Bachelor Thesis Analyst recommendations and abnormal returns

Analyst recommendations and abnormal returns

An event study on OMX Stockholm 30

Bachelor Thesis

Authors: Krenare Salihu & Ludwig Flank

Zetterström

Supervisor: Christopher von Koch Examiner: Håkan Locking

Abstract

The main purpose of this study is to contribute to the previous literature by evaluating positive changes in analysts' consensus recommendations of the stocks listed in OMXS30. We analyze if new positive changes in consensus recommendations correspond with lower abnormal returns. By conducting an event study and performing a series of different statistical tests, we find that positive changes in analyst consensus provide a short lived negative mean abnormal return in certain cases. We argue that this implies that investors might interpret positive changes as a sell signal. Furthermore, we find some pieces of evidence to suggest that it may actually be changes in the mean target price rather than changes in recommendations that causes the movements in abnormal returns.

Key words

Finance, event study, recommendations, abnormal returns, behavioral finance

Acknowledgments

We would like to send a very special thank you to Christopher von Koch for

his key role in the creation of this thesis. We are most grateful for his

continuous support and guidance throughout the writing of this thesis. We

would also like to extend our gratitude to our opponents and our examiner

Håkan Locking for their feedback on our work. Finally, we would like to thank

master’s student Nathalie Sommar Lindskog for her guidance in STATA.

Table of contents

1 Introduction 1

1.1 Background 1

1.2 Problem Discussion 2

1.3 Purpose and Contribution 3

1.4 Disposition 4

2 Theoretical framework 4

2.1 Efficient Market Hypothesis 4

2.2 Adaptive Market Hypothesis 5

2.3 Behavioral Finance 6

2.4 Noise Trading 7

2.5 Piggyback Theory 8

3 Literature Review 8

3.1 Stock Recommendations 8

3.1.1 A Critical View on Recommendations 10

3.1.2 The Aspect of Noise 11

4 Data and Research Method 11

4.1 Data 11

4.2 Research Method 12

4.2.1 Significance Test 15

4.2.2 Spearman Rank Correlation 15

4.2.3 Calculations and Variables 16

4.2.4 A Critical View on the Method and Data 21

5 Empirical Results 23

5.1 Descriptive Statistics 23

5.2 Abnormal Returns 24

5.2.1 Cumulative Abnormal Returns 25

5.3 Bivariate Analysis 27

5.4 Regression Analysis 29

5.4.1 Positive Change 31

5.4.2 Negative Change 32

6 Discussion 33

7 Summary and Conclusions 38

8 References 40

8.1 Scientific Articles 40

8.2 Literature 42

8.3 Media and News Outlets 42

8.4 Websites 43

9 Appendix 45

1 Introduction

1.1 Background

Buy low and sell high is perhaps the most repeated phrase when talking about the stock market and how to approach it. While it is the basic goal of investors, buying low and selling high isn’t always easy, why? Because investors may lack the time, resources and knowledge required in predicting stock price and market movements.

There are strategies with varying degrees of accuracy an investor can implement to forecast movements, most of which are difficult, time consuming and require extensive knowledge of both business and economics.

This is not only true for individual investors, but firm’s as well, which may have the resources but lack the time, knowledge or manpower necessary to analyze large numbers of financial statements, predicting micro- and macroeconomic trends and ultimately translate it to a stock price. This is where financial analysts come in.

Everyone from banks, to financial institutions, to newspapers hire analysts to produce forecasts of such things as growth, earnings per share and future cash flows. As mentioned above, an analyst will also analyze microeconomic factors such as the specific sector the firm is in and the macroeconomic trends of the market as a whole to determine how they may affect the company now

and, in the future

(McClure, 2020). Usually, analysts will produce a recommendation based on either fundamental or technical analysis. Fundamental analysis means that private and public information is a part of the valuation basis. The technical aspect means that historical data is studied and used as a valuation basis to predict future share prices. The analysis may also come with a strategic tool called a SWOT-analysis which assesses the firm's strengths, weaknesses, opportunities and threats going forward. Based on these variables the analyst will produce a recommendation whether an investor should buy, sell or hold the analyzed asset. The scale and terminology used for stock recommendations depends on which bank or firm issues the recommendation, but the purpose is the same, recommendations exist to provide investors with information, in order to make value estimates of companies more accurate (Jegadeesh et al., 2004).

Larger companies tend to receive recommendations more frequently than smaller companies. This may take place in connection with annual reports and interim reports, but also major events that may affect the value of shares. The analysts will revise their forecast and valuation. However, they can repeat themselves with an identical recommendation again, which means that they remain at the same position.

1.2 Problem Discussion

These earnings forecasts and recommendations of stocks from analysts at big banks, financial institutions and even newspapers seem to be plentiful and very easily accessible on the internet nowadays. In March of 2021 the Wall Street Journal reported on the increasing number of so-called “stock influencers”. These influencers include everyone from business leaders, fund managers and billionaires like Elon Musk to more “common” people, for example Keith Gill also known as Roaring Kitty, who use social media platforms to relay trading strategies, recommended stock picks and stock analysis. Gill is most known for his analysis of GameStop which ultimately influenced a number of people on Reddit’s WallStreetBets, an online trading forum, to “declare war” on the hedge funds with short positions in the company. The author describes an almost cult following of these influencers where every comment is taken as gospel (Otani, 2021).

This increase in adherence to stock recommendations, be it an institutional analyst or a “stock influencer”, could very well be correlated with the fact that in just the last ten years, the market activity of retail investors on the U.S. stock market has doubled. In 2020 retail investors accounted for 19,5% of all market activity, with daily peaks reaching as high as 25% (Osipovich, 2020). One explanation for this increase of traders may be because “free” trading apps, such as RobinHood in the U.S. and Avanza and Nordnet in Sweden, have made equity trading more easily accessible for the “common” man and woman. Just in 2020, Avanza saw a surge of 300 000 new users on its platform to a total number of 1,3 million users (Avanza, 2021).

The usefulness and accuracy of analysts' stock recommendations is an extensively studied subject within finance academia. However, there is still no clear consensus from the existing literature on how well, if at all, the analysts' stock recommendations actually correlate with stock price movements.

