The influence of sentiment on risk premiums

Transcription

1 TILBURG UNIVERSITY The influence of sentiment on risk premiums A study of the influence of sentiment on different risk premiums Concerns: Master Thesis in Finance Name: Sander Vrijhoeven ANR: Date of defense: 19 th December 2013 Name of supervisor: dr. J.C. Rodriguez Second reader: prof. dr. F.A. de Roon Page 0

2 Part I Abstract This thesis focuses on the influence of sentiment on risk premiums. It will focus mainly on the effect of sentiment on the market risk premium in the United States by conducting a regression analysis using a panel dataset from the NYSE, AMEX and NASDAQ covering the time period Another study has been done to look for the relationship between sentiment and the variance premium. After empirical research no significant relation was found between sentiment, as measured by Baker and Wurgler (2007), and the market risk premium. However, some evidence was found of the different influence sentiment can have on different industry portfolios which suggests that more stock specific variables such as age, profitability and others are indeed correlated to sentiment and allows further understanding of the anomalies to a model such as the Capital Asset Pricing Model. Page 1

3 Foreword After over five years of studying I can finally say that, if I pass this thesis, I have my first Master degree title. However, it will not yet mark the end of my life as a student. At this moment I have already started writing my thesis in the Strategic Management program. However, finishing my Master in Finance will be a very relaxing feeling as it has been very challenging. The program has thought me a considerable amount of things I wasn t aware of after finishing my Bachelor s degree. Little did I know what was expected of me in the Finance Master before I started. It has been challenging, hard, but in the end very rewarding and enriching. I am already looking forward to implementing the knowledge gained in the five years at the Tilburg University in the real world. Before you as a reader start reading my research problem I manage to resolve I would like to thank my supervisor for his contribution to the project. Mister Rodriguez managed to help me start my thesis even while located in Argentina. He gave interesting insights into my subject and managed to steer me in the right direction from the start on. Much credit goes to him for giving me a stress-free writing process with my self-determined meetings and deadlines. I needed the flexibility since I was still doing a course in Strategic Management and also started writing that thesis. I greatly appreciate the many comments that I needed to address and the insights in what I could change in the thesis. Furthermore, I would also like to thank my parents and girlfriend and others in my direct surrounding for supporting me during my study and also during the thesis. The greatest things I learned from studying are to be critical, practical and that everything requires dedication. Page 2

4 Table of Contents Part I... 1 Abstract... 1 Foreword... 2 Part II Introduction Problem Indication Problem Statement Research Questions Structure of the thesis Literature Review Defining sentiment Defining the different risk premiums The relationship between sentiment and the risk premium The relationship between sentiment and the variance premium Methodology and Data Research Design Data Data analysis Results The influence of sentiment on the market risk premium The influence of sentiment on the risk-free rate The influence of extreme levels of sentiment on the market risk premium The influence of sentiment on the different industry portfolios The influence of sentiment on portfolios sorted by size and value The influence of sentiment on the variance premium Discussions and limitations Discussion Limitations Recommendations Page 3

5 5.4 Conclusion Part III Literature list Appendices Page 4

6 Part II 1. Introduction 1.1 Problem Indication There is a still a considerable amount of debate of what influences stock prices. The classical way to think about risk and return was through the Capital Asset Price Model (CAPM) which was established by Sharpe (1964), Lintner (1965), and Black (1972). The central prediction was that expected returns depend on exposure to systematic (i.e. undiversifiable) risk. The way to measure the systematic risk, was through a beta. However, after empirical testing, which was done by Fama and French (1992), the model turned out to be insufficient to say the least. After their work the so-called Generation 1 anomalies were established by researchers who found other variables that predicted returns over and above beta. These are: - Size: small stocks have higher returns - Leverage: highly levered firms have lower returns - Book-to-market equity: stocks with high BM ratios have higher returns - Earning-to-price ratio: stocks with higher EP ratios have higher returns Fama and French established a new 3 factor model which incurred the size and value effect. It is great in that it explains anomalies based on finance theory, however dubious as it simply defines the anomalies away. In response to this new model, new anomalies were found. Examples of these are: - Long-run reversals: past winners (3 years) will be losers and vice versa - Short-run momentum: short-run winners do well over next 1-12 months - Post-Earning-Announcement Drift: firms with good news do well for up to 4 months Barberis, Shleifer and Vishny (1998) in reaction aligned these anomalies. They identify two families of pervasive regularities namely: underreaction and overreaction. The underreaction takes place over horizons of 1-12 months, and concerns the short-run momentum where security prices underreact to news (Jegadeesh and Titman, 1993). This tends to exhibit a positive autocorrelation for up to four months, which is due to the slowly incorporating of the news into Page 5

