all are important. In order to investigate long-term performance portfolios with a buy-and-hold strategy were formed. None of the three magazines show any long-term over-performance. The results are robust to changes in holding period and to being divided into buy-only or sell-only portfolios. To see if there was a Super star effect present in my sample, that is if one analyst clearly outperforms the rest of his peers, my sample was divided by analyst. During the period I am investigating I found no support for a superstar being present at any of the magazines. In sum, I find that recommendations made in the three main Swedish business magazines have a short-term effect on stock prices. When looking at the long-term, however, I find no support for any over-performance compared to the market. Keywords: Stock recommendations, buy-and-hold, event study Tutor: Assistant Professor Ramin Baghai Acknowledgements: I would like to thank Ramin Baghai for his valuable support and advice throughout the process of writing this thesis @student.hhs.se

2 TABLE OF CONTENTS 1. INTRODUCTION PREVIOUS RESEARCH PERFORMANCE OF STOCK RECOMMENDATIONS ANNOUNCEMENT EFFECTS LONG TERM PERFORMANCE SUPERSTARS OVERREACTIONS IN THE STOCK MARKET HYPOTHESES METHODOLOGY EVENT STUDY Event Definition Confounding Effects Hypotheses Testing LONG TERM PERFORMANCE DATA EMPIRICAL FINDINGS EVENT STUDY Veckans Affärer Affärsvärlden Aktiespararen LONG TERM Veckans Affärer Affärsvärlden Aktiespararen DISCUSSION EVENT STUDY Veckans Affärer Affärsvärlden Aktiespararen LONG TERM Veckans Affärer Affärsvärlden... 26

3 7.2.3 Aktiespararen ECONOMIC ISSUES ECONOMETRIC ISSUES Event Definition The Use of Daily Returns Confounding Effects The Bad Model Problem Sample Size and Standard Errors CONCLUSIONS REFERENCES APPENDIX... 35

4 1. INTRODUCTION In 2006 Lidén investigated the performance of stocks that had gotten a buy or a sell recommendation in one of the Swedish business magazines by looking at buy-andhold abnormal returns (BHARs). The sample he used stopped in 2000, which today is over ten years ago. By comparing his results to mine, which are more current, I will be able to see if the results he got are still valid today. One way in which my investigation is different from his is that he lumped all the analyses from journalists together in order to compare them to analyses from professional analysts. I have instead chosen to compare the different magazines with each other to see which magazine a subscriber should choose to follow when it comes to stock recommendations. Another way in which my thesis is different from the way Lidén (2006) did his paper is that he excludes all repeat recommendation that a magazine does from his sample. My intention is to see what kind of performance a magazine subscriber could expect if he follows all recommendations. That is, even if the same stock got a buy recommendation a year ago this would still prompt the reader to once again purchase that stock. Also, if more magazines recommend the same stock I would consider this to further indicate that the first recommendation was an accurate one. Finally, I have chosen to use the OMXS Index as a benchmark to compare the portfolio returns to. Lidén (2006) used an industry portfolio as a benchmark in an effort to control for an industry effect. The idea is that an analyst that only recommends stocks from an industry that happens to be performing well at the moment, like the IT sector before the IT-bubble for instance, has an advantage to an analyst that is confined to an underperforming industry. As a magazine subscriber, however, you would not care how the recommended stocks perform compared to other stocks of a similar type. You only care about how much money you make from your investment. A big part of choosing which stocks to invest in is to first choose the right industry, and a good advisor will most likely produce more recommendations skewed towards the currently booming industries. Following this introduction I will start with presenting previous research relevant for this thesis. After that, my hypotheses are introduced, followed by a section about the methodology. Next, the data will be described and the empirical findings will then be 1

5 presented. After that, the results of the hypothesis testing are evaluated, the economic and econometric issues described, and the credibility of my findings is discussed. As a final part, the conclusions derived from the thesis are presented. 2. PREVIOUS RESEARCH 2.1 PERFORMANCE OF STOCK RECOMMENDATIONS When studying the effects of stock recommendations it is important to separate the findings into two parts, the announcement effect and the long-term effects. The following two sections will now describe them in turn ANNOUNCEMENT EFFECTS An announcement effect is when abnormal returns are observed around the day of publication of a news item (Desai et al, 2000). When looking at stock recommendations, the announcement effect simply tells us if investors see the recommendation as new information to be traded on, or not. A buy recommendation would prompt a stock purchase, and a sell recommendation a sale or a short sale. Whether or not the stock is over or undervalued is not the issue here, and this is instead the subject of interest in the next section. Several prior studies have found significant announcement effects from stock recommendations. Liang (1999) investigated the performance of stock recommendations made in the Dartboard column of the Wall Street Journal. The results show a significant two-day announcement effect that is reversed within 15 days. Ferreira and Smith (2003) studied recommendations made by reporters in their study of the TV show Wall Street Week with Louis Rukeyser. They found statistically significant positive abnormal returns of 0.65 percent in the first trading day after the show. Yazici and Muradoğlu (2002) investigated stock recommendations published in the Investor Ali column of the journal Moneymatik. They find that the average abnormal return for the event date to be 2.35 percent, and that the abnormal return over the one-week period after the publication date is percent. The fact that these studies comes from different countries but still offer similar results 2

