Financial Magazines impact on the Swedish Stock Market: An event study

(1)

Bachelor Thesis

Financial magazines impact on the Swedish stock market An event study

Authors:

Philip Hausenkamph, 19960204 ph222ph@student.lnu.se

Gusten Hansson, 19931101 gh222fe@student.lnu.se Supervisor: Magnus Willesson Examinator: Håkan Locking Date: Spring 2019

Subject: Finance Level: Bachelor Course code: 2FE32E

(2)

Abstract

The purpose of this study is to investigate the effect of a stock recommendation from the leading financial magazines in Sweden. The study aims to measure the impact a

recommendation illustrates in true value. The measurements are mean abnormal returns (AR), mean cumulative abnormal returns (CAR) and mean abnormal volume (AV).

Conducting an event study to monitor, not only the date of announcement, but to also validate or invalidate the recommendation as a fundamental changer in the stock case. Where the calculations are made before, on and after the event occurs. With the aim to test if the market is efficient and in line with the rational theories, or if there are other explanatory theories, like the behavioral financial approach, that can explain the results. The sample consists of 571 recommendations that have been announced 2017 and 2018, divided into categories of buy and sell. The sample of buy and sell are also tested in subcategories of small and large companies, to measure the impact due to size of the firm, as a dependent variable. The empirical results shows that there are AR and AV existing due to recommendations. Small companies have the highest measured AR, with sell recommendations having the largest effects. The sell recommendations changes the value and the fundamentals of the stocks, while buy recommendations react positive to the recommendations on the day of

announcement, then reverses back to the same price in the end of the event window.

Suggesting that the market act both efficient and rational, but also irrational and ineffective, depending on what type of recommendation that is being released and how large and well monitored the company, that gets the recommendation is.

Keywords

Buy- and Sell recommendation, Event study, Efficient market, Overreaction, Behaviour finance

Thanks

We would like to take the opportunity to thank our supervisor Magnus Willeson for guidance through the whole work. Also Maziar Sahamkhadam for guiding us through the calculations and Håkan Locking for the help with STATA.

(3)

1. Introduction

This chapter aims to create an interest and an understanding of the topic we deal with in the thesis. The reader will be introduced to how the thesis is structured, the procedure and what the main focus is. Here we present the previous studies, the issues that we are set to

investigate and the hypothesis, which will then form the basis of this paper.

1.1 Background

With an option of over a thousand of stocks available on the market for purchasing, how do we choose the correct ones to select? The implicit decision in which the investor has to rank the huge amount of stocks against each other, bound by the rationality, makes the process of narrowing somewhat complicated. It's easier to process the selection when narrowing down the potential stock picking to 10 stocks, rather than over a thousand (Barber and Odean, 2007). According to Barber and Odean (2007) the process of stock picking, or narrowing down the alternatives, are coming from the cognitive mindset. We want to buy the stocks that have recently caught our attention, but the buyers do not engage a transaction on all the stocks that catch our attention. The second part of the decision making process is the behavioral part, the personal preferences and strategies, leades the selection of resource allocation among the stocks that are on our mind (Barber and Odean, 2007).

The selection approach of the individual investor can also be derived from the behavioural perspective of “following the herd” (Nofsinger and Sias, 1999).

The herding is derived from personal preferences in anchoring of information sources, if investors follow the same sources of information, the herding might occur. If the information, or news, are given to much value from the investors, the herding might lead to overreaction (Nofsinger and Sias, 1999). An example of the herding behaviour can be read in Huberman and Regev (2001) article. In May 1998 New York Times wrote a story about a company called ENMD, who might have come up with a cure for cancer. The stock rushed with over

(6)

300% after the publishing of an article in New York Times. This information had already been released over five months before the article occured in New York Times. The article spread optimism on the market like a contagion. This Contagion also spread to the whole sector, and created reactions in stocks traded in the biotechnology sector (Huberman and Regev, 2001).

So how should the stock price reflect the firm value, and can a recommendation change the the fundamental case of the stock, or is it just an overreaction? The price pressure hypothesis suggest that abnormal return, or overreaction, is caused by instinctive investors buy pressure.

While the information hypothesis suggest that new information about the stock presented, is what reflects the fluctuation (Barber and Loeffler, 1993). “The efficient market hypothesis”, however, states that the price of the asset should be reflected by the information available on the market. The anomaly short term perspective will be adjusted on the long term fair price of the asset. Which suggest that there is an over/under reaction on a short term, but on the long term perspective, the asset will move towards the fair price (Fama, 1998).

1.2 Previous Studies

Previous studies made in similar, comparative ares, inside of the topic gives a variation of results. Keasler and Mcneil (2010) does a study on based on the tv-show Mad money, that is televised in the US, with a viewers number of around 400 000 viewers a day. The result showed that abnormal return and volume on recommended stocks were significant on the daily basis, with a turnaround to the normal price level on a 25-day basis, for both sell and buy recommendations. With an exception of sell recommendations on stocks in small cap (Keasler and Mcneil, 2010). Which is in the same conclusionary line as Barber and Loeffler (1993) who studied the abnormal return of stocks recommended in the Financial Journal.

What Barber and Loeffler also concluded, as a possible underlying event, was the herding behaviour as a factor of abnormal return. The conclusion of the article made by Barber and Odean (2007) show that investors tend to buy on recommendation, but not sell on the sell recommendation (Barber and Odean, 2007) which is differentiating the results from the

(7)

article written by Keasler and McNeil (2010). Also Tetlock (2007) interpret the result from Media content to be a significant variable in explaining abnormal return and trading volym (Tetlock, 2007). Engelberg and Parsons (2011) conclude how local media reporting is significant for the local trading and suggests that this also would be applicable for global perspective (Engelberg and Parsons, 2011). Liu et al. (1990), Sant and Zaman (1996),

examines the recommendation section from the Wall street journal and conclude that there is significance between stock recommendations and abnormal return. There is a reversal in the stock return, back to normal levels, which support the price pressure hypothesis Liu et al.

(1990) concludes. While Sant and Zaman (1996) could see a correlation between reaction to the recommendation and number of analysis covering the security. The abnormal return gets lower, the higher number of analysis following the company, supporting the information hypothesis (Sant and Zaman, 1996). While Yazici and Muradoglu (2002) focus on the information potentially being leaked before the recommendation announcement. Suggesting that the information might be shared in advance, with the purpose of making profit.