7 prices by investors. Thus, current good news has power in predicting positive returns in the future. The overreaction on the other hand, shows that over longer horizons of around 3-5 years, security prices tend to overreact to consistent pattern of news pointing in same direction. Thus, securities which had a long record of good news tend to become overpriced and have low average returns afterwards (De Bondt and Thaler, 1985; Fama and French, 1992). This evidence presents a challenge to the efficient markets theory, which states that prices contain all information (depending on the version this explanation changes), because it suggests that in a variety of markets, sophisticated investors are be able to earn superior returns by simply taking advantage of this under/overpricing without taking on additional risk (Barberis and all, 1998). Fama and French (1996) tend to address the problem from an efficient markets viewpoint. Their 3 factor model can account for the overreaction evidence, however not for the continuation of short-term returns (underreaction). The challenge is to explain how investors might form beliefs which lead to both under- and overreaction. These challenged lead to the main topic, sentiment. Researchers in behavioral finance have been working on augmenting the standard model in which unemotional investors always force capital market prices to equal the rational present value of expected future cash flow (Baker & Wurgler, 2007). As a reaction an alternative model was built based on two assumptions; the first one, as laid out by Delong, Shleifer, Summers, and Waldmann (1990), is that investors are subject to sentiment. Two families of sentiment were determined; underreaction of stock prices to news such as an earnings announcement, and overreaction of stock prices to a series of good or bad news (Barberis, Shleifer, Vishny, 1998). Further specification shows that underreaction takes place over horizons of 1 month to 1 year, consisting of security prices underreacting to news (Bernard & Thomas, 1989; Jegadeesh & Titman, 1993). As a result news is incorporated slowly into prices, which comes in hand with a positive autocorrelation over these same horizons. Autocorrelation, in this case, means that current good news has power in predicting future positive returns (Barberis et al., 1998). This thesis will focus on the influence of sentiment on the general market risk premium rather than stock characteristics as done by previous literature. Sentiment will be quantified by a Page 6

8 sentiment index which is established by Baker and Wurgler (2007). Furthermore, it will try to explain industry differences for risk management purposes. Theory on these subjects is not available in established literature. 1.2 Problem Statement As mentioned in the problem indication, study has been done on sentiment and its relation to the securities price; however no study has been done yet concerning the relationship between sentiment and the risk premium of the market and industries. This investigation will focus on the New York Stock Exchange (NYSE), and through theoretical and empirical analysis, will attempt to clarify the effect of sentiment on the market risk premium and industry risk premiums. The main research question therefore is; How does sentiment influence the risk premium? 1.3 Research Questions In order to be able to answer the question proposed in the problem statement, we will cut the question into several parts. 1. What is sentiment and how can it be measured? 2. What are the different risk premiums and their measurements? 3. How does sentiment influence the market risk premium? 4. Does sentiment have a different effect on different industries? 5. Is sentiment a significant addition to the Carhart 4-factor model? 6. What is the effect of sentiment on the variance premium? Page 7

9 1.4 Structure of the thesis This thesis will be structured as follows. Chapter two will provide an extensive review of the literature which is already available on the topic. Chapter three will explain the reader what methodology and data is used in the empirical section of the thesis. Chapter four will focus on the results which are obtained from the regression analysis performed. Chapter five will follow up with a discussion of the results and will highlight the main limitations in this thesis. Hereafter, the bibliography will provide an overview of the literature which has been used and investigated in order to conduct this thesis. The appendices of the thesis display the regression tables and will provide the reader with tabulated results to support the results from chapter three. Page 8

10 2. Literature Review 2.1 Defining sentiment This first subchapter will discuss what sentiment is, and how it came to be. It will start with a recapture of the problem indication and will extent on this by looking at the research which was done prior to the development of the sentiment index. It will look at the anomalies which were formed after different models came to be and it will inform the reader in why sentiment is important. There is a still a considerable amount of debate of what influences stock prices. The classical way to think about risk and return was through the Capital Asset Pricing Model (CAPM) which was established by Sharpe (1964), Lintner (1965), and Black (1972). The central prediction was that expected returns depend on exposure to systematic (i.e. undiversifiable) risk. The way to measure the systematic risk, was through a beta. Markowitz (1959) said that the market portfolio is mean-variance efficient and establish the following formula. Here E(r i ) is the expected return for a certain stock, r f is the risk-free rate. MRP stands for the market risk premium; lastly the Beta should suffice to describe the cross-section of returns. There should be a positive relation between expected returns and betas. Fama and French in 1992 checked the empirical performance of the CAPM model. Their work lead to the Generation 1 anomalies, these were established by researchers who found other variables that predicted returns over and above beta. The first of these was size, which suggested that small stocks have higher returns. The second one is leverage, which suggested that highly levered firms have low returns. The next one is the book-to-market equity (BM) anomaly, which means that stocks with high BM ratios have higher returns. The last one was the earning-to-price ratio (EP), which meant that stocks with higher EP ratios have higher returns. Fama and French (1992, 1993, 1996) show that size and book-to-market equity (together with beta) capture generation 1 anomaly variables. Their view is that the CAPM finally died in 1992 and lead to a new 3-factor model; Page 9