6 indicates that although conditions are different in different countries there are usually some sources of information that is the go to source of information in that country LONG TERM PERFORMANCE This section deals with the question if the recommendations made in business magazines actually contain information about over or undervalued stocks. Desai et al (2000) studied recommendations made by Wall Street Journal all-star analysts. Using a buy-and-hold strategy, they found that recommendations made by all-star analysts outperform benchmarks when controlling for size and industry. Analysts that were focused on covering only one industry outperformed those covering several industries. Several other studies do not support the claim that recommendations made in business magazines outperform the market. When investigating recommendations given in the Dartboard column of the Wall Street Journal, Liang (1999) finds that, on average, investors following the experts advice lose 3.8 percent over a six-month holding period. Yazici and Muradoğlu (2002), who investigated recommendations given in Moneymatik, found that the recommendations do not help small investors to earn excess returns. A Swedish study by Lidén (2006), that examined recommendations made in the Swedish business press, showed that there were no excess returns to be made by following all the recommendations. However, by only following the sell recommendations an investor could earn substantial buy-and-hold abnormal returns. 2.2 SUPERSTARS Superstars can be found in all areas, from business to sports. Malmendier and Tate (2009) investigate how a small number of superstar CEOs in the US enjoy most of the rewards in the form of compensation, status and press coverage. In their study they find that the CEOs that have achieved superstar status underperform both compared to their prior results and when compared with other CEOs. Hausman and Leonard (1997) studied the effect on TV-ratings for NBA games when superstars like Michael Jordan appeared and found that the ratings were substantially higher. This superstar phenomenon is an expression of a more general concept often called the Pareto 3

7 Principle after its originator Vilfredo Pareto ( ). What Pareto found was that in most areas, rewards and performance is severely skewed. For instance, most of the world s wealth belongs to a small number of people, and most of the peas produced in Pareto s garden came from only a few pots. When it comes to finance, it is well known that most of the returns are made by a small group of investors, and that usually only a few stocks in a portfolio are responsible for most of the gains (Koch, 1998). This is a concept fully understood by Warren Buffet, who says, Don t diversify. Put all your eggs in one basket and watch that basket closely. In the context of this thesis, the Pareto Principle would manifest itself by showing that only a few of the analysts clearly outperform the others. Desai et al (2000) shows how important it is to find your niche in order to become a superstar. They found that analysts who were focused on covering only one industry outperformed those covering several industries. As a reporter for a business magazine you are likely asked to cover several industries, which according to the reasoning regarding specialization, would make it harder to produce superior results. 2.3 OVERREACTIONS IN THE STOCK MARKET Previous studies have found evidence of overreactions in the stock market. In their study from 1985, De Bondt and Thaler investigated if Bayes rule, which is a concept from research in experimental psychology saying that most people tend to overreact to unexpected events, could also be applied to the stock market. They found that this was indeed the case. Their study from 1990 also showed that not only the common man overreacts in the stock market, but that seasoned professionals are just as human. Michaely, Thaler and Womack (1995) researched the effects of initiations or omissions of cash dividend payments by public companies. In their study they found evidence of overreactions to the news. They also found that the overreactions were of a greater magnitude when it came to omissions compared to initiations. In my mind, this could be interpreted as people are more afraid of, and therefore overreact more to, negative news than positive ones. 4

8 3. HYPOTHESES To investigate whether or not there is a news effect from recommendations made in Swedish business magazines an event study methodology will be used. When looking at Swedish business media as a whole the following null hypothesis is used: H 1 : There are no significant abnormal returns around the publishing date, when investigating recommendations made by the main Swedish business magazines. In order to see if there is a difference between the three main Swedish business magazines three null hypothesis are used: H 2 : There are no significant abnormal returns around the publishing date, when investigating recommendations made by Veckans Affärer. H 3 : There are no significant abnormal returns around the publishing date, when investigating recommendations made by Affärsvärlden. H 4 : There are no significant abnormal returns around the publishing date, when investigating recommendations made by Aktiespararen. Even if there happens to be a news effect this simply means that investors feel that the announcement of the recommendations is information worth trading on. In order to see if the analysts are actually proficient in finding over or underpriced stocks the long-term performance has to be investigated. For this purpose these three hypotheses are used: H 5 : There are no significant BHARs, when investigating recommendations made by Veckans Affärer. H 6 : There are no significant BHARs, when investigating recommendations made by Affärsvärlden. 5