Concluding that the recommendations might not be set to widen the information about a stock to the individual investor, but instead being misleading to favour the “prefered investors” in making profits. Furthermore their conclusion is that the recommendations only favour the

“prefered investors”, and for the individual investor, they suggest the buy and hold strategy, that would yield far more than trading on recommendations (Yazici and Muradoglu, 2002).

1.3 Purpose

This paper is set to investigate if the leading financial magazines in Sweden has any impact on the stock market. This in the area of the recommendation section, that is frequently used in every stock related newspaper. How does a stock react to a recommendation? That is the fundamental approach that is going to be examined through this whole paper. Are there differences in sell and buy recommendations, small and large companies? Do the investors act rationally and trade in the same direction as the recommendations, or is the information just interpreted as noise? The Paper will distinguish and categorize companies based on their

(8)

size and the type of recommendations, this to evaluate the true power of the recommendations, and measure the differences in impact.

With a changing set to interpret information, due to the digitalization expandation and width, the information is more easily available than it has ever been before. A recommendation reaches the market within seconds, in the flash feed after its being posted, and how does the market interpret the recommendation information feed. The purpose is to investigate whether a recommendation changes the fundamental case of the stock, or if it's just a reaction, due to information being presented by a legitimate source.

There has been some previous studies trying to calculate the effect of the recommendations on the swedish stock market. What has not been investigated before, to our knowledge, is the split of companies into groups based on the size, for investigation and measurement of the impact of recommendation based on the size of the company.

1.4 Issue and Hypothesis

We have two question formulations and hypotheses that will be tested in this paper.

The first one is whether the magazines recommendations on stocks have any effect on the price and volym, of the traded stock?

The magazines will be able to have more effect on their recommendations in small insolvent companies, then huge established firms that are traded on the bigger indices, which support the information hypothesis.

Our second question is whether the recommendations have any substance in the fundamental stock case, or if the reactions to the recommendations, solely is based on the fact that there is a recommendation cited from a legitimate source?

The reactions from a recommendation are overreactions, and the stocks will reverse to their true value, according to the efficient market hypothesis and the price pressure hypothesis.

(9)

This will be tested by calculating the mean abnormal returns and the mean cumulative abnormal returns, to measure the power of the recommendation in a value. Also the mean cumulative abnormal returns will conclude whether the recommendation has had any effect in the event window. The mean abnormal volume will tell if the stock is being traded with a higher frequency than usual, which will gives an understanding of the power of the recommendations.

1.5 Delimitations and Limitations

The limitation of this paper will be to the Swedish stock market. This as a result of the data collection being from the leading financial magazines in Sweden, since the Swedish stock market is the focus. The papers main focus will be the individual investors, as they are the focus group of the magazines recommendations, and as the assumptions are that the institutional investors already have the “recommendation information” printed into their analysis of the valuation of the stock.

1.6 Structure

The first section is the background and previous studies, which will give an understanding of how we motivate and produce our questions and hypothesis. The second section is a theory section that deal with the important behavioral and financial theories that are the pillars of this paper, but will also serve as a basis for analyzing our empirical results. The next section is methodology where the approach will be explained, how a event study works and why we chose this particular method for this paper. In this section, the data description also going to be presented. The reader will get an insight into the three different magazines the will be investigated. The fourth section will present empirical result, this through both graphs and shorter text that will give the reader a better understanding. Next chapter contain analysis that are based on the results from previous chapter. Finally, there will be a summary which will emphasize the conclusions that has been reached in the paper. The thesis is completed with the sources we retrieved the information from.

(10)

2. Theoretical Framework

In this section, presented the theories that are considered important to the reader in order to understand the underlying factors, and why different things happen and behave in their ways.

Theories that we introduce are The efficient market hypothesis, Price pressure hypothesis and Information hypothesis, which are classic and rational theories that explain why the market acts as it does. We have also chosen to include Keynesian beauty contest and Herd behavior as theories, since not everything can be explained by rational theories. These theories are instead about psychological factors, how people behave in relation to each other and how they act. Those five theories are discussed later in the thesis, in the parts where the result is analyzed and then also in the summary.

2.1 The Efficient Market Hypothesis

The efficient market hypothesis is a theory based on perfect capital markets, where there are several assumptions, about market conditions, to be taking into consideration. The transaction costs for trading with securities are assumed being zero. The information is free and available for all market operators. All the market participants have the insight that the price of the security, should be reflected by the information available on the market. Which conclude that the price of the security itself, can be motivated and reflected by the current information available to the public (Fama, 1970). Fama (1998) puts the theory of the efficient market hypothesis in a perspective with the behavioral financial theories, and conclude the significance between them. Anomaly return and behavior is due to an overreaction in the analysis of the information publicly presented. In the empirical study made, the result suggests that there is a reversal to the overreactions, over time. These findings verifies the theoretical suggestions about the efficient market. Over time the prices will emulate the available information on the market. Shleifer (2000) argues against the efficient market hypothesis. Meaning that new studies and evidence, rather suggest that the markets are

(11)

inefficient, or not should be assumed being efficient. With the conclusion that behavioral finance is an alternative model and explanatory variable (Shleifer, 2000).

2.2 Price Pressure Hypothesis

The theory of the price pressure hypothesis can be explained, in this association, as it imposes that recommendations of securities will be exaggerated by the individual investor. Creating a momentary volume pressure of the recommended security. That somewhat develop

incitement for abnormal returns, as a cause from the pressuring investors (Barber and

Loeffler, 1993). The empirical support for the hypothesis is discussed in Barber and Loeffler (1993) where the supporting evidence suggest an abnormal price and volume change in the security as the initial event, followed by a reversal in the upcoming monitored period. The effect of the buying pressure, on the price, is also further significant on smaller firms with lower liquidity flow (Barber and Loeffler, 1993). Sant and Zaman (1996) conclude the same empirical support for the theory, that the effect on a recommendation is only temporary and reversal. They describe the event as a self fulfilling prophecy (Sant and Zaman 1996).