11 In this model α is an intercept which should be zero, β, s and h are loadings on the market and SMB and HML factors. Fama and French interpret that HML and SMB are not anomalies but are risk-factors instead; this means that high BM- and small-stocks are fundamentally riskier and therefore require a high risk-premium in the case that investors are risk-averse. Therefore, the model is viewed in two ways; either it is great as it explains anomalies based on finance theory, however dubious as it simply defines anomalies away. This is where sentiment comes in, one prominent version of the behavioral view focuses on the complement to value stocks, which are growth stocks. Growth stocks can be extremely attractive to naïve investors as growth stocks have a high valuation relative to their book value. Some examples of (past) growth stocks are Google, Apple, Microsoft etc. These stocks are also referred to as glamour stocks. However, why would naïve investors hold these stocks? There are several possible reasons; - Naïve investors might extrapolate past growth into the future - These stocks might have bragging value or excitement - Naïve investors might confuse a well-run company with a good investment Since glamour stocks are attractive to naïve investors, there is excess demand for glamour stocks leading to overvaluation. As a result it lowers returns going forward, as price is inverse of return as shown in the Gordon Growth Model (Gordon & Shapiro, 1956) 1. Therefore, someone has to hold value stocks, which in turn means their returns rise in equilibrium. Barberis, Shleifer and Vishny (1998) conceptualized this by stating that sentiment explains this phenomenon by appealing to a specific set of biases. Demand shifts The demand side of the market shows that some shifts in investor demand for securities are completely rational. They can reflect, for example, reactions to public announcement which can affect future growth rate of dividends, risk, or risk aversion (DeLong et al., 1990). However, not all demand changes appear to be so rational, as already explained in the previous paragraph. Some of the changes in the demand seem to be changes in expectations or sentiment which are 1 Page 10

12 not fully justified by information. These include, for example, pseudo-signals that investors believe to contain information about future returns which cannot be rationalized by, for example, a model like that of Black and Scholes (1973). Such signals can be advice of brokers or financial gurus or can have other sources. De Long et al. (1990) call these kinds of investors, noise traders. These types of shifts are only important when they are correlated across a significant amount of noise traders. If all investors were to go in different directions, they would mostly cancel out, and no aggregate shifts in demand will take place. However because of, for example, pseudo signals, noise and other factors do lead to aggregate demand shifts (De Long et al., 1990). There are many different factors that in the end influence the sentiment in a market. These will be discussed thoroughly in the next paragraph. Measuring sentiment In behavioral finance, sentiment has been an area of interest for a longer period of time and although mentioned long periods a go, no study was able to actually measure sentiment. More recent studies, as that of Baker and Wurgler (2006), utilize interim advances in finance theory to be able to provide sharper tests for the effects of sentiment. In order to understand how they have come to a measure, it is important to understand what De Long et al. (1990) researched. In their study they define two different kinds of investors, namely; rational arbitrageurs who are sentiment-free and irrational traders who are prone to exogenous sentiment. Together they compete in the market and set prices and expected returns. However, arbitrageurs are limited in correcting those prone to sentiment; these limits come from short horizons or costs and risks of trading and short selling. Therefore, prices will not always reflect their fundamental values. Mispricing arises out of the combination of two: for one, a change in sentiment on the part of irrational traders, and secondly a limit to arbitrage from the rational ones (Baker & Wurgler, 2007). In order to quantify sentiment, Baker & Wurgler (2006) identify different factors that are good measures: Page 11

13 - Trading volume Trading volume, or more generally spoken of as liquidity, is viewed as an investor sentiment index. Baker and Stein (2004) note that if short-selling is more costly than opening and closing long positions, which it is in practice, and then irrational investors are more likely to trade. By doing so they add liquidity when they are optimistic and betting on rising stocks rather than when they are pessimistic and betting on falling stocks. Then market turnover, which is the ratio of trading volume to the number of shares listed on the NYSE, is a proxy for this concept. - Dividend premium Dividend-paying stocks resemble bonds which have a rather predictable income stream that represent a salient characteristic of safety. This price-based measure is the relative premium which is paid for dividend-paying stocks and is inversely related to sentiment. Baker and Wurgler (2004) defined the dividend premium as being the difference between the average market-to-book value ratios of dividend payers and non-payers. When dividends are at a premium, firms are more likely to pay them, and less so when they are discounted (Fama and French, 2001). - Closed-end fund discount The closed-end fund discount (sometimes a premium) is the difference between the NAV (Net Asset Value) of a fund s actual security holding and the fund s market price. If closed-end funds are disproportionally held by retail investors, the average discount on the closed-end equity funds is a sentiment index, with the discount increasing when retail investors are bearish (Lee, Shleifer and Thaler, 1991; Wheatley, 1998). - IPO volume IPO volume displays wild fluctuations, sometimes there are over 100 issues per month when in some periods there are near to, or even actually, zero issues per month (Baker & Wurgler, 2007). Therefore, the underlying demand for IPOs is often said to be extremely sensitive to investor sentiment. Page 12