Previous studies such as Jegadeesh et al. (2004) have found that in most cases stocks with new favorable recommendations outperformed the less favored stocks, but in some cases the opposite could be found, where the analysts most favored stocks underperformed both the market and the least favored stocks. Other studies have ended up with similar results, where favored stocks usually outperform those less favored but far from always. For

instance,

Barber et al. (2003) found that for two straight years the stocks most favored by analysts underperformed the analysts least favored stocks. Whether or not analysts' recommendations are deemed effective seems dependent on the chosen time periods, the method

used,

and which variables are considered.

Considering

the absence of a clear consensus over the effectiveness of analyst stock recommendations coupled with the increase in retail trader activity, we feel that a follow up to previous works, such as Barber et al. (2001 & 2003) and Jegadeesh et al.

(2004), is needed. Our thesis will focus on the Swedish stock market and with help from previous studies in the field as well as established economic theories, we will seek to answer the following research question; Can a positive, new, consensus recommendation have an adverse effect on abnormal return? If it does have an adverse effect; In which cases and during which periods may this occur?

1.3 Purpose and Contribution

The purpose of this study is to evaluate changes analysts' consensus stock recommendations of stocks listed on the OMXS30 index and analyze if they correlate with higher or lower abnormal returns.

While our thesis is not the first of its kind, it contributes to a better understanding of the Efficient Market Hypothesis (EMH) and how information, in this case new recommendations, may influence stock prices. Furthermore, our thesis contributes to the behavioral finance field by analyzing both analyst and investor behaviors and providing a better understanding of herding behavior, loss aversion and piggybacking. We also contribute to the concept of noise trading and the effects it has on markets and market prices, in this case the Swedish stock market. We provide an increased understanding of if, how and when analysts’ favorable stock recommendations may have adverse effects on the price of a stock.

Our contribution, per sei, is not to explicitly construct an optimal portfolio or to offer investment advice, instead we seek to provide the growing number of novice investors a clearer understanding of how this specific kind of information may affect returns and why.

1.4 Disposition

The thesis consists of seven chapters, where the first part explains our problem, the relevancy, purpose and background. The second and third part contains our theoretical framework and a review of previous literature. Part four presents the data and methodology used to answer the research question. Part five and six is a presentation and analysis of the results obtained from our study and a discussion, and the final part, part seven explains the conclusions we draw from the results.

2 Theoretical framework

2.1 Efficient Market Hypothesis

The effective market hypothesis is fundamental in financial theory and assumes that the financial markets are efficient. A market that is efficient means that all available information is reflected in the share price. To predict future share prices, investors will implement the information that is available and change the share price to the new correct level. In order for the share price to be adjusted at the time of advertising, there must be unexpected information available. If the information is expected, the price will already contain the new information at the time of advertising (Bodie et al., 2009). If analysts possess knowledge that enables them to beat the market, it disputes the theory of the effective market hypothesis (Fama et al. 1979). There are three grounds for a market to be seen as effective. The first is rationality, which means that new information is released on the market and investors adjust their estimates of the price in a rational way. The second basis is deviations from rationality, which is

since

people do not always act rationally but that emotions can influence. The third basis is arbitrage which means that there are imbalances between valuations of assets in markets.

Since all available information is reflected in the price directly, the possibility of arbitrage wins is limited. Depending on how accessible the information is, the market’s efficiency can be divided into three different categories.

Weak

Weak market efficiency means that by studying historical share prices it is not possible to take advantage of risk-adjusted excess returns. By applying technical analysis, for example, risk-adjusted excess returns cannot be achieved (Fama, 1970).

Semi

The semi-strong form of market efficiency means that all available public information is taken into account in the share price i.e., annual reports and shared analysis.

Creating excess returns with technical or fundamental analysis is not possible under semi efficient market conditions. The only reliable way to generate excess returns is by obtaining insider information. Insider information is information that has not been announced to the public (Bodie et al, 2009).

Strong

The strong form of market efficiency means that all information, historical-, public- and insider information about a share is reflected in the price. Because the share price already reflects all the information that analysts could possibly have, they cannot add any value to investors. Therefore, the strong efficient market hypothesis cannot give investors an excess return (Bodie et al, 2009).

2.2 Adaptive Market Hypothesis

The theory is based on the basic assumptions within EMH but includes behavioral finance as an additional aspect. The AMH believes investors to be rational but that increased market volatility can lead to irrational behavior. Cognitive factors from the behavioral finance-field are applied to explain irrational behavior among investors.

Examples of these cognitive factors are that investors are loss averse and overstating their ability to generate profits (Lo, 2004). Investors are believed to adapt to events that affect market conditions and the cognitive factors described in behavioral finance may strengthen the argument that investors are not necessarily rational (Lo, 2004).

An example of this is that if an investor were to make an incorrect investment decision, he or she will most likely start to form an alternative investment strategy at a later time i.e., adapting their behavior (Lo, 2004).

2.3 Behavioral Finance

Behavioral finance is a field that uses cognitive factors to explain financial market behaviors that cannot be explained by traditional, economic or quantitative models (Ritter, 2003). Related aspects to the cognitive have to do with psychology, how human behavior can influence investment decisions and how restrictions on arbitrage for an investor can arise.

A common behavior in finance is loss aversion, meaning that people in general are more sensitive to losses than gains. In other

words,

investors are less likely to choose higher gains with high risk than lower gains and low risk (Benartzi and Thaler, 1995). Since investors are believed to be loss

averse,

they may also tend to sell “winners'' too early and hold on to “losers” for too long. For the winners this means that investors will be quick to secure a profit and sell the asset to avoid the risk of the profit diminishing or disappearing. For the losers the opposite is likely to happen, the investor will instead hold on to the asset for far too long in order to avoid realizing the loss (Shefrin and Statman, 1985). In theory this means that an investor who buys a stock based on a new positive consensus recommendation is less likely to hold on to the stock if the stock price increases a lot, since the investor wants to

“secure” the profit.

Another common

behavior

in finance is herding behavior. This refers to the instances where investors start to follow other individual investors or groups of investors, instead of relying on their own analysis they rely on the wisdom and knowledge of others. An investor who shows a herd instinct will therefore often attract the same, or similar, investments as other investors. This may either be because investors receive the same information as others around them, or because they start to imitate the investment strategies of more profitable investors (Jegadeesh and Kim, 2010).