9 H 7 : There are no significant BHARs, when investigating recommendations made by Aktiespararen. 4. METHODOLOGY When investigating abnormal returns one is faced with the problem of determining what constitutes normal returns. The most widely used model is the Capital Asset Pricing Model (CAPM). However, this model is not without problems and in 1992 Fama and French suggested an improvement. By adding firm size and Book to Market the model would more accurately reflect reality. In this thesis, however, I have chosen to use the CAPM. This is a better approach in a small market like Sweden, with too few companies and too many unique ones (Simonov, 2008) 4.1 EVENT STUDY The event study methodology has many applications and is therefore widely used. It is used both for firm-specific and economy-wide events (MacKinlay, 1997). As in the majority of applications I am investigating the effect of an event on the stock price using financial market data. Given that markets are rational, the informational content contained in the event must be incorporated into prices instantly and therefore the effect of the event on prices should be seen over a relatively short time period (MacKinlay, 1997) Event Definition The event window is set from the day before publishing date (τ = 0) and three days forward (see Figure 1). This is to capture any effect of the announcement of the recommendations. In other words, the event window is defined as the period where the abnormal returns are accumulated. Since I use daily data, the event window should only have to include the day of the announcement. In practice, however, the event window is normally expanded to include at least one day after the announcement (MacKinlay, 1997). Ryngaert and Netter (1990) proved empirically that a short event window will most likely capture 6

10 the significant effect of an event. After confirming with my tutor, Ramin Baghai, and following common practice, I decided to use an event window of three days, starting the day before publishing and ending on the day after, in an attempt to best capture whether or not the market had anticipated the event. Figure 1 Timeline of Estimation and Event Window Estimation Window Event Window T 0 T 1 τ T Days 3 Days The estimation window consists of 120 trading days prior to the event window. This is suggested by MacKinlay (1997) and is in line with common practice. 120 trading days is approximately half a year and can be considered a sufficiently long period to estimate normal returns. The exact date when the outcome of the event is announced is important when performing an event study. The event should also be unexpected (MacKinlay, 1997). Even though it is well known when the magazines are published the recommendations themselves are not, making it an unexpected event. Normally the event window is set so that the day before the announcement is the first day of the event window, but when looking at the individual days, as a small addition to this thesis, I have not included the day before publishing. The idea behind this is to see the world from the point of view of a normal magazine buyer. Any information that might leak out before the actual publishing, about which stocks who will receive a certain recommendation, will not be available for the average reader, and it is therefore of no interest to him. 7

11 4.1.2 Confounding Effects A problem with other event study methodologies, using a longer event window, is that it becomes difficult to isolate the effect of the event from the effect of other events that might impact stock returns. The shorter the event window you use, the less likely it is that confounding events will occur. When performing an event study one assumes that the effect of the event is isolated from the effects of other events (McWilliams and Siegel, 1997). Since I have chosen an event window of three days, which can be considered to be relatively short, I assume no confounding effects Hypotheses Testing To measure the impact of an event, one needs to calculate the abnormal returns. In order to be able to identify abnormal returns one must assume efficient markets (McWilliams and Siegel, 1997). Abnormal returns are defined as actual ex post returns over the event window minus the normal returns over the event window. Therefore, for firm i and event date τ abnormal returns are calculated as follows (MacKinlay, 1997): AR!" = R!" E(R!" X! ) R!" stands for the actual return and E(R!" X! ) denotes the normal return, which is equivalent to the expected return if the event would not occur. The expected normal return is estimated using the market model in the estimation window of 120 trading days prior to the event window. Actual returns are calculated as follows: R!" = P!" P!,!!! 8

12 P!" is the closing price of stock i on day t. Returns for my benchmark, the OMXS Index, is calculated using the same methodology as for the stock returns, and the results are then used in the market model to calculate abnormal returns. For any security i the market model is defined as: R!" = α! + β! R!" + ε!" R!" is the market return observed at t, and α! and β! are estimated using the ordinary least squares regression (MacKinlay, 1997). α! and β!, from the regression, are then used to calculate the sample abnormal returns: AR!" = R!" α! β! R!" Another common model used to model normal returns is the constant mean return model. Compared to this, the market model has the advantage that it removes the portion of the return that is related to variation in the market s returns (in this case the OMXS Index). This reduces the variance of the abnormal returns and this gives an increased ability to find the true effects from the events. A higher R2 means a greater reduction in the variance of the abnormal returns (MacKinlay, 1997). Under the null hypothesis that the events have no impact on the means, or the variances of the returns, the distribution of the sample abnormal returns of the observations in the event window is normally distributed with (MacKinlay, 1997): AR!" ~N(0, σ! (AR!" )) This assumption will be further discussed in section 7.2 Econometric Issues. To calculate the variance of the abnormal returns the sum of the squared abnormal returns is divided by the number of days in the event window (L! ), adjusted for the loss of the degrees of freedom: Var AR!" = (AR!")! (L! 2) 9

13 In order to draw overall inferences about the events, the abnormal returns need to be aggregated over time and across securities (MacKinlay, 1997). Define CAR! (τ!, τ! ) as the sample cumulative abnormal return (CAR) from τ! to τ! where T! < τ! τ! T!. Then the CAR from τ! to τ! is:!! CAR! τ!, τ! = AR!"!!!! With increasing L, the variance of CAR! approaches σ!!! τ!, τ! = (τ! τ! + 1)σ!! The individual abnormal returns (AR!" ) are then used to calculate the average abnormal return (AR! ) for N events: AR! = 1 N AR!" To get the average variance of the abnormal returns the sum of the variances of the individual abnormal returns are divided by the square of the number of events: Var AR! = 1 N! Var(AR!" )!!!! Then the average abnormal returns can be aggregated in order to obtain the cumulative average abnormal returns over the event window (from τ! to τ! ):!! CAR τ!, τ! = AR!!!!! 10