14 - First-day returns of IPOs Sometime initial public offerings earn remarkable returns on their first trading day making it difficult to explain it without including investor enthusiasm and thus investor sentiment. An example of this is the return on Netscape the day it went public, the return was 108 percent. Interestingly, IPO first-day returns are not idiosyncratic, average first-day returns display peaks and troughs which are highly correlated with the IPO volume and other sentiment proxies which are fundamentally not related (Baker & Wurgler, 2007). - Net equity share in new issues A broad measure of equity financing activity is the equity share of total equity and debt issues by all corporations. The measure does not only include IPOs, but all equity offerings. High values of the equity share indicate low stock market returns, and thus suggest that this pattern reflects firms shifting successfully between equity and debt to reduce the overall cost of capital (Baker and Wurgler, 2000). They state that this pattern does not need to imply that individual firms of their managers can actually predict prices on the market as a whole but rather that correlated mispricings across firms may to lead to correlated managerial actions. This in turn forecasts correlated correction of mispricings that is, it will forecast market returns (Baker and Wurgler, 2007). Thus summarizing in Table 1, these are the six measures that create the sentiment index by Baker and Wurgler (2007); Table 1: Measure Trading volume Dividend premium Closed-end fund discount Initial public offering volume First-day returns on IPOs Equity issues over total new issues Sentiment index variables Baker and Wurgler (2007) Description Relative trading volume on the NYSE Relative value of safe stocks How much investors want to invest in equities Number of IPOs The first-day returns on IPOs Demand for equity Page 13

15 Some of the sentiment proxies reflect economic fundamentals at least to some extent. An example of this is the IPO volume, this depends on prevailing investment opportunities. In order to remove these influence, at least partially, Baker and Wurgler (2007) regress each proxy on a set of macroeconomic indicators (growth in industrial production, real growth in durable, nondurable, and services consumptions, growth in employment, and an NBER recession indicator) and then use the residuals from these regressions as the sentiment proxies. In order to achieve a better index, another adjustment was made. As these sentiment proxies, given that major macroeconomic influences have already been removed, will have a common sentiment component, the remaining idiosyncrasies will be removed by averaging them together into an index (Baker & Wurgler, 2007). The sentiment index The historical value of this sentiment index is shown in the graph below: Here we can see that the sentiment index, as established by Baker and Wurgler (2007) matches well observed pattern of stock market bubble and burst periods. We go further into this data when we try to establish relationships between the different variables in this thesis. Page 14

16 2.2 Defining the different risk premiums This part will start with introducing the market risk premium. It will explain what it is, and what its role is in today s economics. Hereafter, we will go into the variance premium. The average return on equity has far exceeded the average returns of the risk-free asset, defined as short-term virtually default-free debt (Mehra and Prescott, 1985). Mehra and Prescott (1985) found that over a ninety-year period, from , the average real annual yield on the S&P 500 has been seven percent, where the average yield on short-term debt was less than one percent. They investigated whether this difference could be due to transactions costs, liquidity constraints and other frictions, but there conclusion is that it is most likely due to some equity premium otherwise referred to as the risk premium. The market risk premium can be defined as the spread between investor required returns on safe and average risk assets (Harris and Marston, 1999). It has played a central role in finance for a long period of time, and is a key factor in asset allocation decisions. It is there to help determine the portfolio mix of debt and equity. Furthermore, the market risk premium plays a crucial role in the Capital Asset Pricing Model (CAPM) which is already discussed in this thesis. More recently, the practical significance of estimating the market premium has become more important as more financial frameworks are being used to analyze corporate and investment performance Defining variance premium The first area of interest is the variance premium. In all investment decisions, investors face a certain risk. This can be measured easily by looking at either volatility or variance; however, we can make a case in saying that volatility on itself is uncertain. For example, the U.S. market has had very large volatility shocks in events like the 9/11 incident and the recent financial crisis. These were times with extremely high volatility rates which in turn tend to be negatively related to index returns. Therefore, we can say that investors do not only face risk from returns, but also from the variance in these returns. The variance premium is not only a theoretically defined premium, but there is also a market for this. The first variance related contracts date from the mid-1990s, but it was not until the late 1990s that investors started trading these contracts on a larger scale. The over-the-counter market, which trades the most common volatility contract, the variance swap, has increased Page 15