One reason this behavior occurs is due to the fear of missing an opportunity to make money. The fear of missing a good investment opportunity can lead to impulsive actions and in the end, this can be a risky and wrong investment that one would not have made otherwise (DeMarzo et al., 2008).

Both investors and analysts may exhibit herding behavior. Analysts usually do not stray too far from the consensus recommendation of a stock, especially when downgrading. Analysts’ herding is sometimes said to introduce increased noise on markets. However, assuming that the EMH holds, markets will expect herding from analysts’ and prices will already have adjusted to the correct level when a revision is made (Jegadeesh and Kim, 2010).

2.4 Noise Trading

Noise traders are a group of investors who trade stocks independendent of the firm’s fundamentals or new information. In the literature they are usually described as uninformed traders who spend no time on research or due diligence and thus are unaware of the information investors usually require to assess stocks (Black, 1986).

They are however aware that market prices should reflect all current information about the firms and thus the market as a whole (Grossman, 1976).

Noise traders are what makes financial markets possible but also what makes them imperfect. Without noise trading, the trading of individual assets would be almost non-existent. People would still hold assets, such as stocks, but be unwilling to trade them. If one investor has information that makes them willing to sell a stock, the investor at the other side of the trade must have information that makes him or her willing to buy the stock. As such, one of them must have faulty information (Black, 1986). Noise traders solve this

market problem

by trading on noise they believe to be information. For the informed traders this creates an incentive, because they can now use their information and trade with uninformed investors, who as opposed to actual informed traders are willing to trade even though they, objectively, will be worse off (Black, 1986) Other benefits on markets include increased trade volumes, reduced ask-bid spread and a reduced price impact from trades. However, they also prevent market prices from

efficiently

adjusting to new information (Bloomfield et.

al, 2009). This noise effect may cause stock prices to drift away from the actual value, the farther away it drifts the faster it will adjust back. (Black, 1986).

2.5 Piggyback Theory

The piggyback theory suggests that analysts' recommendations often “piggyback” on other news or announcements from or about the company. When accounting for the effects of the news or announcement,

analysts’

recommendations have no measurable effect on the stock price and the reason why the price moves at all is because of the “piggybacked” event (Altinkilic and Hansen, 2009) If true, the piggyback theory would disprove a large number of previous studies on the

effects

analyst recommendations presented in the next chapter.

3 Literature Review

3.1 Stock Recommendations

In the article Discrete Expectational Data and Portfolio Performance, written by Edwin J. Elton, Martin J. Gruber and Seth Grossman (1986), the authors analyzed over 10 000 recommendations per month and found out that an investor will benefit by investing in new buys rather than sells, by as much as +4,5%. Dividing stocks into different portfolios with similar values for beta, based on receiving positive or negative updates to their recommendation level, they can observe how different changes affect the stocks. Their study differs from past studies, since Elton et al.

(1986) utilizes a much larger sample than what was commonly used prior to the release of their study and applies a more robust method when adjusting for risks. By removing the highest beta stocks from portfolios until the betas are within just 0.01 of each other, their procedure only assumes that beta is simply a measure of risk. They go on to conclude that there is indeed information value in

analysts’

recommendations and that buying from “buy-lists” can generate excess returns, however, not as great as the excess returns generated by acting on changes in recommendations.

Ten years later Womack (1996) found a similar pattern in the article Do Brokerage Analysts’ Recommendations Have Investment Value? By analyzing 14 brokerage firms he found that new buy recommendations had an immediate positive effect on the stock price, described as a

“short

lived modest drift” of +2,4%.

One interesting observation was that when a stock received a new sell recommendation, the effect was much larger and more long lived. The observed drift for the stocks receiving a sell-recommendation in Womack’s study was -9,1% and the effects of the new information were prevalent for six months. He also points out that analysts were more likely to issue buy than sell recommendations.

A key aspect of Womack’s study was the fact that only 9% of recommendations were issued in conjunction with quarterly earnings reports. He goes on to conclude that analyst recommendations have a significant effect on stock prices.

Womack’s results are put to test by Barber, Lehavy, McNichols and Trueman (2001 & 2003) who take a more “practical” approach by forming investment strategies based on the findings published by, among others, Womack (1996). Barber et al. (2001) found that if an investor employs a strategy based on

recommendations,

he or she can make a profit greater than 4%, if they invest in the analyst's most favored stocks compared to their least favored. They do, however, conclude that none of the investment strategies yield positive abnormal returns since they all require a lot of trading and thus high transaction costs. This means that none of the strategies reliably can generate abnormal net returns larger than zero. The authors also believe that the results show that the market efficiency is semi-strong, excluding transactions costs. It would not be possible to take advantage of abnormal risk-adjusted returns if strong market efficiency prevailed. This depends on that all available information is already included in the price.

In their follow up study from 2003 they can observe a negative correlation between analysts' most favored stocks and their least favored stocks in both 2000 and 2001. The most favored stocks by analysts significantly underperform, compared to the least favored stocks but also the index. They do however conclude that this could be a case of analysts doubling down on their positive view on small cap, high growth stocks which significantly underperformed the market in these years.

From the

groundwork

provided by the articles above the study by Jegadeesh, Kim, Krische and Lee (2004) examines when

analysts’

recommendations add value.

They note that when comparing value stocks, growth stocks and “glamor-stocks'' seems to be more popular among analysts. According to the results of the study, this is claimed to lead to “noise trading” in the financial market.

In short, noise trading means that actions such as selling and buying in the market are done irrationally. Parts of the study’s results also show that issued recommendations can be useful when considering different investment strategies.

They find similar results as the studies by Elton et al. (1986), Womack (1996) and Barber et al. (2001) which means that it can be of value for an investor to act on analysts' stock recommendations. Like the previous studies they also find that stocks favored by analysts tend to outperform the less favored stocks.

For a more “local” perspective on the Swedish stock market the dissertation by Lidén (2005) discusses the effects of stock recommendations in print media in Sweden between 1995 and 2000. The dissertation is mainly based on two economic theories called the price-pressure hypothesis and the information hypothesis. An interesting observation is that Lidén received advice from Brad Barber on how to conduct the event study used in the dissertation. Lidén finds evidence that journalist recommendations are much more influential than analyst recommendations. He also finds evidence that buy recommendations were more likely to be released on a weekday, while sell recommendations were more likely to be released on a weekend.