14 After that, the variance of the CAR can be calculated by summing the variances of the AR! over the event window (MacKinlay, 1997): Var CAR τ!, τ! = Var(AR! )!!!!!! It is now possible to test the null hypothesis that the abnormal returns are zero during the event window since the cumulative average abnormal returns follow the following distribution: CAR τ!, τ! ~N 0, Var(CAR(τ!, τ! )) The following t-statistic can now be used to test the null hypothesis (MacKinlay, 1997): t!"# = CAR(τ!, τ! ) a ~N(0,1) Var(CAR(τ!, τ! ))!/! The longer the estimation window (L 1 ), and the larger the amount of securities (N), the more reliable the results. Worth noting is that the results are not exact since an estimator of the variance is used in the denominator. The appropriate estimator is the sample variance measure from the market model regression in the estimation window (MacKinlay, 1997). If any significant results are found these will be tested for robustness using the nonparametric Wilcoxon sign and sign-rank tests. These tests are free of assumptions regarding the distribution of returns, and do therefore not rely on the assumption of normality. This way of using these non-parametric tests, to test the results from parametric tests for robustness, is common practice (MacKinlay, 1997). 11

15 4.2 LONG TERM PERFORMANCE When investigating long-run abnormal performance of stocks you have to choose between two methods, the buy-and-hold abnormal return method (BHAR) or the cumulative abnormal return method (CAR). One reason for why the BHAR method is to be preferred is that it is more intuitive in the sense that it captures the investor experience of holding a security for a long post event period (Lidén, 2006). The method is not without problems, but Barber and Lyon (1997) still argue that even though BHARs can give rise to biased test statistics, it is nevertheless preferred for detecting over-performance of published recommendations. By using a simple regression of buy-and-hold abnormal returns against cumulative abnormal returns they showed that cumulative abnormal returns are a positively biased predictor of buy-and-hold abnormal returns. Given this evidence they promote buy-and-hold abnormal returns in tests intended to detect long- run abnormal stock returns. Not all firms which received a buy or a sell recommendation remained on the Stockholm Stock Exchange until In order to minimize survivorship bias I estimated abnormal performance for these firms for as long as they remained on the exchange. This procedure was used in Kothari and Warner (1997) and Lidén (2006). The buy-and-hold return of firm i (BHR! ) is obtained by compounding its returns over a chosen time period following the publication day. This replicates an investment strategy that consists in buying and holding shares for a period of time. The same logic applies to the reference portfolio associated with the issuing firm i. The difference between the BHR of the issuing firm and the BHR of its reference portfolio, which in this thesis corresponds to the OMXS Index, is the buy-and-hold abnormal return (BHAR). To calculate the BHARs for each recommended stock i during the period T the following formula was used:!! BHAR!" = 1 + R!"!!!!!! 1 + R!" T has been set to 1 month, 6 months and 12 months, as well as the entire period from publishing day until This, the longest holding period, is referred to as the 12

16 Max holding period. In this thesis I is the return from the OMXS index. The mean BHAR for each portfolio, or type of recommendation, has been calculated as a simple mean, which means that each stock in the portfolio is equally weighted. This way of calculating the mean BHAR has the advantage that it mimics an investor who invests an equal amount of money in each recommended stock (Lidén, 2006). 5. DATA The data consists of stock recommendations made in Swedish business magazines during the period The reason for choosing this period is that the aim of this thesis is to evaluate current performance and the recommendations therefore have to be fairly recent. In order to see how the recommendations perform with a 12-month holding period no new stocks could be added to any portfolio after For example, the portfolio for Veckans Affärer then contains 135 stocks and the only reason why this should change is if a stock is delisted. The magazines considered in this thesis are Veckans Affärer (18200 copies per issue 2 ), Affärsvärlden (21100 copies per issue 3 ) and Aktiespararen (61600 copies per issue 4 ). The only other Swedish business magazine currently giving stock recommendations is Privata Affärer, but since they during 2009 did not offer proper stock recommendations this magazine was not included in this study. I am therefore confident that the area of stock recommendations made by Swedish business magazines has been sufficiently covered. The total number of recommendations made during in the three magazines combined were of these came from Veckans Affärer, 173 from Affärsvärlden and 72 from Aktiespararen. For different reasons, some of these had to be removed from the final study. For example, some were not available in DataStream, and some had not been listed long enough on the Stockholm Stock Exchange to have a sufficiently long enough estimation window for the event study. The total sample for the period ended up containing 354 recommendations. 2 Tidningsstatistik, Tidningsstatistik, Tidningsstatistik,