17 significantly in size. It shows the importance investors assign to the variance premium. The variance swap its payoff is equal to the difference between future realized variance over a given period and a predetermined rate and is similar to other forwards contracts. The rate for the variance swap sets the value of the contract equal to zero. Therefore, it represents the risk-neutral expectation of future variance for a certain period of time (Zhou, 2010). Whaley (2000), defined the variance premium as being the market gauge of fear. It is measured by the Chicago Board Options Exchange s market volatility index, or shortly the VIX. The VIX itself is determined by investors and express their view about the expected future stock market volatility, and can be seen as the higher the value of the VIX, the greater the fear. To determine what the variance premium is influenced by, it is important to understand the concept of risk-aversion. As implied by the market risk premium, people in general tend to be risk-averse. When an investor buys a call option, someone must pay the premium to own this call option, this has to do with the holder of the underlying asset taking the risk. If the asset drops in value, the buyer of the option will not exercise the options and solely loses what he initially paid. However, the person who holds the asset does lose money and wants to be compensated for the risk he takes, and therefore a premium is paid. This also goes for the variance premium. Zhou (2009) says that the driving force behind the volatility spreads is the notion that risk-neutral probabilities are physical probabilities revised by investor s risk preference which is determined by the pricing kernel, which is a stochastic discount factor. This stochastic discount factor reflects the fact that the price of an asset can be computed by discounting future cash flows by the stochastic factor, and then taking the expectation (Back, 2010). Thus, in order to cover extreme losses, these investors are willing to buy protection against exposure to extreme losses. In turn, this desire drives up the risk-neutral probability relatively to the actual probability of occurrence (Bakshi and Madan, 2006). In this paper we will use data on the variance premium as provided by Londono (2010), who calculated it as the difference between the VIX squared and the expected realized variance, where the expected realized variance is calculated using auto-regression analysis. Page 16

18 2.3 The relationship between sentiment and the risk premium In this part the relationship between the sentiment and the market risk premium will be discussed. First will be discussed why there is a relationship and hereafter the actual relationship will be discussed. Classical finance theory gave no role to investor sentiment, which suggested investors to be rational and would at all times diversify to optimize the statistical properties of their portfolios (Baker and Wurgler, 2006). This in turn leads to an equilibrium in which prices are equal to the rationally discounted value of future cash flows, and thus make the cross-section of returns dependent on the cross-section of systematic risks (Gomes, Kogan, and Zhang, 2003). There was some room for irrational investors; however their irrationality would be offset by arbitrageurs. In this paper, we suggest that there is room for sentiment and that it has effect on the risk premium. Therefore, we start with some theoretical basis. A mispricing is the outcome of both an uninformed demand shock as well as binding arbitrage constraints (Baker and Wurgler, 2006). As a result, a wave of sentiment is then predicted to have cross-sectional effects, as a rise (or lowering) of all prices by the same amount is not likely to be the case. Since sentiment is more likely to influence certain stocks, the influence on the risk premium of a whole market will likely be less prominent. This can be explained by saying that if sentiment does influence certain stocks in a positive way, it will likely influence contrary stocks in an opposite way. However, this research takes the stand that sentiment premium has a negative influence on the general risk premium. This is because if sentiment is high, it is likely that certain stocks are to be overpriced and others underpriced, and in response arbitrageurs cannot fully offset this mispricing (De Long et al., 1990), thus enforcing limits to arbitrage. This is where arbitrageurs are likely to be risk averse and will have reasonably short horizons (De Long et al., 1990), and as a result, the arbitrageurs willingness to take positions against these so called noise traders, which are the irrational investors, will be limited. This consists of different sources of risk; the first one is the fundamental risk that these arbitrageurs face. Figlewski (1979) showed that it can take a very long time for noise traders to lose most of their money if arbitrageurs must bear fundamental risk in betting against them and therefore, take limited positions. This is shown by for example Shiller et al. (1984), who shows that aversion to this fundamental risk can, by itself, severely limit arbitrage even when arbitrageurs have infinite horizons. Page 17

19 There is another source of risk, which is the risk that noise traders beliefs will not revert to their mean for a long period of time and can even become more extreme (De Long et al., 1990). For example, if noise traders are pessimistic about a certain asset and hereby drive down the price, an arbitrageur can catch up on this and act upon this mispricing. However, he must keep in mind that these noise traders can become even more pessimistic and drive down the price even further. Now there is a problem; these noise traders can stay irrational for a longer period than the arbitrageurs can stay solvent. So then, if the arbitrageur has to liquidate before the price goes up, he will suffer a loss. Fear of this loss should then in turn, limit his original arbitrage position (De Long et al. 1990). Of course, this is not only the case of pessimistic behavior by noise traders, but also when they are optimistic but then exactly opposite. Since the noise traders future opinions are unpredictable, prices can diverge significantly from fundamental values even when there is no fundamental risk (De Long et al., 1990). An example of this fundamental mispricing is in the Royal Dutch Shell stock. Royal Dutch and Shell are twin shares, as they merged on a 60:40 basis while remaining separate and distinct entities (Froot and Dabora, 1999). Thus, if the market values of securities were equal to the net present values of future cash flows, the value of Royal Dutch should always be one-and-a-half times the value of Shell. However this is not the case as shown in the table below which shows the Log Deviations from Royal Dutch/Shell parity (taken from Froot and Dabora, 1999): Figure 1: Royal Dutch/Shell parity Page 18

