Investor sentiment and stock market returns : the influence of the state of the economy

(1)

Investor sentiment and stock market returns: the

influence of the state of the economy

_____________________________________________

Duncan Jansen

10015620

Bachelor Economics & Business

January 2014

Supervisor: P. Trietsch

(2)

2

Abstract

This thesis examines if the economic state has a significant influence on the relationship between investor sentiment and US stock market returns. Behavioural assumptions are incorporated in a multifactor asset pricing model to test this relationship for the period 1997-2010, thereby capturing two of the most important financial bubbles of the last decades. Previous research has shown that investor sentiment is present in the market and affects specific types of stocks. In literature there is however debate whether sentiment also affects aggregated return and when sentiment affects aggregate returns. This research concludes that investor sentiment significantly affects returns during an expansion state, while investor sentiment does not significantly affects returns during a recession state.

Keywords: Behavioural finance, Investor sentiment, asset pricing model, stock market

(3)

3

1. Introduction

Several events in the world of finance including the recent financial turmoil and the dot.com bubble have not only caused dramatic changes in stock prices, but also challenged the explanation offered by neoclassical finance. The neoclassical assumption that “unemotional investors always force capital market prices to equal rational present value of expected future cash flows” does no longer seem to offer perfect insight into asset pricing anomalies (Baker & Wurgler 2007). There is therefore an increasing need to behavioralize finance and replace the neoclassical assumptions with behavioural counterparts (Shefrin, 2010).

Researchers in behavioural finance have been working on an alternative model to capture these anomalies based on the assumption that there is investor sentiment present in the market. Investor sentiment, defined broadly, is “the propensity to speculate” (Baker & Wurgler, 2007). When the propensity to speculate is high, it is both costly and risky for rational investors to bet against these sentimental investors, which results in irrational investors driving up prices above fundamental value (Shleifer & Vishny, 1997). When irrational investors drive out the rational investors in the market, bubbles can emerge that lead to severe recessions. It is therefore important to determine the effect investor sentiment has on stock returns.

Several researchers such as Baker & Wurgler (2007), Lemmon & Portniaguina (2006) and Qiu & Welsch (2006) have investigated the relationship between investor sentiment and stock market returns. Their findings conclude that investor sentiment has the most predictive power on stocks that are difficult to value and are more prone to speculation. There is however no definite conclusion about the relationship between investor sentiment and aggregate returns. In more recent research, Chung, Hung and Yeh (2012) claim that investor uncertainty about the state of the economy predicts the presence of asymmetries in the predictive ability of investor sentiment over different economic states. Over the economic cycle, they conclude that investor sentiment has predictive power only in the expansion state and not in a recession state on small stock, growth stock, non-earning stock and non-dividend paying stock. These finding imply that the economic state is an important factor that should be taken into consideration when investigating the relationship between investor sentiment and stock market returns. Besides the research done by Chung, Hung and Yeh (2012), little research takes this factor into account. This research therefore aims to investigate the influence of the economic state on the relationship between investor sentiment and US stock market returns.

This research aims to answer this research question by performing a time series regression on the Carhart four factor model, including investor sentiment for the period 1997-2010, thereby capturing the two most recent financial bubbles. Data for the US stock market returns and factor model are conducted from the Fama & French website. The US stock market returns are represented by 25 portfolios formed on size and book to market ratio, conducted from the CRSP database. Data for investor sentiment is conducted from Jeffrey Wurgler’s website. An interaction term is created to capture the effect of the economic state on the relationship between investor sentiment and US

(5)

5

stock market returns and is added to the Carhart four factor model. The economic state is determined by the NBER business cycle status listed on their website.

After the introduction, this research will start with a literature overview in chapter 2. In this chapter, traditional asset pricing theory will be discussed first. Then the influence of investor sentiment and how it is measured will be clarified. After that two most recent financial bubbles will be discussed and its implications on investor sentiment. In chapter 3, the collection of data and the methodology used in this research will be explained. At the end of this section, testable hypothesis will be formed based on the conclusions drawn from the literature. In chapter 4, first the statistics will be summarized. After that, the results from the time series regression will be discussed and analysed. In the last chapter, concluding remarks will be made as well as limitations to the research and further research recommendations.

2. Literature review

Several crises in the last decades did not only cause dramatic change in stock prices, but also challenged the explanation offered by neoclassical asset pricing models. The neoclassical assumption that all investors are rational has difficulties explaining asset-pricing anomalies. Black (1986), DeLong et all (1990) and Schleifer & Vishny (1997) proposed behavioural assumptions and measurements that capture behavioural aspects, defined as investor sentiment. Researchers are still debating about the correct measure of investor sentiment but agree that investor sentiment exists in the market and may inhibit some predictive power on aggregated stock returns. More recent research done by Chung, Hung and Yeh (2012) shows that investor sentiment only has predictive power in the expansion state and not in the recession state. This in contradiction to previous researcher such as Brown & Cliff (2005), Baker & Wurgler (2007) & Qiu & Welsch (2006) who find that investor sentiment may inhibit predictive power regardless of the economic state.