17 The magazines produce far more buy recommendations then sell recommendations. From the 161 recommendations made in Veckans Affärer, 140 recommendations remained in the final sample. 102 of these were buy recommendations (73%). For Affärsvärlden, 122 of the 147 recommendations (83%) were buy recommendations. Finally, for Aktiespararen 63 of 67 (94%) recommendations made were recommendations to purchase. As noted by DeBondt and Thaler (1990), it is well known that buy recommendations issued by brokerage houses far exceed sell recommendations. According to Linden (2004), it is not self evident that the same would be true when the source is a business magazine. He says that There are typically no incentives for journalists to give either a favorable or an unfavorable recommendation., but the skewness in my sample indicates that there might be. There are several reasons why a magazine would want to keep the companies they investigate happy. For one, magazines need advertisers, and if money is not the issue then they are at least in need of information. A firm that has gotten a sell recommendation might be less willing to offer interviews etc. The financial data comes from DataStream. This is a highly reliable source of information. Dividend adjusted daily returns were converted into monthly portfolio returns. As a benchmark the OMXS All Share has been used. When the holding period is set to something less than Max there are some trading days when there are no stocks in the portfolios (during the magazines summer break for instance). In order to make the portfolios comparable to the market portfolio these days have been removed from the market portfolio s performance as well. 14

18 6. EMPIRICAL FINDINGS 6.1 EVENT STUDY The event study is aimed to find if there is a news effect from the publishing of the recommendations. That is, if there new information in the recommendations that have not yet been priced into the market. In all cases the estimation window is 120 trading days and the event window is the publishing day and the day before and after publishing. Table I - Summary of Event Study For more detailed information see 10. Appendix Magazine Related,Hypothesis Significance N t7value Average,Price,Impact Average,Daily,Price,Impact All H1 Yes Veckans,Affärer H2 Yes Affärsvärlden H3 Yes Aktiespararen H4 Yes When using a sample of all three magazines the results are significance on a five percent level with a t-value of This means that I can reject the null hypothesis of no abnormal returns in the event window. The average price impact is 1.1 percent, which implies an average daily price impact during the event window of 0.37 percent. (Table 1 in 10. Appendix) The results were confirmed on a five percent level with both a Wilcoxon sign test (Table 2 in 10. Appendix) and a Wilcoxon sign-rank test (Table 3 in 10. Appendix). The next three sections will now test whether the news effect will persist when limiting the sample to a single magazine Veckans Affärer When the sample is limited to only contain recommendations made by Veckans Affärer the results are significance on a five percent level with a t-value of Thus, the null hypothesis of no abnormal returns in the event window for recommendations made by Veckans Affärer can be rejected. Average price impact is 15

19 1.6 percent giving an average daily price impact during the event window of 0.52 percent. (Table 4 in 10. Appendix) Both a Wilcoxon sign test (Table 5 in 10. Appendix) and a Wilcoxon sign-rank test (Table 6 in 10. Appendix) confirms the result on a five percent level. Table 7 in section 10. Appendix describes the statistics for the individual days. Looking at the individual days following the publishing of the recommendations I find that on both the publishing day and the day after there are significant abnormal returns. The result for the publishing day is highly significant with a t-value of and a mean abnormal return of 0.97 percent. For day PD+1 the t-value is also high at and the abnormal return is 0.53 percent. This means that for these days I can reject my null hypothesis of no abnormal returns at the five percent level, and this would suggest that there is a news effect from recommendations made in Veckans Affärer. Interestingly, there also seems to be common that this results in an overreaction since the following day shows a negative return, which is significant on a five percent level. Day PD+3 did not offer any significant results, but the last day in the event window shows negative mean abnormal returns of percent on a five percent level with a t-value of Affärsvärlden With a sample limited to recommendations made by Affärsvärlden the results are similar to the ones from Veckans Affärer. The null hypothesis of no abnormal returns during the event window can be rejected on the five percent level since the t-value is The average price impact is 1.0 percent, giving an average daily price impact during the event window of 0.35 percent. (Table 8 in 10. Appendix) A Wilcoxon sign test (Table 9 in 10. Appendix) and a Wilcoxon sign-rank test (Table 10 in 10. Appendix) confirms the results on a five percent level. Table 11 in 10. Appendix lists the statistics for the individual days. For Affärsvärlden, the results were only significant on the five percent level on the third day of the event window. This day offered a mean abnormal return of percent, with a t-value of The conclusion here must be that I cannot reject the null hypothesis of no 16

20 abnormal returns on a five percent level for Affärsvärlden other than for day three of the event window Aktiespararen Limiting the sample to only include recommendations made by Aktiespararen renders results similar to the other two magazines. This time, however, the significance is lower with a t-value of Nevertheless, this still means that the results are significant on a five percent level and therefore the null hypothesis of no abnormal returns can be rejected for recommendations made by Aktiespararen. Average price impact is 0.4 percent, meaning an average daily price impact in the event window of 0.12 percent. (Table 12 in 10. Appendix) The results are confirmed on a ten percent level by a Wilcoxon sign test (Table 13 in 10. Appendix) and on a five percent level by a Wilcoxon sign-rank test (Table 14 in 10. Appendix). Table 15 in 10. Appendix lists the statistics for the individual days. For Aktiespararen, the results for the publishing day are significant on a five percent level with a t-value of and a mean abnormal return of 0.47 percent. For the following days the results are not significant on even a ten percent level. Hence, I conclude that the null hypothesis of no abnormal returns can be rejected on a five percent level for the publishing day, but not for the days following. This would suggest that there is a news effect for Aktiespararen, but that it is very short lived. 6.2 LONG TERM The long term analysis aims to discover if there was indeed some substance to the buy and sell recommendations, i.e. if the reporters had found over or underpriced stocks that later had a price correction Veckans Affärer When using the entire sample for Veckans Affärer the results are not significant on a five percent level for either the 1-month, 6-month, 12-month or the Max holding 17