20 So, if we apply this to our thesis, we should see that on general, the market premium will be lower in case of a positive sentiment and vice versa. Thus our first hypothesis is: - Sentiment has a negative influence on the market risk premium As often the relationship between sentiment and stock returns, rather than the risk premium, is researched it would be an addition to research what part of the return it is mostly connected to. Therefore, the first part investigated is the risk premium. However, the return consist of as well the risk premium as the risk free rate, if we look at the CAPM model. As the relationship between the return and sentiment is established and we do not know yet whether a relationship between sentiment and the risk premium exists, it would be an addition to the research to see whether the sentiment explains the risk-free rate part of the return rather than the risk premium part. Therefore, our third hypothesis will focus on the relationship between sentiment and the risk-free rate. Therefore, sentiment has a positive influence on the risk-free rate since noise traders tend to overinvest in risky securities rather than risk-free alternatives thus lowering demand for risk-free alternatives and increasing returns. Thus; - Sentiment has a positive influence on the risk-free rate Investor sentiment in the behavioral asset pricing literature has been thoroughly discussed already in this paper. As previously mentioned, different researchers have focused on different aspects. However, another missing piece in this literature is the research whether extreme levels of sentiment have a different effect relatively to regular levels of sentiment. Baker and Wurgler (2007) roughly suggest that the index generally captures the bubbles and crashes written by authors such as Brown (1991) and others. They suggest that eyeball tests thus explains the differences. However, in order to know whether these extreme levels of investment actually imply also higher predictability of the stock returns is not tested. Therefore, we will focus on the extreme levels of sentiment, and whether its predictability of the market risk premium is actually empirically present. Therefore, the fourth hypothesis is as follows: - Extreme levels of positive (negative) sentiment explain a lower (higher) market risk premium Page 19

21 The influence of sentiment on individual stocks Another important relationship between sentiment and the risk premium concerns the influence of sentiment on individual stocks. First of all, let us look at the possibility that sentiment-based demand shocks vary across firms, which already has been suggested in this paper, however arbitrage is equally difficult across firms. Let us look at investor sentiment as the propensity to speculate by an investor and this investor is akin to play the lottery. Then sentiment in this case is almost by definition a higher demand for more speculative stocks, and therefore such speculative stocks have temporarily higher returns. Then the question is; what makes some stocks more speculative than others? Baker and Wurgler (2007) believe that the main reason lies in the difficulty and subjectivity of determining true values of certain stocks. Let us look at a young and currently unprofitable, however, potentially extremely profitable growth firm. It has no earnings history and it has a highly uncertain future, this allows investors to defend valuations ranging from very low to very high, as befits their prevailing sentiment. During a bubble, as speculation is high, investment bankers can join the corus arguing for high valuations (Baker and Wurgler, 2007). However, on the other side, a firm which has a long earnings history, tangible assets, and stable dividends is much less subjective to value and therefore, less prone to sentiment. Many different psychological arguments can be given in order to defend this statement ranging from the effect of overconfidence (Daniel, Hirshleifer, and Subrahmanyman, 1998), representativeness and conservatism (Barberis et al, 1998) and many more. Miller (1977) states that, even when investors have the same basic information, opinions can vary largely. Looking at it differently, it is possible to view investor sentiment as solely being optimism or pessimism about stocks in general, and allow the limits to arbitrage to vary across stocks. Looking at arbitrage, as previously discussed in this research, it is shown by research that some stocks can be particularly risky and costly to arbitrage. These stocks are; (1) young, (2) small, (3), unprofitable or, (4) experience extreme growth. These stocks are on general more costly to both buy and to sell short (D Avolio, 2002). These stocks have a higher degree of idiosyncratic variation in their returns, which in turn makes betting on them riskier (Wurgler and Zhuravskaya, 2002). These stocks higher volatility can then lead to second-guessing by the investors who in turn provide the funds to the arbitrageur, which then leads to withdrawals from contrarian Page 20

22 arbitrageurs exactly when the mispricing is the greatest (Shleifer and Vishny, 1997). Therefore, we can expect that sentiment has a greater effect on these types of stocks. The take-away here, is that the securities that are often difficult to value also tend to be difficult to arbitrage. The influence of sentiment on different industries After having established a connection between sentiment and individual stocks, the focus will be put on the effect of sentiment on different industries. A growing body of research addresses the impact of sentiment on for example initial public offerings (Bradley et al., 2009), liquidity commonality (Choe & Yang, 2010), soccer game results (Palomino et al., 2009) and stock split announcement (Kim & Byun, 2010). Other related literature explores how sentiment affects the valuation on the firm-level as explained in previous paragraph. However, few studies actually investigated what the effect of sentiment is on industry differences. For individual investors, knowledge of the relative sensitivities of industry stock returns, or in our case risk premium in special, to sentiment can be of great benefit for risk management purposes. Therefore, this paragraph will focus on the influence of sentiment on the different industries their risk premium which is in accordance with our general research. However, in order to separate the whole economy into different sectors we must first decide on how many different industries will suit this paper best. Deciding which firms belong to an industry is not an easy task. Standard Industrial Classification (SIC) codes, which aggregate firms selling related end-product or which use similar production processes into an industry, are often used in order to answer this question (Chan et al., 2007). Fama and French (1997) used the four-digit SIC codes and organized them into a different number of industries. Their classifications have been highly influential, as it is widely used in many academic studies on asset pricing (for example; Daniel and Titman (2006), Ferson and Harvey (1990)). The Fama-French specification of industries is virtually the default choice in academic studies, and is therefore also applied in this research. However, they have different classifications ranging from five industry portfolios (consumer, manufacturing, high-tech, healthcare, other) to 49 industry portfolios. For this research their classification of twelve different industries will be used as more will lead to an over complication of the data and will be of no added value. These industries are summarized in the table below: Page 21