2.3 Information Hypothesis

The information Hypothesis is described by Barber and Loeffler (1993) in the situation of an analysis recommendation, as relevant information is presented and revealed, which

contributes with input to fundamental case of the security. Even though there might be an abnormal anomaly on the day of announcement (Barber and Loeffler, 1993). The empirical studies presented in Barber and Loeffler (1993) gives weak significantly support for the theory. Even though that they conclude that the Empirical significant support for the price pressure hypothesis, does not mean that it extracts the possibility for the information hypothesis to also be significant (Barber and Loeffler, 1993). Another approach for this hypothesis is argued in Sant and Zaman (1996) where the authors suggest that the

information presented in a recommendation might already be the fundamental approach of an analysis of people already owning the stock, which suggest that it is already visible in the

(12)

traded price. From the empirical study, Sant and Zaman (1996) conclude that the effect of abnormal return is reduced in significance, with the numbers of analysts following the security, which gives support to the previous statement (Sant and Zaman 1996).

2.4 Behaviour Finance

The Behaviour financial perspective is the backlash for the efficient market assumptions.

This perspective takes the human decision-making process into consideration, and may be associated with the inefficient markets and irrational decision. Where one of the

argumentations against the efficient market hypothesis are the high volatility in stocks, or overreactions, that instead could be explained with psychological phenomenons (Burton and Shah, 2013).

2.4.1 Keynesian Beauty Contest

The concept of Keynesian beauty contest is based on a prize competition by a newspaper in London, where readers would choose a number of faces that they thought were most

beautiful, those who chose the most popular faces won. In order to win, they would therefore choose faces that they thought, that others thought were most beautiful, and not necessarily those that they themselves liked. In other words, information and knowledge of what others considered was central. Keynes (1936) saw that the stock market shared the same

phenomenon, as many investors who traded on a short-term looked at the stock market in the same way. Instead of going after their own cases, they choose to invest in the companies that most others believe in the moment (Keynes, 1936).

For investors who make use of this philosophy, both private and public information play a role, but most important is the public information as it reaches out to all investors. Even though the noise of both private and public signals enter the individual demands of the investor, the independent noise of private signals get cancel when the individual demands are aggregated. Unlike private signals, noise in the public signal remains in the aggregate

(13)

demand since the individual demands share the same noise term. As a effect, the public signal influences the price by its information value. This role of public information biases stock prices away from the fundamental value unto public information (Allen, Morris and Shin, 2006).

2.4.2 Herd Behaviour

People believe that they have full control of their own behavior and thoughts, but according to social psychology this is not true. Social psychology can be defined as "The scientific study of how we think about, influence, and relate to another". 70-80 percent of our waking time, we devote to communicating in some way, which confirms the certainty that we are social beings, and increases the likelihood that we will affect each other. An important lesson to take from social psychology is the influence others have on us, and therefore herd

behaviour has been researched a lot within social psychology (Rob Henderson, 2017). To describe more specifically for herd behavior, it can be said that it’s a state where the person are acting according to the actions of others instead of using their own signals, private information, to act. Meaning a person chooses to act differently when he is aware of another individual's actions, which differs from what he would do without this knowledge. That is why this phenomenon is called herding, a person changes his own path and instead follow the herd (Banerjee, 1992).

Banerjee who is a economist at MIT, present a model that shows if we trust your own private signal, or if we follow other, and what is going to happen. He gives a common real world example to make the argument clearer. The example is a group of people who will choose between two restaurants, A and B that are located beside each other. It is known that the restaurants are generally as good as each other, but the individuals know their own best choice, so they have their own private signal that tells them which restaurant suits them best.

Say that 100 people are facing this decision, where 99 percent of the people have the signal that restaurant A is the best choice for them, and the rest have the signal that restaurant B is the best choice for them. Everyone's private signals have the same quality, no one has a

(14)

stronger signal than the other. The individuals arrive at the restaurants in sequence. Let's say that the person who has signal B get to choose first, then he of course going to pick restaurant B. Next person in line going to know that that the first person favored B, but his own signal is A. Because both signals are of equal quality, they will cancel each other out and the rational choice going to be the same as the previous person, restaurant B. Regardless of its personal signal A, he chose B. This pattern will go on and affect the next person in line, which will result in all ending up in restaurant B (Banerjee, 1992).

2.4.2.1 Herd Behaviour And Its Effect In The Stock Market

In various empirical studies, it has been found that the share price is more volatile than the expected return, which has worried the market. An explanation for this is the hardening effect that has influence the market (Christie and Haung, 1995). People has become increasingly aware of this phenomenon in connection with economic crises, it’s frequently argued that crises being a result of herding behavior. There is extensive hardening among investors today, which is well known to both economists and practitioners (Devenow and Welch, 1996). Bikhchandani and Sharma (2000) defines it as an investor is part herding if he is aware of and influenced by acts of other investors. Herding, in context to investors, can be divided into intentional and spurious herding. Intentional herding means that investors choose to resemble others with intent, which may lead to the market becoming inefficient. Spurious herding occurs when investors face similar decisions and problems, which also causes them to act similarly, unlike intentional herding, this leads to an efficient market (Bikhchandani and Sharma, 2001).

Herding behavior is not only of interest to economists, but also practitioners. Herding is of interest to economists because the behavioral effects affect stock prices. The behavior might affect both return and risk characteristics, and therefore has consequences for asset pricing models. On the other hand, practitioners are interested in this kind of behavior among investors because it might create profitable trading opportunities. The influence of investor herds has the strength to drive stock prices away from their fundamental values (Tan, 2008).

(15)

3. Methodology

To monitor and measure an isolated price impact due to information content, the most successful technique is to use an event study. The event study captures the abnormalities in the event and the surroundings of an event (Khotari and Warner, 2006). Since the purpose of the paper is to measure the impact from information, in the content of recommendations, the most suitable and acknowledge method to the best of our knowledge, is the event study method. This section also presents the sample and a brief introduction to the magazines that the recommendations are retrieved from.

3.1 Event Study

The event study is a well known, and certified scientific method used to measure, or evaluate, a current events impact on the value of the firm. The convenience of using an event study, is based on the fundamentals of efficient markets, the results will be interpreted in the firm value immediately, making the event measurable as an isolated event (MacKinley, 1997).