21 period. The reporters are less prone to give a sell recommendation, for reasons discussed in section 3. Data, so they would likely be highly convinced that a stock will underperform in order to publish a sell recommendation. In order to investigate this, the sample was limited to only the stocks receiving a sell recommendation. This did not result in any abnormal returns significant on a five percent level for either the 1-month, 6-month, 12-month or the Max holding period. For the 1-month holding period the results were a lot closer to being significant (with a t-value of ) than the other holding periods. The negative BHAR of -2.3 percent indicates a bad performance for this portfolio. (Table 16 in 10. Appendix) The same lack of significance on a five percent level was the result when only stocks receiving a buy recommendation were included in the sample. (Table 16 in 10. Appendix) With the 1-month holding period, Analyst 3 comes closest to being significant with a t-value of The BHAR of , however, indicates poor performance. The highest BHAR with the 1-month horizon comes from Analyst 6, but the t-value of is low. (Table 16 in 10. Appendix) For the Max holding period, only one analyst was close to have abnormal returns significant on a five percent level. He had a BHAR of 1.1 percent with a t-value of None of the others were close to being significant on the five percent level, and the analyst with the second highest t-value actually had a negative BHAR of -1.0 percent. (Table 16 in 10. Appendix) When dividing the sample by analyst, the indications are that some have shorter and some have a longer investment horizon in mind when giving their recommendations. For example, Analyst 1 gets a negative BHAR of -0.9 percent for the 1-month period and a positive BHAR of 1.1 percent for the longest holding period. This can be compared to Analyst 6, who has a BHAR of 1.7 percent for the 1-month holding period, but whose results decline the longer the holding period. With the maximum holding period his BHARs have declined to However, none of the results for the analysts were significant on the five percent level. (Table 16 in 10. Appendix) 18

22 Table II Summary Long-term findings for Veckans Affärer For more detailed information see 10. Appendix Total Sell.only Buy.only Period BHAR t n Period BHAR t n Period BHAR t n 1.Month 10,196 10, Month 12,348 11, Month 0,442 0, Months 10,388 10, Months 10,685 10, Months 10,421 10, Months 0,023 0, Months 10,664 10, Months 0,243 0, Infinity 0,120 0, Infinity 0,255 0, Infinity 0,059 0, Analyst.1 Analyst.2 Analyst.3 Period BHAR t n Period BHAR t n Period BHAR t n 1.Month 10,863 10, Month 11,477 11, Month 12,846 11, Months 11,776 11, Months 10,937 11, Months 11,729 11, Months 10,630 11, Months 0,028 0, Months 10,829 10, Infinity 1,121 1, Infinity 10,087 10, Infinity 10,433 10, Analyst.4 Analyst.5 Analyst.6 Period BHAR t n Period BHAR t n Period BHAR t n 1.Month 0,626 0, Month 13,424 10, Month 1,677 0, Months 0,986 0, Months 10,844 10, Months 1,631 0, Months 0,506 0, Months 0,010 0, Months 0,269 0, Infinity 0,355 0, Infinity 11,032 10, Infinity 10,748 10, Analyst.7 Period BHAR t n 1.Month 1,588 0, Months 10,635 10, Months 10,861 10, Infinity 10,400 10, Affärsvärlden The results when using the entire sample of recommendations made by Affärsvärlden did not show that the null hypothesis of no abnormal returns could be rejected on a five percent significance level for either the 6-month, 12-month or Max holding period. For the 1-month holding period the results were much closer to being significant than for the other holding periods with a t-value of Compared to Veckans Affärer, all BHARs were positive (however low) when all recommendations made by Affärsvärlden were included. (Table 17 in 10. Appendix) When dividing the sample into only sell or only buy recommendations, the sell recommendations gave rise to BHARs significant on a 25 percent level with a 6- month holding period, and on a 15 percent level with a 1-month holding period. This is, however, not below the accepted significance level of five percent. The sell-only portfolio produced positive BHARs for all holding periods, indicating that sell recommendations by Affärsvärlden are more reliable than the buy recommendations, which produced both positive and negative BHARs, depending on the holding period. 19