2.1 Neoclassical asset pricing models

The literature of neoclassical asset pricing model goes back to the Capital Asset Pricing model (CAPM) of Sharpe (1964) and Lintner (1965), who in short developed a model where returns are explained by only one factor, the market risk premium, which is a measurement of systematic risk. According to this model all other risk is non-systematic and can be diversified away by holding an efficient portfolio. The CAPM model is a widely applied model because of its easiness to use, but fails to explain the returns on stocks with certain firm characteristics such as a size effect (Banz, 1981) and a value effect (Chan, Hamao & Lakonishok, 1991).

The CAPM model is later reviewed by Fama & French (1992), who extended the CAPM model with a size factor (small minus big) and a value factor (high minus low) to capture cross-sectional variations in average stock returns. By including these two additional factors the model tends to adjust for the fact that value and small cap stocks outperform the market on a regular basis. In later research this model is expanded with a momentum factor (Jagadeesh and Titman, 1993). They find substantial evidence that

(6)

6

indicates that stocks that perform the best over a three-to-twelve month period tend to continue to perform well over the subsequent three-to-twelve month period and vice versa. Taking into account the momentum factor, Carhart (1997) added the momentum factor to the three factor model of Fama & French (1992) and constructed the Carhart four factor model:

Rpt-Rft = αt+ β(Rmkt – Rf)t + s SMBt + h HMLt + m MOMt + εt

With Rp-Rf being portfolio returns over the risk free rate, β the market risk premium,

SMB the size factor, HML the value factor and MOM the momentum factor.

In this model there is no room for investor sentiment as an explanatory variable for returns. This can be explained by the “Efficient Market Hypothesis” of Fama (1970), which states that all available information is reflected in market prices. Investors trade according to the information available, driving asset prices to their intrinsic value. This implies that erroneous beliefs about future cash flows and risks are inefficient and that their demands are offset by arbitrageurs with no significant impact on prices.

2.2 Sentiment and stock returns

Many events in the world of finance, including the two most recent economic crises have questioned the efficient market hypothesis. Each of these events caused a dramatic level of stock prices returns and the neoclassical finance models have considerable difficulties fitting these patterns (Baker & Wurgler, 2006). As a result, several researches have come up with assumptions different from the efficient market hypothesis. One example is the “noise” theory developed by Black (1986), who points out that noise in financial markets is possible and can cause markets to be inefficient. Another theory developed by DeLong et al. (1990) is the “noise traders theory” suggesting that the unpredictability of noise traders belief creates a risk in the prices of assets in the short run that withdrawals rational arbitrageurs from betting against these assets. As a result, stock market prices can diver significantly from fundamental value. Betting against noise traders is costly and risky, which results in the fact that rational investors are not as aggressive in forcing prices to fundamental value as traditional finance theory suggest (Shleifer & Vishny, 1997).

Researchers in behavioural finance have therefore been working to augment the neoclassical model with these assumptions by focussing on investor sentiment. Investor sentiment is broadly defined. Anderson, Ghysels and Jeurgens (2005) defined sentiment as the erroneous beliefs about future cash flows and risks. In later research, Baker & Wurgler (2006) define investor sentiment as “the propensity to speculate”. They state that high sentiment should be associated with high stock valuations, especially for stocks whose valuations are highly subjective and are difficult to arbitrage, since these stocks are more prone to speculation. On the other hand, low sentiment should be associated with stocks that are easy to value and easy to arbitrage. Baker & Wurgler

(7)

7

(2006) summarize this perspective into a simple, unified view of the effects of sentiment on stocks (figure 1).

Figure 1: Theoretical effects of investor sentiment on different types of stocks. Stocks that are more speculative are overvalued

in a high sentiment state and undervalued in a low sentiment state. Stocks that are safe and less prone to speculation show an inverse relationship. The x-axis orders the stock according to how difficult they are to value and arbitrage, the y-axis measures prices with P* being fundamental values. The lines illustrate the hypotheses about how stock valuations are effected by sentiment.

When sentiment increases, all stock prices go up, but according to Baker & Wurgler (2006) some stocks more than others. The effect of sentiment on aggregate returns will be muted because stocks are not all moving in the same direction. According to these statements, they conclude that when sentiment is low at the beginning of the period, subsequent returns tend to be relatively high on small stocks, young stocks, high volatility stocks, unprofitable stocks, non dividend-paying stocks, extreme-growth stock and financial distressed stocks. When sentiment is high at the beginning of the period, the pattern largely reverse, suggesting that these stocks are overpriced in high-sentiment state (Baker & Wurgler, 2007). Both Lemmon & Portniaguina (2006) and Qiu & Welsch (2006) draw similar conclusions, although a different measure of sentiment has been used.