The event studies also serve another purpose, which is to test the efficiency of the market, if the event studies are constructed to monitor the long term horizon after the event occured.

Than the market efficiency can be tested (Khotari and Warner, 2006). The method of using an event study is well recognized and was first used in 1933 by James Dolly, where he examined the effect on firm value after a stock split (MacKinley, 1997).

3.1.1 Procedure For The Event Study

The initial process for conducting an event study is to determine the event and the time period of the event. The examination period should be in the surrounding window of when the event occurs. Including, at least, the day of notification and the following day. The prior period can also be of importance since there is a risk, or opportunity, that the market has acquired the information prior to the announcement. The main focus is to capture the price effect of the

(16)

firm value, due to the announcement. The selection criteria is a main part for the constraints of event analysis, this for making the samples comparable and consistent with each other. The monitoring time is varying from paper to paper (MacKinley, 1997). Barber and Odean (2017) uses 100 days of monitoring, with an event window consisting of 50 days. While MacKinlay (1997) suggests a 120 days window is standard. Liu et al. (1990) even had an estimation of 200 days. In similar reports made by Lidén (2004), he uses the standard 120 days estimation and a window of 41 days, +/- 20 days of when the event occurred (Lidén, 2004).

3.1.2 Abnormalities

With the abnormal returns and the abnormal volume the actual effect of the recommendations can be measured with a value. These values can then be interpreted to test whether the

recommendations create abnormalities significant from zero. The abnormalities are used as the measurement to test the stated hypothesis and theories presented. The cumulative

abnormal return/volume are used to test the overall effect of the recommendations and also if the market is efficient. The CAR/CAV tests the power of the recommendation, to see if there is any fundamental substantial change after a recommendation, or if the recommendations are overreactions.

3.1.3 Abnormal Return

The interesting part that is measured, as a result from the occurred event, is the abnormal return. The abnormal return is the measured return from the event, subtracted from the normal return. The normal return is derived and defined, as the expected return, without the event taking place. There are two legitimate ways to estimate the normal return. The first one is called constant mean return model, assuming that X is constant. The second way is called market model, assuming that X is the market return. There are different market models that can be used to estimate the expected market return, suggested models are APT and CAPM.

What should be mentioned is that this brings new problems related to the restriction of APT

(17)

and CAPM model (MacKinley, 1997). With the abnormal returns the actual effect of the recommendations can be measured with a value.

The formula of abnormal returns:

ARit= Rit -(E)Rit

Where,

ARit= Abnormal return for stock i on day t.

Rit= Return of stock i on day t

(E)Rit= The expected return of the stock i on day t

The formula for the market model:

(MacKinley, 1997).

Where,

Rit= The expected return of stock i

i= The estimated intercept for stock i and _iRmt= The estimated beta of the stock stock.

α β

These values come from the estimated OLS from the estimation period.

i=The error term for stock i ε

The constant mean return model is used by calculating the market, as the average return of all the stocks during the estimation window. This method might be the one of the simplest, but the result is considered to be as true as more sophisticated and complicated models (Brown and warner, 1980). To give more depth to the paper, the market model is also used in the calculations of abnormal returns. The market used is the SIXRX index, since this index is the most comparable to our stock universe. Using the CAPM formula for estimating alpha and beta for each stock to get the expected return, then subtracting the expected return from the actual returns, to get the abnormal return. The risk free rate is assumed to be zero, since the key rate is negative during this period. The market model is suggested to be more

sophisticated in sense of reducing the variance of the returns, compared to the constant mean model (MacKinley, 1997).

(18)

3.1.4 Cumulative Abnormal Return

For interpreting any result from the event study, the abnormal return have to be aggregated.

The aggregation will have to be done in two manners, through time and between samples.

The determination of the mean cumulative aggregated return can be described with the following formula:

Where,

CARi(t1,t2)= The cumulative abnormal return for stock i under the period t1 to t2 (The event

window)

ARit = The sum of t days abnormal returns for stock i under the period t1 to t2 (The event

∑^t2

t=t1

window)

(MacKinley, 1997).

3.1.5 Abnormal Volume

The abnormal volume can be measured as the sum of traded volume, subtracted from the estimated average. To get the differences in traded volume within the event window, divide the abnormal volume with the average estimated volume (Joseph et al, 2011).

Ajinkya and Jain (1989) suggests that there can be a market model derived from the traded volume, just as their are a market model for the returns (CAPM). The model is derived from the principles that announcement of information affects the whole market, including the trading volume. Since the assumption that trading volume of the market is correlated with trading volume of the individual firm, the model can be confirmed as a market model (Ajinkya and Jain, 1989). To be able to compare the samples in sense of time horizon and between samples, the mean cumulative abnormal volym will also be investigated (Lidén, 2004).

(19)

The formula for abnormal volume:

AVit=(Vit-Via)/Via

(Joseph et al, 2011).

Where,

AVit= The abnormal traded volume for stock i on day t.

Vit= The traded volume for stock i on day t.

Via= The average traded volume for stock i.

The formula for cumulative abnormal volume:

CAVi(t1,t2)= ∑^t2 AVit t=t1

(Joseph et al, 2011).

Where,

CAVi(t1,t2) = The cumulative abnormal volume for stock i under the period t1 to t2 (The event window)

AVit = The sum of t days abnormal volume for stock i under the period t1 to t2 (The event

∑^t2

t=t1

window)

When calculating the abnormal volume the presented formula above was used. The reason why a market model was rejected, is because of the lack of a comparable index to use. Since the sample includes some stocks that are insolvent and have low trading volume. Using the OMXSPI as a base for the market model, which would be the best alternative, would possibly erase the significance in the trading volume for these kind of companies. The second

approach is the fact that some of the stocks are listed outside of the NASDAQ market indices, which means that the index would not be comparable, since some of the stocks are not

included in the index.