3.1.1 A Critical View on Recommendations

A critical view on the impact of recommendations is provided by Altinkilic and Hansen (2009). They

analyze

a number of studies, among others Elton et. al (1986) and Womack (1996) and claim that recommendations tend to “piggyback” on earnings announcements and firm specific news. When removing these effects, the recommendations do not produce any meaningful stock price reaction. Another study that puts analysts in a critical light, is the study Do Analysts Herd? An Analysis of Recommendations and Market Reactions by Jegadeesh and Kim (2010). The authors find that analysts tend to herd around the consensus estimate, especially when they are analysts at a big firm. They find evidence that the market reacts more strongly to new recommendations that are away from consensus. The question of the effect of recommendations is further debated by Loh and Stulz (2011) who found that just 12%

of recommendations can be considered influential given that they are from an influential analyst, accompanied by earnings forecasts from high growth firms.

The observation that they

must

be accompanied by an earnings forecast validates Altinkilic and Hansen’s piggyback theory. Similar to Jegadeesh and Kim (2010) they found that new recommendations away from consensus have the biggest impact.

Meaning that herding could potentially hurt analyst reliability.

3.1.2 The Aspect of Noise

In Grossman’s (1976) study On the Efficiency of Competitive Stock Markets Where Trades Have Diverse Information an insight is brought to light. If the market prices reflect all available information, there is no incentive to do any further research than to just observe prices. Grossman states that some form of noise is required for informed traders to hide information from uninformed traders. A further development of the noise theory is provided by Black (1986) in his article simply called Noise.

Black (1986) states that noise traders are what makes financial markets possible, but also what makes them imperfect. Without noise trading, the trading of individual assets would be almost non-existent. Black (1986) explains that people would still hold stocks, but there would not be any trade. Black (1986) also states that noise traders solve this market problem by trading on noise they believe to be information.

For the informed traders this creates an incentive, because they can now use their information and trade with uninformed investors, who as opposed to actual informed traders are willing to trade even though they, objectively, will be worse off.

4 Data and Research Method

4.1 Data

In order to answer our research question daily price and volume data was collected via a Thomson Reuters Eikon terminal for the 30 Swedish stocks in OMXS30 (as of Q1 2020, hence SSAB A for example will be included instead of Evolution Gaming), the full list of stocks can be found in Appendix 1. All of the data on prices, volumes, indices and analysts’ recommendations are obtained from the Thomson Reuters Eikon terminal with a daily frequency and then summarized weekly. This in order to make it easier for the reader and easier to track changes between weeks as well as within weeks. The week in which a change occurs will later become our event window.

Our data spans three full years (2017, 2018 and 2019) and two quarters, one on either side of the primary time period (Q4 2016 and Q1 2020).

This is done in order to avoid the market effects of the COVID-19 pandemic and the ability to obtain entry values for the time period.

We will use the all-share index OMXS PI as the benchmark for the economic climate in Sweden in each time period. The new recommendations will not be analyzed individually nor will the type of analysis used to produce the recommendation or be analyzed. We will instead use changes in the analyst consensus recommendation during the period and study how they correlate with abnormal stock returns. We will also touch briefly on mean target price changes and other variables related to analysts, but these will not be the primary focus in this thesis.

4.2 Research Method

This thesis applies a deductive approach, meaning that based on existing models and theories, we develop a hypothesis which is then tested. The chosen research strategy in this thesis is a quantitative analysis, more specifically an event study based on panel data. This is the most appropriate way to obtain an answer to our research question.

It means that we will use changes in analyst consensus recommendations as our events and the week where the event occurs as our event window.

Timeline 1. The timeline for the event study (in weeks). The estimation period is the weeks prior to a change that falls outside of the observation period of a previous change. The observation period for an event will always be

replaced by the observation period of a more current event in the case where two (-or more) events overlap.

We have taken inspiration from the methods and tests used by Womack (1996), Barber et al. (2001 & 2003), Jegadeesh et al. (2004) and Lidén (2005), in order to discover if and how analyst recommendations correlate with abnormal returns.

Like Lidén (2005) we have chosen abnormal returns based on a market model as our dependent variable, for the simple reason that it shows if returns for our sample are unusually high or low when compared to the expected returns.

The mean analyst consensus recommendation is calculated by using a weighted average of all outstanding recommendations for every stock during every week. Like Jegadeesh et al. (2004) we have assigned the recommendations a numerical weight on a five-point scale where 5 is the highest (strong buy recommendations) and 1 is the lowest (strong sell recommendations). The number is then divided with the total number of outstanding recommendations for the stock in the period, in this case during the week. For example, if a stock has three strong-buy recommendations, seven buy recommendations, ten hold recommendations and five sell recommendations, the analyst consensus recommendation for the stock would be 3.32¹. If the consensus recommendation for the stock in the example changes to 3.42 in the following week, there would be an increase of 3% in consensus compared to the previous week². Worth nothing is that we will focus solely on changes in analyst consensus and not the actual level of the aforementioned analyst consensus stock recommendation.

Since we have a large number of analyst changes with multiple overlapping events, we have decided to do the following; the changes in analyst recommendations happen in week t=0, this will be our event window. The estimation period, seen in Timeline 1, is used to estimate performance during the weeks where a change is not present, and the week is not within a 4-week period after the latest change. The observation period consists of the event window and evaluation period and is used to evaluate the returns in week t=0 to week t=4. If a change takes place in t=0 week and a new change happens in week t=1, the more current change will override the previous period. For example, in week 8 the analyst consensus recommendation for stock A increased, thus opening the event window. In week 9 there is no change in the consensus, so the evaluation period of the event in week 8 is still ongoing.

1A specification of the equation is presented in section 4.2.3.

2A specification of the equation is presented in section 4.2.3.

However, in week 10 the analyst consensus recommendation decreases for stock A, this means that a new event window and evaluation period begins, replacing the evaluation period of the change that happened in week 8.