23 The sample with only buy recommendations did not give any BHARs significant on a five percent level with either the 6-month, 12-month or Max holding period. (Table 17 in 10. Appendix) When looking at individual analysts, one person (Analyst 1) had a t-value of for the 12-month holding period. This means that for Analyst 1, with a holding period of 12 months, the null hypothesis of no abnormal returns can be rejected on the five percent level. He managed to produce a BHAR of 2.7 percent with this holding period. (Table 17 in 10. Appendix) For Analyst 3 and 6, the indications are that their recommendations are made with a short investment horizon in mind as their BHAR decrease with increasing holding periods. Analyst 4 appears even more shortsighted as her 1-month BHAR is 3.1 percent and her second highest BHAR (with the 6-month holding period) only measures 0.9 percent. For the other analysts no clear pattern emerges. (Table 17 in 10. Appendix) For the maximum holding period, the analyst that was closest to have BHARs significant on the five percent level with a t-value of interestingly enough yielded negative BHARs. The person with the second most significant BHARs, with a t-value of 1.206, managed to yield positive BHARs. (Table 17 in 10. Appendix) Table III Summary Long-term Findings for Affärsvärlden For more detailed information see 10. Appendix Total Sell.only Buy.only Period BHAR t n Period BHAR t n Period BHAR t n 1.Month 1,270 1, Month 3,946 1, Month 0,794 0, Months 0,293 0, Months 1,565 1, Months <0,087 <0, Months 0,056 0, Months 1,003 0, Months <0,362 <0, Infinity 0,116 0, Infinity 0,998 1, Infinity <0,209 <0, Analyst.1 Analyst.2 Analyst.3 Period BHAR t n Period BHAR t n Period BHAR t n 1.Month 3,097 1, Month <1,527 <0, Month 0,249 0, Months 1,329 0, Months <2,455 <1, Months 1,571 1, Months 2,664 2, Months <0,666 <0, Months <0,797 <0, Infinity 0,717 1, Infinity <0,328 <0, Infinity <1,113 <1, Analyst.4 Analyst.5 Analyst.6 Period BHAR t n Period BHAR t n Period BHAR t n 1.Month 3,061 1, Month 0,077 0, Month 1,514 0, Months 0,864 0, Months <0,501 <0, Months 1,206 0, Months 0,693 0, Months <0,172 <0, Months 0,534 0, Infinity 0,723 1, Infinity <0,142 <0, Infinity 0,153 0,

24 6.2.3 Aktiespararen When the entire sample is used the results are not significant on the five percent level and therefore the null hypothesis of no abnormal returns cannot be rejected for recommendations made by Aktiespararen. (Table 18 in 10. Appendix) When separating buy and sell recommendations, the sample with only sell recommendations and the Max holding period comes closest to being significant on the desired five percent level with a t-value of This sample also has the highest BHAR of all the portfolios for Aktiespararen with a BHAR of 1.8 percent. Although not significant on the five percent level, this gives some indication that the sell recommendations by Aktiespararen, as for Affärsvärlden, are more reliable than the buy recommendations. The sample with only buy recommendations only produces negative BHARs, and the results are far from being significant on the five percent level. (Table 18 in 10. Appendix) When looking at the 1-month holding period, Analyst 2 produces a BHAR of -2.0 percent with a t-value of This means that the null hypothesis of no abnormal returns can be rejected on a 15 percent level, but not on the desired five percent level. (Table 18 in 10. Appendix) For the maximum holding period, the analyst that comes closest to being significant, with a t-value of 1.479, also produced a positive BHAR of 0.8 percent. Even though far from significant, it is noteworthy that the other analyst produced a negative BHAR. (Table 18 in 10. Appendix) Although not significant on the five percent level, it is interesting that for all holding periods the two analysts produce BHARs of a similar magnitude but with opposite signs. Not surprising then is that when looking at the performance of the entire magazine the BHARs are close to zero and with low t-values. (Table 18 in 10. Appendix) As with the section for Veckans Affärer I must point out that none of the results were significant on the desired five percent level. 21

25 Table IV Summary Long-term Findings for Aktiespararen For more detailed information see 10. Appendix Total Sell.only Buy.only Period BHAR t n Period BHAR t n Period BHAR t n 1.Month 0,050 0, Month 0,144 0, Month 0,354 0, Months :0,268 :0, Months :0,722 :0, Months :0,259 :0, Months :0,380 :0, Months 0,741 0, Months :0,291 :0,59 36 Infinity 0,083 0, Infinity 1,765 1, Infinity :0,025 :0, Analyst.1 Analyst.2 Period BHAR t n Period BHAR t n 1.Month 1,918 1, Month :2,016 :1, Months :0,662 :0, Months 0,532 0, Months :0,462 :1, Months 0,248 0, Infinity :0,119 :0, Infinity 0,751 1, DISCUSSION 7.1 EVENT STUDY The first step in this part of the thesis is to start on an aggregated level and determine if recommendations published in the main Swedish business magazines give rise to significant abnormal returns, i.e. if there is a news effect in the recommendations. To investigate this the following null hypothesis is used: H 1 : There are no significant abnormal returns around the publishing date, when investigating recommendations made by the main Swedish business magazines. In my sample I find that there are significant abnormal returns when considering the 354 recommendations made in the three main Swedish business magazines used in this study. This is in line with studies like Liang (1999), who finds a significant announcement effect when investigating recommendations made in the Wall Street Journal. Ferreira and Smith (2003) also find that journalists achieved a significant announcement effect in their study of the TV show Wall Street Week with Louis Rukeyser. Yazici and Muradoğlu (2002) similarly find an announcement effect when investigating recommendations made in Moneymatik. All this indicates that although conditions are different in different countries there are usually some sources of information that is the go to source of information in that country. 22