(20)

3.2 Data Description

The recommendations are collected from the leading financial newspapers/E-newspapers in Sweden. The recommendations data are collected from Affärsvärlden, Börsveckan and SvD Börsplus, which are all presented and described below. The variety of the study is limited to two years, 2017-2018. The constraints for the collection of recommendations are limited to only actual recommendations. This means that the words “buy” and “sell” has to occur. In sense of a magazines writing that a stock might be “underpriced” or the “future prospects are promising”, or “we are keeping off the stock”. Which by some means can be interpreted as a recommendation, is not interpreted in this study as a recommendation and are not collected as a sampledata. The stock data collection for making the event study is collected from the Thomson Reuters database. Where the event of the recommendation will be called T0 and the following, or previous, period will be T+/-days after/before the event occurred. The prices used is the closing price. If the event occur on a red day or during weekends, the T0 will occur on the first day that the market is open after the event occured. If a recommendations is occuring during the evening, when the market is closed, the T0 will be assumed the day after.

This to make T0 consistent, when the information can be traded on, in the market. The estimation window is T-135 to T-15, which is 120 days. The event window is set to be 46 days, T-15 to T+30, including T0. The reason of using a period previous to the event is to capture the effect of potential information being released before the publication as well. The usage of a longer period after the T0 is motivated by trying to capture the aggregated cumulative abnormal return, to see if there is a reversal of the abnormalities during the event window.

This to test the power of the market efficiency and see if there is an overreaction with a reversal, or if the stocks keep drifting.

Figure 3.1: Timeline of the Event study (Estimation- and Event window).

(21)

The figure below shows how an overreaction and is intended to serve as an example that will make it easier for the reader to understand how it looks graphic. The diagram shows an overreaction and then a reversal.

Figure 3.2: Diagram showing a overreaction (Viswanath, 1996).

3.3 Sample

There is one sample that is tested in different constellations. The sample divided into buy and sell recommendations that are tested separately (Table 3.1). Furthermore the sample is

divided into size, or what market they are listed on (Table 3.2 and 3.3). The test made, based on size of the companies, are divided into two groups, large and small companies. The large companies are defined as companies operating on Large Cap and Mid Cap. The small companies are defined as companies operating on Small Cap, First North, Spotlight and NGM. Due to variation in recommendations for each market, the decision of splitting the sample into two groups of small and large companies arise. The qualification for each group is derived from the market capitalization.

There should also be mentioned that some stocks in the selected data set did not fulfil the criteria of having daily data within the estimation window. This due to being newly listed to

(22)

the stock exchange. These samples were removed from the population due to potential bias result in the interpretation of the average return. This also affects the recommendations on buying and selling stocks that have an upcoming listing on an exchange, and the samples affected from that were therefore not considered into the population.

Magazines/Recommendations Buy Sell Total

Börsveckan 224 17 241

SvD Börsplus 140 34 174

Affärsvärlden 120 36 156

Total Recommendations 484 87 571

Table 3.1: Number of buy and sell recommendations, divided into magazines.

Market/Recommendations Buy Sell Total

Small Cap 93 9 102

First North 73 15 88

Spotlight 24 4 28

NGM 2 2 4

Total Small Companies 192 30 222

Table 3.2: Number of buy and sell recommendations, divided into smaller markets.

Market/Recommendations Buy Sell Total

Large Cap 108 36 144

Mid Cap 184 21 205

Total Large Companies 292 57 349

Table 3.3: Number of buy and sell recommendations, divided into larger markets.

(23)

3.3.1 SvD Börsplus

SvD Börsplus is Svenska Dagbladets platform where they collect their articles and reports from the stock market. SvD Börsplus's most important task is to help their subscribers to make money with the help of their analyzes, according to Josefin Sigedal, CEO of Börsplus.

In addition to helping people get more money in the account, they fulfill an important

function for society as stock analyzes are needed to expose both good and bad companies. At the beginning of 2018, SvD had almost 350 000 readers (svd, 2018). The recommendations will be taken continuously as the recommendations are released online on an ongoing basis.

The T0 for the occurred recommendation event will be the next day that the market is open.

3.3.2 Affärsvärlden

Affärsvärlden is a magazine that has entered the market slightly later than the other news sources that is analyzed. At the end of 2017, they had about 250 000 readers, of which 80 000 read VA-finans which is the newspaper's niche monitoring of the stock market. The magazine gives a deeper insight into the stock market and what happens in the center of power.

Affärsvärlden focuses on providing quick news, comments and analyzes about the business sector and the stock market development, says Erik Wahlin, editor-in-chief of Affärsvärlden (affarsvarlden, 2018). The recommendations will be taken from the weekly magazine that is published every thursday morning. The T0 for the occurred recommendation event will be the same day.

3.3.3 Börsveckan

Börsveckan purpose is to find the best shares on the stock exchange and thereby inspire readers to good stock trades. They give their subscribers sharp buying advice every week and access to a couple of different portfolios, this according to the Börsveckans website

(borsveckan, 2017). Börveckan does not want to show public figures for the number of readers, neither on their website nor after mail contact. But one can see that they affect the

(24)

market with its recommendations, to the same extent as the other two magazines that has been investigated. The recommendations will be taken from the weekly magazine that publishes online every saturday. The T0 for the occurred recommendation event will be the first day that the market is open.

3.4 Method Issues/Bias

When conducting an event study the basics is to estimate the normalities in the “estimation window”. That later on should be subtracted from the samples in the “event window” to get the abnormalities as a reaction to the event. There might be bias in the estimation window due to events occurring in the estimation window is not taken into consideration in the estimation results. We have discovered that some stocks have several recommendations from the same, or comparing, magazine. Which means that in the estimation window their might be abnormalities occurring that are not taken into consideration in the estimation. There is also a possibility that other market-driven “players” are affecting the stocks irrationally from the normalities during this period as well. Which might cause a bias in the “normality”

estimation. Since these events occurring, they can also be seen as normal market activity appearing. If we believe in efficient market, these event will not have any effect to the estimation, since the market will adjust to its true value, and no bias will occur due to events happening under the estimation window.

When managing with the recommendations, there is always the possibility of selecting the wrong announcement date as T0. MacKinley (1997) points at this being a issue especially for the printed newspaper, with the information being known to the public prior to the

publication date (MacKinley, 1997). In the study, we do not believe that this is a huge error, since the recommendations collected is from the online articles of the printed magazines, which can give us the exact date and time of the announcement.