In order to find out if the change in analyst consensus recommendations has a significant effect, this thesis will conduct a number of statistical tests to determine the relationship between abnormal returns and changes in analyst consensus recommendations for the chosen stocks. First, we will use a two tailed t-test to see if mean abnormal returns, when changes are positive or negative, are significantly different from the mean returns of the index. This test will be conducted for both abnormal returns and cumulative abnormal returns. This should give us a first clue of whether or not the changes actually produce a reaction.

Secondly, we will conduct a bivariate analysis to see if we can observe a statistically significant correlation between abnormal returns and the changes in analyst consensus recommendations. A bivariate analysis is a test of correlation between two variables. In this thesis we have chosen to test for correlation with the Spearman rank correlation-test, described and motivated in detail in the next section.

When the data from the tests is obtained, we can assess if (positive) changes in analysts’ recommendations indeed show signs of (negative) correlation for our chosen sample.

Third, we will conduct a multiple linear regression analysis with abnormal returns as our dependent variable in order to see if and how much of its variation is explained by the changes in analyst consensus recommendations, and if so at which significance level(s). Although we focus on recommendations, we will include a few other variables used in previous studies that are believed to be relevant.

4.2.1 Significance Test

In order to discover if the mean abnormal returns and the mean cumulative abnormal returns of the chosen stocks are statistically different from the mean returns of the

index,

we will conduct a two tailed t-test. In this thesis we have gone for a two tailed t-test which assumes that the two groups are unpaired and have unequal variances.

The null hypothesis is that the two samples have a mean difference of zero, and the equation is formulated as following;

H0: The two samples have a mean difference of 0 H1: The two samples do not have mean difference of 0

Where

𝑥

₁ denotes the mean of the first group and

𝑥

₂ denotes the mean of the second group. The variables

𝑠

₁² and

𝑠

₂² denotes the standard deviation of group 1 and 2 respectively,

𝑛

₁ and

𝑛

₂ denotes the sample size in each of the two groups. The tests will show if there is a statistically significant difference and at which level, meaning that if the p-value is less than significance level, there is a significant possibility that the null hypothesis (H0) is untrue. However, it does not automatically mean that the alternative hypothesis (H1) is true, nor does it mean that H0 is automatically true if we

cannot

reject it.

4.2.2 Spearman Rank Correlation

Like Jegadeesh et al. (2004) we have opted to use The Spearman rank correlation-test as a test for correlations. The Spearman rank-test allows us to test for monotonic relationships between variables, in this case between the abnormal return and changes in analyst consensus stock recommendations. We will run the test with all stocks as one group but with changing conditions, so we will look at each year individually and if there is a report or not. The Spearman rank correlation-test is often used if there is a monotonic relationship and no linear relationship between the variables, this is however not required to run the test (Laerd Statistics, 2018).

The difference between a monotonic relationship and a linear relationship is that in a monotonic relationship the two variables do not have to move in the same direction at a constant rate like in a linear relationship, so it imposes fever restrictions on the relationship.

The coefficient, rho, obtained from the Spearman-test takes a value between -1 and +1, where -1 indicates a perfect negative relationship and +1 indicates a perfect positive relationship between the variables (Laerd Statistics, 2018). The hypothesis test of the Spearman rank correlation-test in this thesis is formulated by STATA as:

H0: Variable A is independent of Variable B H1: Variable A is not independent of Variable B

The test shows if the correlation coefficient between the chosen variables is statistically significant and at which significance level. Meaning the same thing as described in the previous section 4.2.1.

4.2.3 Calculations and Variables

This section provides a presentation of the equations used to calculate the variables, and a presentation of said variables, that will be used in this thesis.

Realized return³: 𝑅_'( =^*+,-'./_*+,-'./^{012*+,-'./0134}

0134

Realized return for stock i is the difference between closing price on the last trading day in week t and closing price on the last trading day in week t-1, in relation to the closing price on the last trading day in week t-1. It gives us a percentage return for the stock(s) and OMXS PI index.

Abnormal return: 𝐴𝑅_'( = 𝑅_'(− 𝛼_'(+ 𝛽_'𝑅_:(

Abnormal return is based on the market model used by Lidén (2005), where 𝐴𝑅_'( denotes the abnormal return for stock i in week t. 𝑅_'( denotes the stocks realised return in week t and 𝑅_:( denotes the market’s return in week t.

3Used for both stocks and OMXS PI, the return for OMXS PI is denoted Rmt.

The values for a (intercept) and b (slope) are estimated through running an OLS regression using the stocks realized return and the markets return. This provides us with the abnormal return for the stock given its level of risk. The mean abnormal return 𝐴𝑅₍ is simply the abnormal return in relation to the number of observations in period t. The mean cumulative abnormal return 𝐶𝐴𝑅 is the sum of the mean abnormal returns between two points in time. In this thesis it will be measured each week for the 4 weeks following a change in analyst consensus.

Change in volume: ∆𝑉_'( =^?01_?^2?0134

0134

The change in volume is calculated the same way as the realized return. The change in volume for stock i is the difference between the volume in period t and volume in period t-1 in relation to the volume in period t-1. This will provide us with a percentage change.

Abnormal volume: 𝐴𝑉_'( = 𝑉_'( − 𝛼_'(+ 𝛽_'𝑉_:(

Abnormal volume is, as the abnormal return, based on the same model used by Lidén (2005), where 𝐴𝑉_'( denotes the abnormal volume for stock i in week t. 𝑉_'( denotes the stocks trade volume in week t and 𝑉_:( denotes the market’s trade volume in week t.

The values for a (intercept) and b (slope) are estimated through running an OLS regression using the stocks trade volume and the markets trade volume. The mean abnormal volume 𝐴𝑉₍ is the abnormal volume in relation to the number of observations in period t.

Analysts change variables

Analyst consensus recommendation:

𝐶𝑜𝑛_'(= 𝑁_B.DEF'( 5 + 𝑁_HEF'( 4 + 𝑁_J,+K'( 3 + 𝑁_BM++'( 2 + 𝑁_B.-M++'( 1 𝑁_NMO'(

The analyst consensus recommendation is the weighted average of all outstanding recommendations for stock i in period t. A recommendation is assigned a weight between 1 and 5, depending on their level.

Then multiplied by the number of recommendations in that category, the total is then divided by the total number of outstanding recommendations for stock i in period t.