26 The following sections will divide the analysis by magazine. Affärsvärlden and Veckans Affärer are the main competitors, and of the two Affärsvärlden is widely considered to be the one to offer the most in depth knowledge. The simple fact that Affärsvärlden offers a target stock price in their buy and sell recommendations shows why this is the case Veckans Affärer In this section the following hypothesis is the main focus: H 2 : There are no significant abnormal returns around the publishing date, when investigating recommendations made by Veckans Affärer. Here the null hypothesis can be rejected on a five percent level and I therefore conclude that there are significant positive abnormal returns for the sample consisting of only recommendations made by Veckans Affärer. This means that there is a news value in the recommendations given by Veckans Affärer. The recommendations have not been anticipated and therefore have not been incorporated in the stock prices. My results are in line with Liang (1999), Ferreira and Smith (2003) and Yazici and Muradoğlu (2002) who also find significant announcement effects from recommendations made by journalists. When looking at individual days following the announcements I find that the publishing day and the day after offers significant abnormal returns. Interestingly, there also seems to be common that this results in an overreaction since the following day shows a negative return, which is significant on a five percent level. Previous studies have also found evidence of overreactions in the stock market. De Bondt and Thaler (1985) investigated if Bayes rule, which is a concept from research in experimental psychology saying that most people tend to overreact to unexpected events, could also be applied to the stock market and found that this was indeed the case. Michaely, Thaler and Womack (1995) found evidence of overreactions when looking at initiations or omissions of cash dividend payments. They also found that the overreactions are of a greater magnitude when it comes to omissions compared to 23

27 initiations. This could be interpreted as people are more afraid of, and therefore overreact more to, negative news than positive ones Affärsvärlden This section aims to investigate if the following null hypothesis can be rejected: H 3 : There are no significant abnormal returns around the publishing date, when investigating recommendations made by Affärsvärlden. The null hypothesis can be rejected on the five percent level and I therefore find that there are significant abnormal returns resulting from recommendations made by Affärsvärlden. In the same way as for Veckans Affärer I get results that point in the same direction as Liang (1999), Ferreira and Smith (2003) and Yazici and Muradoğlu (2002) who also find significant announcement effects from recommendations made by journalists. When investigating individual days the results were only significant on the five percent level on the third day after publishing. What is surprising here is that the more highly regarded magazine, Affärsvärlden, did not offer a news effect on the publishing day and the day after, but its lesser competitor, Veckans Affärer, did. A possible reason for this could be that Veckans Affärer has a higher circulation ( copies in ) than Affärsvärlden ( copies in ). The difference was even higher at the start of the period I am investigating since Veckans Affärer sold fewer copies in 2011 than in 2010, a reduction of 14%, whereas Affärsvärlden increased their numbers by five percent last year (Lundqvist, 2012) Aktiespararen Looking at the final magazine, Aktiespararen, the null hypothesis becomes: H 4 : There are no significant abnormal returns around the publishing date, when investigating recommendations made by Aktiespararen. 5 Lundqvist, Lundqvist,

28 When limiting the sample to only recommendations made by Aktiespararen I get similar results to the other magazines. The null hypothesis can be rejected on a five percent level and I therefore conclude that there are significant abnormal returns as a result of recommendations made by Aktiespararen. Once again, this is a result in line with Liang (1999), Ferreira and Smith (2003) and Yazici and Muradoğlu (2002). For the individual days the results are only significant on a five percent level when looking at the day of publishing. For the other days the results are not significant even on a ten percent level. This implies that there is a news effect for recommendations made by Aktiespararen, but that the effect is very short lived. Liang (1999) produce similar results when he finds a significant two-day announcement effect that is later reversed within 15 days. The fact that information is quickly priced into the market supports the claim by for instance Fama (1998) that stock markets are efficient. 7.2 LONG TERM Veckans Affärer This section aims to investigate if the following null hypothesis can be rejected: H 5 : There are no significant BHARs, when investigating recommendations made by Veckans Affärer. When looking at recommendations made by Veckans Affärer I did not find any significant BHARs so the null hypothesis cannot be rejected. This is in line with Lidén (2006) who found that recommendations made by Swedish reporters did not result in significant BHARs. Others that claim that analyst recommendations do not outperform the market include Liang (1999) and Yazici and Muradoğlu (2002). Desai et al (2000) on the other hand found that recommendations made in the Wall Street Journal resulted in significant returns when using a buy-and-hold strategy. When dividing the sample into buy-only or sell-only recommendations Lidén (2006) found that the recommendations resulted in significant negative BHARs for a buyonly portfolio with a 6-month holding period, and a sell-only portfolio with a holding 25