(25)

3.5 Regression Specification

The multivariate regression model will be used to conduct the regressions for the market model of the abnormal returns and the cumulative abnormal returns. The test is specified as a joint test, which means that the mean abnormal returns and mean cumulative abnormal returns are presented in the empirical result for all recommendation. This method can be used when the events are all occurring at the same time, or the data is formatted for all the events to occur at the same time. To conduct a multivariate regression model, each day in the event window is specified with a dummy variable (Binder, 1985). For example D-15 is coded with 1´s for all the events occurring on day T-15, and the rest is coded with a 0`s for this specific dummy. By constructing a set of 46 dummies, the event window of t-15 to t+30 is covered.

The formula:

Rt= α β_t+ ₁Rmt +γ_1tD-15+γ_2tD-14+...+γ_46tD+30

Where,

Rt= The mean return of all stocks on day t.

t= The intercept.

α

1Rmt= The estimated mean market return on day t.

β

1tD-15= The estimated abnormal return for T-15. γ

(Binder, 1985)

The assumptions for running the regression are that the returns are normally distributed, temporally identically and independent. Otherwise the result would be asymptotic, if not assuming normality (MacKinley, 1997). In event studies one of the most common misspecifications are clustering (Khotari and Warner, 2006). To overcome the clustering problems, the usage of a multivariate regression model is suggested, this to erase the potential problem of the securities overlapping in the event time (MacKinley, 1997). To overcome potential autocorrelation and heteroskedasticity the robust standard errors are used.

(26)

4. Empirical Result

This chapter consists of the results that has been obtained through various tests in Excel and Stata. The results will only be commented briefly in this section, in order to analyze in the next part.

4.1 Data

The purpose of the study is to determine whether a recommendation generates abnormal return and Abnormal volume on the day of announcement (T0) and in the surrounding event window. If the aggregated average AR and AV for the sample are not equal to zero,

abnormalities are occuring. What is also tested is the average cumulative abnormal returns and volume, that can indicate if the recommendations have an impact during the whole event window, it can also show if there is a reversal in the return after the recommendations. Which is another purpose of the study, to investigate whether the recommendation changes the fundamental approach of the stock, or if it is just an overreaction. If the average CAR and CAV are close to 0 in the end of the event window, one might suggest that the

recommendation has no effect on a longer perspective.

Since the AR for the buy and sell recommendations has been calculated using the market model and the constant mean model, we will perspicuously present the numbers for the constant mean model and have a larger focus on the market model estimates. The reason for selecting the market model, over the constant mean model, is that the regression is more sophisticated and is taking the risk for each stock into consideration, which makes the model more advanced and so the result more reliable. Even though the constant mean model has provided confirmation to that the model is adequate in comparison to the market model.

(27)

4.2 The Reaction To A Recommendation

Below presented are charts showing CAR and CAV during the event window.

4.2.1 Buy/Sell

Figure 4.1 showing CAR for both buy and sell recommendations. The line that shows buy recommendation, indicates that shortly after the recommendation, there is reversal. The case is not the same regarding the sales recommendations, where figure 4.1 instead show that CAR continues to drift in the same direction and in the end of the event window, CAR has past -15%.

Figure 4.1: Line chart of CAR showing buy and sell recommendations.

4.2.2 Buy/Sell Large Companies

Figure 4.2 shows that the impact is not as great when looking just at large companies. One can still distinguish the same pattern as figure 4.1, as buy recommendations show a reversal and sell recommendations drifts down, but not as drastic.

Figure 4.2: Line chart of CAR showing buy and sell recommendations in large markets.

(28)

4.2.3 Buy/Sell Small Companies

Figure 4.3 shows that the smaller companies are to a greater extent affected by the magazines recommendations. Buy recommendations makes CAR increases, and then remains on the same level, only a small reversal occurs. Sales recommendations, on the other hand, continue in a downward trend.

Figure 4.3: Line chart of CAR showing buy and sell recommendations in small markets.

4.2.4 Volume

Figure 4.4 shows the volume of both buy and sell recommendations. The chart shows that trading, when the recommendations are given, increases sharply for about ten days, then a reversal occurs.

Figure 4.4: Line chart of CAV.

(29)

Table 4.1: AR and CAR day by day Table 4.2: t-value day by Table 4.3: AV and CAV using market model. day using market model. day by day using constant

*Significant on 10% level, **Significant on 5% level, ***Significant on 1% level mean model.

(30)

4.3 Sell Recommendations

Presented in the appendix 7, are the average abnormal returns (AR) and the average

cumulative abnormal returns (CAR) on a daily basis, using the constant mean return model.

The abnormal return for the T0 suggest that the sell observations are significant (- 3,89%).

The cumulative abnormal return shows a downward trend that starts already shortly before T0, but intensifies at T0 and continues downwards. The CAR for T-1 are summarized as (-2,81%). The average summarized CAR for the constant mean model ends with the value of (-17,76%). This suggest that there is no reversal in the stock, since the value is far away from 0. Which might suggest that the sell recommendations have an effect on the fundamental stock case.

Presented in Table 4.1 are the average abnormal returns and the average cumulative abnormal return in the event window, using the market model. The abnormal return on T0 is significant on the 1% level with a measured abnormal return of (-3,93%). The T+1 is significant on the 5% level with a measured abnormal return of (-1,16%). Other significant days in the event window are T+14 and T+21 on the 5% level and T+9, T+10, T+11, T+19 T+22, T+25 on the 10% level.

The cumulative abnormal return is negative (-2,75%) for T-1. The CAR for the sell

recommendations drifts in a downward trend and ends on (-17,17%) on T+30, which also is the maximum value for the CAR in the event window. This suggests that there is no reversal in the stock during the event window.

4.4 Buy Recommendations

The empirical results for the constant mean model are presented in the appendix 7. The value of the AR at T0 can be measured to 2,63%. The CAR for T-1 is -0,54%. The CAR peaks at T+2

with a value of 2,41%. The CAR then drifts in a negative trend to reach 0,40% at T+30. This suggests that there is a reversal in the event window since the ending CAR value is close to 0.

(31)

The empirical results for the market model are presented in table 4.1. T0 is significant on the 1% level and is measured with an abnormal return of 2,70%. The T+1 is significant on the 10% level and is measured to 0,20%. T+6 is significant on the 5% level with a negative (-0,22%) abnormal return measured. Other significant values are T12 and T+9 on the 5% level.