23 Table 2: Twelve industry portfolios As defined by Fama and French Industry Examples Abbreviation Consumer Non-Durables Food, tobacco, textiles, apparel, leather, toys Nodur Consumer Durables Cars, TV s, furniture, household appliances Durbl Manufacturing Machinery, trucks, planes, paper, com printing Manuf Energy Gas, and goal extraction and products Enrgy Chemicals Chemicals and allied products Chems Business Equipment Computers, software, electronic equipment Buseq Telecommunications Telephone and television transmission Telcm Utilities Utilities Utils Shops Retail, and some services (laundries, repair) Shops Health Healthcare, medical equipment, and drugs Hlth Money Finance Money Other Mines, construction, hotels, entertainment Other As previously discussed, the available empirical literature supports the pricing implication of sentiment risk across several developed markets, the question of generalization of sentiment pricing still remains. Industry analysis rarely been explored in the behavioral asset pricing literature (Dash and Mahakud, 2013). Looking from the view point of the practitioner, active investment strategy explores tactical asset allocation with respect to the industry analysis. So, although behavioral asset pricing researchers and active investment practitioners share a common belief of the inefficient market, the commonality of sentiment pricing across industry groups has rarely been a source of research (Dash and Mahakud, 2013). As, in a financial market, the assets are perfect substitutes and supply and demand of an asset is infinitely elastic, the effect of sentiment at the aggregate market level may have a subtle effect, but the question is how this applies to the industry level. Therefore, in the fifth hypothesis we will test the following statement; - Industry portfolios are affected differently by sentiment Page 22

24 The influence of sentiment on portfolios sorted by size and value In addition to finding industry differences, it might be interesting to look at other types of portfolios. The famous Fama and French (1992) paper researched the influence of size and value on returns on stocks. They established two anomalies based on size and value, size was measured by market capitalization and value as market equity over book equity. Their research showed promising results of the influence of size and value on returns on stocks. They found that smaller stocks have higher returns than bigger stocks while stocks with higher value perform better than growth stocks. Later Carhart (1997) added the momentum factor, which captures the trend that stocks which perform well in the recent past tend to perform better in the narrow future and vice versa. In addition to this model this research will try to find significance in adding the sentiment factor to the model to enrich it. Therefore our next hypothesis will be: - Sentiment is a significant addition to the Carhart 4-factor model 2.4 The relationship between sentiment and the variance premium In order to establish a relationship between sentiment and the variance premium an in depth view of the influences of both is required. The variance premium is measured as the VIX squared, minus the expected realized variance (Londono, 2010). The variance premium tends to be higher in times of crises and high volatility. If we recapture sentiment, the index takes a high value when the trading volume, on the NYSE, is high. It is hard to define whether there is a relation to the variance premium in general, but we can suggest a connection between this part of the sentiment index and the volatility index. On general we can say that there is a positive relationship between volatility and volume of sales (Shalen, 1993). The next component of the sentiment index is the dividend premium, which is the relative value of safe stocks. As already discussed it has an inverse relationship to sentiment. The relationship with volatility seems to be rather clear, as in times of high volatility; investors are willing to pay a premium for safer assets (Baker and Wurgler, 2004). Thus this would suggest that in case of a higher volatility, the dividend premium would also be higher. The next variable in the sentiment index is the closed-end fund discount. It is the difference between the net asset value of a fund s actual security holdings and the fund s market price (Baker and Wurgler, 2007). If investors are bearish the discount tends to be increasing (Baker Page 23