Change analyst consensus recommendation:

∆𝐶𝑜𝑛_'( =𝐶𝑜𝑛_'(− 𝐶𝑜𝑛_'(2P 𝐶𝑜𝑛_'(2P

The analyst consensus recommendation changes for stock i is the difference between consensus in period t and period t-1 analyst consensus recommendation in relation to period t-1 analyst consensus recommendation. As mentioned before, this variable is the main focus in this thesis. Mean analyst consensus recommendation change is the sum of the changes in relation to the number of observations.

Change analyst mean target price:

∆𝑇𝑅𝐺_'( =𝑇𝑅𝐺_'(− 𝑇𝑅𝐺_'(2P 𝑇𝑅𝐺_'(2P

The mean analyst target price change for stock i is the difference between mean target in period t and period t-1 mean analyst target price in relation to period t-1 mean analyst target price. The change in the analyst mean target price is calculated in order to include it in the regression analysis, to see if it can offer any additional explanatory power.

Change total number of analyst recommendations:

∆𝑁_*,.= 𝑁_*,.1− 𝑁_*,.134

The change in the total number of analyst recommendations is simply the difference between the total number of analyst recommendations in the current period and the previous period. The change in the analyst mean target price is calculated in order to include it in the regression analysis.

As was the case for the change in the analyst mean target price, the change in the total number of analyst recommendations is calculated in order to include it in the regression analysis, to see if it can offer any additional explanatory power.

STATA variables

Table 1. Variables used in STATA, with hypothesised relationship to the dependent variable.

Table 1 above presents the variables that will be used in this thesis in order to run our significance test, Spearman rank correlation-test and our regression analysis. The first two tests will only be between abnormal returns and the analyst consensus recommendation change. The regression analysis will include all variables. The second left most column is the abbreviation for the variable when analyzed in STATA. The center column is the name of the variable as it appears in the section above and the rightmost column shows the hypothesized relationship with the dependent variable. For example, we hypothesize that an increase in volume will have a negative relationship with the dependent variable abnormal return, as stated by Jegadeesh et al. (2004). If an annual report and/or interim report is released, we expect it to have a positive relation to abnormal returns. We expect both mean analyst consensus changes and mean target price changes to be negatively related to abnormal returns, since a number of investors may try to “make a quick” buck on recently upgraded stocks, thus hinder it from rising and send it the other way. The change in the total number of analyst recommendations is hypothesized to be positive since an increased number of analysts studying a stock might, at least in theory, lead to better estimates and may give legitimacy to the stock. We also hypothesize that positive market returns will provide a better basis for stocks to generate abnormal returns.

We have chosen the variables based both on the intuition that they are all the most relevant in explaining abnormal returns in the immediate to short run, but also since the previous studies used as a base for this thesis use these variables or some form of variation of them. Abnormal return and abnormal volume is used by for instance Lidén (2005) in determining if stock recommendations generate price reactions, however our abnormal volume is based on the change in the total number of trades rather than turnover, this is discussed in further detail in section 4.2.4.

Jeegadesh et al. (2004) use volume changes as an indicator of momentum and to determine an expected direction of returns, since they state that high volume stocks should correspond with lower returns. They also use the variable analyst consensus recommendation changes to track if consensus changes have an effect rather than just the consensus level.

The variable mean target change isn’t commonly used in the studies that serves as a base for our thesis, however we felt it was relevant to include it in the regression analysis since target prices often seem to accompany a new or repeated recommendation. The change in the total number of recommendations is often mentioned in previous works but is rarely included as an explanatory variable in the models. Again, we felt that it would be relevant to analyze this effect to some extent as well in our regression. The variable will only change in positive or negative integers, meaning that it does not measure a percentage change. As previously stated, an increased number of analysts studying a stock might, at least in theory, lead to better estimates and may give legitimacy to the stock. The market return variable serves as our benchmark during the period and may provide us with information about why the stocks might not react as expected to firm specific phenomenon’s. Based on the variables presented in this section we have formulated a multiple linear regression for the regression analysis as follows:

4.2.4 A Critical View on the Method and Data

The quantitative method chosen offers a possibility to mathematically assess the relationship between abnormal returns and changes in analysts’ recommendations, we will touch briefly on mean target prices as well, but it will not be of focus in this thesis. Event studies are widely used for this type of study, see for instance Womack (1996), Barber et al. (2001 & 2003), Jegadeesh et al. (2004) and Lidén (2005). Using the Spearman rank correlation-test to assess correlation is also a previously used method, see Jegadeesh et. al (2004). We have also opted to use a t-test to see if mean abnormal returns differ from each other and the index under different scenarios, which is a widely used practice.

Since we have opted to calculate our analyst consensus recommendation based on a weighted average of all outstanding recommendations in the period, and then calculating a percentage change either up or down, and not a dummy variable for increase or no increase, our results may differ from previous studies. However, the method using a weighted average of recommendations to calculate changes in consensus is used by Jegadeesh et al. (2004).

A key difference to previous studies is that our estimation-, event- and evaluation windows are constructed in a different manner. Our estimation period is simply the weeks where no change is present, and the week falls outside the 4-week evaluation period of a previous change. When two (-or more) events overlap, we have chosen to use the evaluation period of the more recent event and the 4 weeks following that event, unless a new event occurs in that time, then the procedure is repeated. This is described in more detail in previous sections. Since we have accounted for both negative and positive changes with a large number of overlapping events, this could cause our estimation windows to be significantly shorter than in previous studies, giving us fewer observations to base our estimation of normal performance on. This may lead to different results from previous studies. Transaction costs are not accounted for in this thesis, this could provide us with differing results from other studies.

The data obtained from the Thomson Reuters Eikon terminal might also be constructed in a different manner than previous and other studies which could give differing results.

The chosen time periods are different from previous works and might return different results. Furthermore, our sample contains a number of 5 490 observations over three full years and two quarters and the number of observations may differ from the previously mentioned studies which also might cause our results to differ.

Since annual reports are few and far between, there were only a handful of observations in this category which limits our ability to get meaningful results and draw conclusions. Interim reports are more frequent, but this category also suffers from few observations.

Our variable for volume is based on the average number of stocks traded each week rather than the average value of the traded stocks. Which could potentially provide us with different results from previous studies. Furthermore, the regressions could potentially suffer from omitted variable bias, since there could be relevant variables which affect the outcome of the regressions that are not accounted for in our thesis. There may also be non-linearity and/or correlation between independent variables which could bias the results.