T+14 is significant on the 10% level. The cumulative abnormal return for the T-1 is close to 0, 0,08%. The peak is on T+2 with a cumulative value of 2,90%. After the peak on T+2 there is a downtrend and the cumulative abnormal returns ends on T+30 at 1,51%. This suggest that there is a reversal in the return during the event window.

4.5 Volume

Table 4.3 showing the changes in volume for both buy and sell recommendations. The volume is not divided between buy and sell because trading in the shares is affected in the same way, regardless of whether a share is bought or sold. The abnormal volume (AV) at T0

is 103%, which indicates a high impact from the recommendations. AV also shows that trading in these companies increases ten days in a row around the recommendation (T-2 - T+7).

As AV increases quite clearly already two days before the recommendation, T-2 (19%) and T-1

(37%), one can imagine that there is leaked information in advance. The cumulative abnormal volume (CAV) reaches its peak at T+9 (225%), after that a reversal occurs and at the end of the event window T+30 is 33%.

(32)

Table 4.4: AR and CAR day by day using Table 4.5: AR and CAR day by day using market model for large companies. market model for small companies.

(33)

4.6 Sell Recommendation Divided Into Large And Small Companies

Presented in Table 4.4 and 4.5 are the divided samples categorized into small and large companies. The empirical results shows that the AR for small companies, with a sell

recommendation, for T0 can be measured to -6,49%. T1 is measured to -2,61%. The value of the CAR for T-1 is -6%. The CAR keeps going in an negative trend through almost the whole event window. At T+30 the CAR peaks at -36%. For the large companies with a sell

recommendation the measured AR for T0 has a value of -2,61%. The CAR for T-1 is -0,84%.

The CAR peaks at T+28 with a value of -7,27%. The CAR for the last day of the event window (T+30) is measured to -6,95%. This suggests that there is no reversal in stock prices after the recommendations. The CAR for T-1 is low for large companies compared to small companies.

4.7 Buy Recommendation Divided Into Large And Small Companies

When dividing the sample into two groups of “small” and “large” companies the empirical result changes. Presented in Table 4.4 and 4.5, the AR for small companies at T0 can be measured to 4,65%. The T0 is significant at the 1% level. The CAR is close to 0 at T-1 with a value of 0,47%. The CAR peaks as T+2 with a maximum value of 5,57%. After T+2 The CAR drifts in a downward trend and ends at 3,13% on T+30. Which shows some kind of reversal, but not in the same latitude as with the whole sample. The empirical results for the large companies shows us an AR of 1,42% on T0. The T0 is significant on the 1% level. The CAR for T-1 is negative at (-0,56%) which is a value close to 0. The CAR peaks at T+2 with a value of 1,14% and then goes in a downwards trend and end at a value of 0,09% at T+30. The empirical results can be interpreted as, that the recommendation has a large impact on the stock value of small companies, while for large companies the recommendations have small impact, in terms of measured AR.

(34)

5. Analysis

In the following part, the empirical results will be analyzed. Results will be explained with support of the theories previously presented. The analysis section will also investigate the issues that started the thesis and see if the hypotheses are correct.

5.1 Sell Recommendations

For the sell recommendations, the days of significant abnormal returns are spread on various days without any rationality within the event window. Since the central aim of the study is to investigate the abnormal returns around T0 , the focus will be on analyzing the days around T0. But the significance for T+9, T+10, T+11, T+14, T+19, T+21, T+22, T+25 might suggest that the sample is to small, and therefore the result could be interpreted as random or biased. It can also suggest that there are other events occurring within the event window. The spread significance impairs the power of the result.

The sell recommendations empirical results suggests that the recommendations have a large impact on the stock returns. After the negative recommendations have been made there is a clear downtrend in the stocks, that lasts through the whole event window. This is not in line with the theories of the efficient market hypothesis (Fama, 1970) or the price pressure

hypothesis (Barber and Loeffler, 1993), that suggests the recommendations are overreactions or anomalies. The results could be associated with the financial behavioural approach and an inefficient market as Shleifer (2000) suggests. Or it could be interpreted as herding behavior that drives the stock away from its fundamental value (Tan, 2008). What´s also interesting from the empirical results are the fact that the CAR is already high (- 2,75% at T-1) before the announcement day. This would suggest that the information about a sell recommendation might be leaked in advance. It might also be the case of a where the recommendation is a reverse causality, meaning that a series of negative events, or negative future prospects, is the

(35)

underlying concrete for the paper to write a sell recommendation, suggesting that the stock is already in a negative downtrend, before the announcement.

The conclusion that can be made is that the sell recommendations have a large impact on the stocks in the sample. What is interesting with the result is that a trading strategy can be formulated. Based on shorting stocks after the sell recommendations would give an average return, in our sample, of 10,49% if the stock is bought on T0 closing price and sold on t+30.

There should also be mentioned that the sample of the sell recommendations are less then ⅕ of the buy recommendations (484 buy recommendations and 87 sell recommendations). A smaller sample size can have effect on the empirical results to be somewhat biased, or at least not as reliable as the buy recommendations result. Another interesting interpretation that can be made from the empirical results, based on the variation of buy and sell recommendations, are that it could be suggested that the sell recommendations have a larger impact due to the moderation of them. The case might be that investors frequently gets the buy

recommendation in their information feed, while the sell feed is more restricted by the

magazines. Due to that, a more serious approach and consideration is taken from the investors when a sell recommendation is issued.

The empirical results of the sell recommendations are varying in the result compared with previous studies made in the area Keasler and Mcneil (2010) and Barber and Loeffler (1993) for example, conclude that they can see a reversal in within the event window, with the exception of small cap companies for Keasler and Mcneil (2010). While Barber and Odean (2007) interpret the result that people buy on buy recommendations, but does not sell on sell recommendations, which is far from the results from this study. While the result is close to the same as Lidén (2004).

(36)

5.1.1 Sell Recommendations In Large And Small Companies

The empirical results when dividing the sample into groups of large and small companies, show that the reaction to a recommendation in small companies are more evident then large companies. It can be explained with the information hypothesis (Sant and Zaman 1996). For small companies the T0 is measured to -6,49%, while for large companies the value is -2,61%. The recommendation effect for small companies in the category of sell is highly significant. What is also interesting is the CAR for the small companies, that is measured to -36% for the event window. The conclusion that a sell recommendation change the

fundamental case and the value, is adequate to assume for our sample. The psychological aspect of behavioural finance could also be an explanatory variable for the stock reactions.