25 and Wurgler, 2007). However, no clear relationship to volatility can be found, it is known that the volatility of the closed-end fund price itself is considerably higher than that of its underlying assets (Pontiff, 1997), but no clear relation to the general volatility has been established. The demand for initial public offerings is often seen as an indicator for sentiment. The suggestion made by theory is that, during times of high volatility, companies are less likely to go for an initial public offering. Thus, this would suggest that there is a negative relationship between the number of IPOs and volatility. Closely related to this variable, is the first day returns on IPOs which is also considered to be a measurement for sentiment, as if sentiment is high, returns on IPOs tend to be higher as well. However, the relationship towards volatility is unclear. The last indicator used in the sentiment index, is the equity issues over total new issues. It can be translated into the demand for equity, this relationship seems to be easily understood, as people in times of high volatility tend to demand safer goods, and therefore, the demand for equity will drop during high volatility periods. In order to better understand this paragraph, it will be summarized in one table. Table 3: Overview relation sentiment and the variance premium Contains: Influence on sentiment: Influence on volatility: Suggested relationship: Trading volume Positive Positive Positive Dividend premium Negative Positive Negative Closed-end fund Negative Unclear Unclear discount Initial public offerings Positive Unclear Unclear volume First day return on initial public offerings Positive Negative/Unclear Unclear In this paragraph we have looked at the relationship of the separate indicators of the sentiment index as used by Baker and Wurgler (2007) and looked at its relation to the variance premium. However, since there was no theory available the focus has been on volatility instead of the variance premium as an indicator of can be expected from the empirical research later in this Page 24

26 paper. Looking at Table 2, we cannot see any clear directions of what we can expect. There are positive as well as negative relationships and three unclear ones. Therefore, it will be hard to establish a hypothesis that will likely be tested with significant result, however, empirical results can be conclusive and therefore this will be tested; - There is a negative relationship between sentiment and the variance premium Page 25

27 3. Methodology and Data 3.1 Research Design This chapter will focus on the methodology and the data that will be used throughout this research. It will explain what sampling strategies have been applied to protect the research from biases. Furthermore, it will give an elaboration on several methodological issues related to the regression analysis. Hereafter, an explanation of the dependent, independent and control variables is given in order to provide the reader with a detailed explanation of the inclusion of these variables Purpose of the research The purpose of this research is to give answer to the problem statement which is given in Part II of this thesis, in Chapter 1.2. It will do so by testing several hypotheses which have been formulated on the basis of the previous literature given in Chapter 1.3. Previous literature has focused on the relation between sentiment and the returns on stocks; however, there is a gap in the theory concerning the effect on the risk premium. Connections have been established between sentiment and the return; however which part of the return is explained is not investigated. Therefore, this research will focus on the risk premium rather than the returns and it will also indicate the effect of sentiment on the other part of return, namely the risk-free rate. Furthermore, research has focused on the influence of sentiment on stock characteristics such as age, profitability and others. There is a gap in the research on the market-wide influence of sentiment. Furthermore, no research has focused on the influence of sentiment on industry portfolios in the United States which can be of great benefit to investors in their allocation of resources as allocating resources based on stock characteristics is very time consuming and often investing in individual stocks is very expensive. This research will attempt to mature the research on these points. Maturity is reached in research when three different criteria have been met; (1) a substantial number of studies have been conducted, (2) these previously named studies have generated reasonably consistent and interpretable findings, and (3) the research has led to a general consensus concerning the nature of key relationships (Palich, Cardinal, Miller, 2000). The results from this research will be able to add to the works of for example Baker and Wurgler (2007), DeLong et al. (1990). Page 26

28 3.1.2 Type of research This study will use an empirical investigation, an explanatory research using secondary quantitative data on the effects of sentiment on the different risk premium with mainly focusing on the (market) risk premium. This thesis contributes to the existing literature in that extends it by looking at the effect of sentiment on risk premium instead of returns on stocks and by looking at market and industry wide effects of sentiment. The main mode of analysis used in this thesis is a regression analysis with the use of panel data. The data used are secondary quantitative data gathered with the help of the Wharton Research Data Services (WRDS) from the University of Pennsylvania. In addition Thomson Reuters DataStream will be used to collect historical data on specific stocks which have been selected for the research. The data that will be considered ranges from 1966 to Sampling strategy The focus of this thesis is on the United States market. Therefore, we look at the NYSE, NASDAQ and AMEX and other markets are excluded. In order to be able to test the effect of the sentiment index on the market risk premium we needed to size down the sample to the years because of the limitation of data from the sentiment index (Baker and Wurgler, 2007). Therefore our population consists of pieces of data which include the market risk premium (as defined by Fama and French (1992)), sentiment index (Baker and Wurgler, 2007), small-minus-big factors, high-minus-low (Fama and French, 1992) and momentum (Carhart, 1997). This leads to 540 consecutive months of data, our units. This data is not subject to any sampling bias as we include the whole United States market and have no exceptions. All the data is complete for all variables and is therefore representative for the whole United States market. Therefore, there is no need to apply any sampling technique despite for having the United States market rather than the World Market. 3.2 Data This section will focus on the method of how the data has been operationalized into the dependent, independent and control variables. The data used in this is thesis is in the form of a panel. Panel datasets have the important characteristic in that they explain cross-sectional as well as longitudinal variation, which means that this allows for explaining differences between stocks as well as differences within the same stock over time. In this research the cross-sectional aspect Page 27