5 Empirical Results

5.1 Descriptive Statistics

In table 2 below our summary statistics of the data sample can be observed. The table shows the mean values for our sample in the first column as all periods combined and the following column presents the mean values each year.

Table 2. Summary statistics from STATA for entire sample

The analyst changes in Table 2 were on average small, with the average analyst consensus change only being 0.0004. The only year it took on a negative value was in 2017 with an average of -0.0012, meaning that during 2017 the average change in analysts’ consensus recommendation was negative. Mean target price was on average changed in bigger steps than the average consensus change. The change in the total number of recommendations is expected to be larger since it only moves in positive or negative integers. The biggest change was in 2019 when the average change in the total number of recommendations was -0.0468.

There was a total of 5 490 observations meaning that the stocks were each measured on average over 183 weeks. However, the total number of observations suggests that Essity B also was observed over a total of 183 weeks, this is not the case, the stock only has 147 observations in total. This is accounted for in the tests and analysis later on, where 2017 returns an accumulated 1537 weeks for our sample.

Table 3. Total number of observed analyst changes in consensus recommendation for all stocks.

Table 3 shows our total number of observed changes in analyst consensus recommendations since it is the main focus on this thesis. As the table shows, there were a total of 2 416 changes in analyst consensus recommendations for the entire sample across all periods. This means that there were on average 0.44 changes in analysts’ consensus recommendations per week. The changes overall were slightly positive at 51.7384% positive changes and 48.2616% negative changes. This should indicate that analysts during the analyzed period were somewhat more likely to upgrade a stocks recommendation in our sample, than to downgrade it.

5.2 Abnormal Returns

The difference between mean abnormal returns and the mean returns of OMXS PI are presented below. Table 4 only includes the mean difference between the groups and a more detailed presentation can be found in table 1 and 2 in appendix 2.

Table 4. Two tailed t-test of difference between mean abnormal returns for the stocks and the mean return for OMXS PI during our observation window. Unpaired and with unequal variances. 2020 and 2016 only contain Q1

and Q4 respectively.

As shown in table 4 the mean abnormal returns of our stocks shows significant mean difference to the index in the event week and 3 weeks after the analyst recommendation change when looking at the left most column. The fact that the mean difference is negative in the event week may indicate that the stocks on average underperformed the index in the week where they were upgraded. The same thing can be found in 2020 where the mean difference in returns between the stocks and OMXS PI was almost -4%, followed by a mean difference of -2,34% in the first week after the analyst consensus recommendation was changed. We are not surprised by this, and we will discuss why this happened later on. A somewhat similar pattern can be found for the weeks following a decrease. For the most part, the weeks after a decrease did not exhibit much of a statistically significant difference, on any level, to the returns of the index. The only case where this happened was in week 3 in 2018.

5.2.1 Cumulative Abnormal Returns

Mean cumulative returns are summarized from the mean abnormal returns. The mean cumulative abnormal returns for the stocks and the mean cumulative returns of OMXS PI are presented in a slightly different manner to before, the table does not include the difference between the two. The mean cumulative (abnormal) returns for the stocks and the index is presented in Table 5, a more detailed presentation can be found in table 3 in appendix 3.

Table 5. Two tailed t-test of mean cumulative abnormal returns for the stocks and the mean cumulative return for OMXS PI during all of our observation windows of an increase or decrease. Unpaired samples and with unequal

sample variances.

Looking at the summary of all years combined in table 5, one can see that only the mean cumulative abnormal returns following a decrease in analyst consensus recommendation shows a statistically significant difference from the mean cumulative returns of the index. On average the stocks generated a mean cumulative abnormal return very close to zero after 4 weeks following a decrease, with the biggest changes happening in week 3. The weeks following an increase did not return a mean cumulative abnormal return statistically different from the mean cumulative return of OMXS PI. As can be observed from graph 1 below, the cumulative abnormal return for the stocks seems to almost move exactly the opposite to the cumulative return of the index. When seeing the lines in graph 1 it becomes clearer why it should lead to an overall difference between the mean values that is not statistically different from 0.

Graph 1. Shows the mean cumulative abnormal returns for the stocks from week 0 to week 4 and OMXS PI mean cumulative returns, when there is an increase.

The stocks in graph 1 seem to generate a higher total mean cumulative abnormal return than the mean cumulative returns of the index when comparing the results in week 4. If we compare it to the mean cumulative abnormal returns from graph 2, following the negative changes to analyst consensus recommendations, it seems to indicate a sharper increase in returns following the initial decline.

Graph 2. Shows the mean cumulative abnormal returns for the stocks from week 0 to week 4 and mean cumulative returns of OMXS PI, when there is a decrease.

Turning to graph 2 we can see a visualization of the mean cumulative abnormal returns of the stocks and the mean cumulative returns index in the weeks following a decrease in analyst consensus recommendation. Instead of exhibiting a pattern like in graph 1, that seemed like a mirror opposite to the index, the stocks seem to follow the general movements of the index more closely. As one can see, there are a few differences, most obvious is in week 2 following a decrease, where the stocks seem to show an increase in returns compared to both the previous week and the index, which shows a pattern of a decline in the week. After a sharp decline in week 3, there seems to be a rather steep incline in the last week.

5.3 Bivariate Analysis

In order to check for correlation between abnormal returns and the analyst change variables we conducted a number of Spearman rank correlation-tests with different time frames and scenarios.

Table 6 below shows the obtained rho-value, number of observations and p- value for the Spearman Rank Correlation-test for the entire sample for each time period as well as all periods combined. The null hypothesis is that abnormal return is independent of changes in analyst consensus recommendations.

The decision to accept or reject the null hypothesis is given in the left most column and the stars show at which significance level the null hypothesis is rejected.

Table 6. Spearman rank correlation coefficient-test between abnormal return and the observation periods of changes analyst consensus recommendations for all stocks, with increase and decrease. 2020 and 2016 only contain Q1 and

Q4 respectively.