The herding behavior could be an explanation for the sell pressure and the negative trend. A disclaimer of the results are highly relevant to post. The sample for sell recommendations in small companies can be measured to only 30 recommendation. Therefore the empirical results for this category is not as reliable as the rest of the study. Even though that the conclusion are in line with relevant theories and the results with the expectations, the power of the empirical results should be kept in mind.

5.2 Buy Recommendations

The empirical results shows a negative significance on the 5% level for T-12. The significance for this date could be explained as a random event or other events occurring within the event window. The negative AR for this date is not in line with previous studies and there is no good explanation for it to occur. Since the main focus of the study is the surrounding days of T0 and the CAR for the whole event window, this day will not be analyzed furthermore.

The empirical results for the buy recommendation conclude that there are abnormal returns existing due to recommendations from the Swedish stock related magazines that were investigated in the study. The abnormalities for the buy recommendations can be interpreted as overreactions, when analyzing the CAR for the event window. There is a reversal in the

(37)

AR for the stocks. The negative significance level for T+6, T+9 and T+14 supports the reversal with negative AR for each day. This is in line with what the theories of the efficient market and price pressure hypothesis. The recommendations are just anomalies or overreactions and the market will always adjust to it´s true value, which is defined as the information available on the market (Fama, 1970). Also the price pressure hypothesis can explain the empirical results as the overreaction can be explained as an exaggerating from the investors, creating a momentary pressure that is defined as the abnormalities (Barber and Loeffler, 1993). It could also be explained with intentional herding (Bikhchandani and Sharma, 2001) or selecting stocks based on what other believe will be a good case (Keynes, 1936). These strategies supports a potential overreaction as a result of using behavioral finance as a trading strategy.

Buying stocks based on attention driven philosophy and buying stocks that other people are interested in. The empirical results are in line with previous studies as well, such as Keasler and McNeil (2010), Tetlock (2007) and (Barber and Odean, 2007) for example. Who

conclude that recommendations and abnormal returns are significant. They also conclude that there is a reversal in stock prices within the event window.

5.2.1 Buy Recommendations In Large And Small Companies

The empirical results when dividing the sample into two categories of large and small companies, gives a larger understanding of the effect of the recommendations. There is clearly more market influence, with a recommendation in small companies than in a large.

The empirical results shows that the AR for small companies on the announcement day is more than 3 times as large, as for large companies (4,65% for small companies and 1,42% for large companies, using the market model). There is a reversal within the event window for large companies, while small companies still show a value significant from zero in the CAR.

This could be explained by the information hypothesis, that explains that the stock price are reflected by the information available for the stock (Barber and Loeffler, 1993). Sant and Zaman (1996) conclude that AR is significant with the number of analysts following the stocks. Which gives support to draw a conclusion that large companies have more

information available, and therefore a recommendation, or the presentation of information,

(38)

might already be printed in the stock price. While small companies who are less followed, and have less information available, there a recommendation might change the fundamental stock case, and have a larger significant effect.

5.3 Volume

The empirical results show that there are distinct abnormal differences in the volume when a recommendation from the financial magazines occur. One explanation for this can be the phenomenon of the beauty contest, who says that you buy or sell because of the information of what others are doing, and not because of your own case or thoughts (Keynes, 1936).

Herding behavior can also describe why the volume increases to a great extent, as people often abandon their own signals when one gets information how other people act (Banerjee, 1992). The volume increases explosively t-1 to t + 1, as CAV increases by 194% during these three days. This can be interpreted as a clear result of the magazines recommendations having an impact on the market, as people react and the volume of the share increases as much as it does.

(39)

6. Conclusion

This paper has investigated whether stock recommendations from the leading Swedish financial magazines affects the value and the volume of the traded stocks. The hypotheses was that the recommendations would have an effect on the the short run, but that the recommendations would be classified as overreactions. What was also assumed is that the recommendations would have a larger impact on small insolvent companies, then on larger companies.

The conclusion is that there are abnormal returns (AR) and abnormal volume (AV) existing due to recommendation from the financial magazines. The AV more than dubbles on the day of recommendation announcement. The buy recommendations have significant AR that can be explained as an overreaction. There is a clear reversal in the cumulative abnormal returns (CAR), within the event window, which is in line with the theories of the efficient market hypothesis and the price pressure hypothesis, but also with psychological approaches, such as behavioural finance and herding behaviour. While the sell recommendations keeps drifting in a negative trend after the recommendations. The sell recommendations can be concluded as a changer of the fundamental stock change. The CAR for the sell recommendations are

significant from zero before the date of announcement, which suggests two things. The first one is that the information might be leaked before the announcement date. The other

perspective, that could explain this, is the reverse causality, that the recommendation is rising from the stock being in a negative trend before the recommendation. The sell

recommendations are not in line with theories as the efficient market hypothesis and the price pressure hypothesis. The negative trend after the recommendation could be explained with theories of behavioural finance, such as herding behaviour or the trading strategy explained in the beauty contest. The recommendations have a larger impact on smaller companies than large companies. This could be explained as an effect of more information availability and analytics following large companies, which is supported by the information hypothesis and previous studies made on the topic. There are still existing AR for large companies, but with

(40)

a smaller value. Only large companies with buy recommendation reverse in stock price within the event window. The recommendations have the most effect on small companies in the category of sell. The results are somewhat in line with previous studies made on the topic, when only having categorizing the sample into buy and sell recommendations. When

dividing the sample into small and large companies as well, the sample of buy

recommendation for small stocks are showing differentiating results from the previous studies. Where there is no significant reversal in within the event window, and there is no support for the recommendation being categorized as an overreaction.

For further research it would be interesting to design and concrete a profitable trading strategy, based on recommendations from financial magazines. There is somewhat support and evidence for constructing a profitable trading strategy shorting stocks on sell

recommendations in this study, but more investigation and tests has to be made to make it more concrete and reliable. Also a larger sample of sell recommendation have to be mined, to verify the results and conclusions that we have made, based on our sample.

Financial Magazines impact on the Swedish Stock Market: An event study

Bachelor Thesis