Overall, from what can be seen from table 6, the statistically significant correlations between abnormal returns and our change-variable are very weak. As the rho-values indicate, there is only a statistically significant relationship between the variables in two of the years when a positive change happens, 2020 and 2017. The only negative correlation between the abnormal returns and a positive change in analyst consensus recommendations happens in 2020 (Q1), this may be expected however, we will discuss why later on. When there is a negative change 2017 and the combined period exhibits a statistically significant correlation, albeit both very weak.

Table 7 shows the results from the Spearman rank correlation-test when we only include observation periods where an annual or interim report was made public or exclude all observation periods where an annual or interim report was made public.

The first column only shows correlations between abnormal return and our change- variables during observation periods where an annual report was released. The second column only shows observation periods where an interim report was released and the third and fourth column excludes observation periods where annual and interim reports were made public from the sample.

Table 7. Spearman rank correlation test between abnormal return and changes in analyst consensus recommendations, with regards to the disclosure of annual and interim earnings reports.

For the entire sample during observation periods where annual reports are released and there is a decrease in analyst consensus recommendation, there is a moderate correlation of 0.5131 between abnormal returns and changes in analyst consensus recommendations. There is also a very faint correlation for the sample when no report is released and there is a decrease in analyst consensus recommendation. The relationships are positive, indicating that abnormal returns and changes in analyst consensus should move in the same direction, for example a negative change in analyst consensus recommendation correlates with lower abnormal returns in that period.

5.4 Regression Analysis

The point of the regression analysis is to analyze how much and to what degree the changes in the dependent variable are explained by the independent variables. A total of 5 454 observations were obtained from our regression analysis. Table 8 below shows the results from our regression with abnormal return as the dependent variable.

The table shows that for all stocks across the entire period, the variables abnormal volume, changes in mean target price, changes in the total number of recommendations, and interim reports all have a significant effect on the changes in abnormal return, at the one percent significance level.

Table 8. Regression analysis, all stocks during all periods

When looking at all periods combined, we find no evidence that changes in analyst consensus recommendation explains the changes abnormal returns. The only years where we can find some evidence that the changes in abnormal returns are explained by changes in analyst consensus recommendations is in 2019 and 2017. Abnormal volume shows a significant effect on the changes in abnormal returns at the one percent significance level, when we look at the leftmost column. We also find that a change in mean target price of one unit corresponds to a 0.172 unit change in abnormal returns and a one unit change in the total number of analyst recommendations corresponds to a -0.004 unit change in abnormal returns, when looking at all periods. Both are statistically significant at the one percent significance level. Interestingly, interim reports show a statistically significant effect on the variation of abnormal return at the one percent level, despite it not being correlated with abnormal returns according to the Spearman rank correlation-test.

The R-squared value is 0.017 for the entire sample across all periods, meaning that only 1,7% of the variation in the dependent variable abnormal returns is explained by the independent variables.

5.4.1 Positive Change

When running the regression again but only including positive changes to analyst consensus recommendations, the results in Table 9 below were obtained.

Table 9. Regression results when only observation periods of a positive change in analyst consensus recommendations are present

One can observe that only two variables are statistically significant on the one percent significance level in the leftmost column, mean target price change and interim reports. It seems that mean target changes play an even bigger role in explaining variations in abnormal returns when only positive changes in analyst consensus recommendations are present. A change in mean target price with one unit now corresponds to a 0.210 unit change in abnormal returns, when looking at all periods.

In 2020 a change of one unit in mean target price corresponded with a 0.697 change of abnormal returns. Looking at 2017 we find that a change in analyst consensus recommendations of one unit correspond to a -0.177 unit change in abnormal returns, almost the exact opposite of what could be found in table 8. Interim reports again show a statistically significant effect on the changes in abnormal return at the one percent significance level. Abnormal volume shows no significant effect on the change in abnormal returns, except for in 2019.

R2 is higher than in table 8, now 2,6% of abnormal returns are explained by the independent variables as opposed to 1,7% before. Meaning that our independent variables are slightly better at explaining changes in the dependent variable than in the previous regression.

5.4.2 Negative Change

Running the regression yet again when there is a negative change in analyst consensus recommendations yields the following results shown in Table 10 below. There are a few differences to before.

Table 10. Regression results when only observation periods of a negative change in analyst consensus recommendations are present

One can observe that now all of the analysts change variables as well as abnormal volume show a statistically significant effect on the variations in abnormal returns at the one percent significance level. In contrast to Table 9 it now seems that volume change plays a statistically significant role in explaining changes in abnormal returns yet again. Like in the previous regressions, changes in mean target price continue to offer statistically significant explanations for the variations in all periods, 2020, 2019 and 2016. Interim reports are now statistically insignificant when looking at all the years combined. When studying the years individually one can observe that interim reports are statistically significant at the ten percent significance level in 2019 and the one percent significance level in 2018.

R2 is higher than in both Table 9 and 10, now 3,3% of abnormal returns are explained by the independent variables as opposed to 1,7% and 2,6% respectively before. Meaning that our independent variables are slightly better at explaining changes in the dependent variable than in the previous regression.

6 Discussion

The main goal of this thesis is to see whether or not positive changes in analyst consensus recommendations correlate with lower abnormal returns i.e., a negative relationship. Our descriptive statistics show that overall analyst changes were on average small, the average changes in consensus recommendation especially. The number however can be somewhat expected since it displays the mean change of the mean recommendation level, but it is far smaller than the average percentage change of the mean target price. Maybe analysts regard a recommendation change as a bigger step than when adjusting the target price to reflect new information. When something substantial effects the stock analysts can use a recommendation change to reflect more important new information. Other reasons why there are small averages in analyst changes this may be due to; a) analysts are in fact exhibiting herding behavior and do not usually stray “too far” away from consensus, b) analysts does not react as fast to changing market conditions as investors, or c) the recommendation and/or target price is based on a long-term forecast and thus will not be adjusted unless an event is believed to have long term effects on the stock.

The t-test shows a significant difference in some of the cases for both mean abnormal returns and cumulative abnormal returns. Like previous research has shown, this means that there is evidence that changes in analyst consensus recommendations can generate a reaction which affects the stocks returns. According to the efficient market hypothesis, this should mean that markets are imperfect and there may be differences in the information that traders possess. The adaptive market hypothesis however, states that market efficiency varies, and certain periods may exhibit less efficiency than others, so this may be expected to happen if we are in a less efficient period.