These papers find strong evidence on the predictive power of investor sentiment on small stocks and conclude that that there is a positive relationship between the level of investor sentiment in the market and subsequent stock market returns. There is however still some uncertainty whether this relationship also holds for aggregated stock market returns. Baker & Wurgler (2006) state that investor sentiment may inhibit predictive power over aggregated stock returns. Brown & Cliff (2005) and Charoenrook (2005) on the other hand, find supporting evidence that investor sentiment predicts aggregated market returns over the next 1-3 years. According to Brown & Cliff (2005) their findings have some important implications. First implication is that the theories of

(8)

8

behavioural finance do in fact affect asset level prices. Second implication is that asset-pricing models should consider the role of investor sentiment. Finally, they state that investors should be aware of the impact sentiment can have on their investment strategies.

2.3 Sentiment measures

The mentioned literature claims that investor sentiment exists in the market. There is however no consistency about how investor sentiment is measured or how to proxy for it. Two types of measurements can be distinguished from the literature; direct measures and indirect measures. A direct sentiment measure approach uses survey results as measurement, while an indirect measure approach uses financial proxies as measurement.

2.3.1 Direct measurement of sentiment

Direct measurements are widely used in literature and is seen as a bottom up approach (Baker & Wurgler, 2006). Brown & Cliff (2004) uses the American Associate of Individual investor survey (AAII) as a measurement for individual investor sentiment and uses the Investor Intelligence survey (II) as a measurement for institutional sentiment. The AAII asks individual investors what will be their expectations for the stock market in the next 6 months (up, down or the same). The II marks newsletters as bullish bearish or neutral based on the expectations of future market movements. According to the findings of Brown & Cliff (2004), these two surveys are related to other popular direct measures of investor sentiment proxies and have a strong co-movement with the market. In Brown & Cliff (2005) the II survey used as sentiment measure predicts stock market returns over the next 1-3 years.

An indirect measurement is also used in the research of Qiu & Welsch (2006). They examine two potential proxies for investor sentiment, one direct and one indirect measure; the consumer confidence index (CC) and the closed end fund discount (CEFD). They find that the indirect measure CEFD is not a proxy for investor sentiment and a direct measure of consumer confidence contains a component related to investor sentiment that correlates with the excess return on small firms.

(9)

9 2.3.2 Indirect measurement of sentiment

According to Baker & Wurgler (2006), direct measurements of investor sentiment consist biases due to the fact that there is a possibility that investors fill in the survey that is not consistent with their own actions in the financial markets. Instead of using a direct measure, they use a top-down approach which focuses on aggregate sentiment and traces its effect back to market returns and individual stocks. An index of investor sentiment is composited based on common variations in six underlying proxies for sentiment; the closed-end fund discount (CEFD), New York Stock Exchange turnover (TURN), the number of Initial Public Offerings (NIPO’s), the average first day returns on IPO’s (RIPO), the equity share in new issues (S) and the dividend premium (PDND). The six sentiment have a common sentiment component because the major macroeconomic influences have been removed. The remaining idiosyncrasy components are filtered out by averaging them in an index. The index is standardized with a mean of zero and a unit variance. The unit variance determines the level of investor sentiment based on the variations in the six underlying proxies. Descriptions of the underlying proxies are listed in table 1 of the appendix. With this index, Baker & Wurgler find strong co-movement with the stock market and predictive power on stock that are hard to value and arbitrage.

A different indirect measurement is used by Lemmon & Portniaguina (2006), who use a consumer confidence index as an indirect measure of investor optimism. They estimate the components of consumer confidence related to economic fundamentals and investor sentiment by freeing the consumer confidence index from macroeconomic variables. Investor sentiment measured using the component of consumer confidence related to sentiment has predictive power on small stocks and stock with low institutional ownership.

Both the indirect investor sentiment measurement of Lemmon & Portniaguina (2006) and Baker & Wurgler (2006) predict returns on small stock. These findings leave room for discussion which of the two is the most unbiased measurement. Lemmon & Portniaguina (2006) check whether the two measurements are correlated and find that the sentiment component of consumer confidence is not strongly related to the sentiment index of Baker & Wurgler, suggesting that the different measures either capture some unrelated component of investor sentiment or fail altogether to capture some important aspects of sentiment. Chung, Hung and Yeh (2012) apply both the measurement of Baker & Wurgler as the measurement of Lemmon & Portniaguina in their research. They find that both measurements are able to predict stock market returns in a high volatility recession state, while there is no evidence of the predictability in a low volatility expansion state. There is however a difference, since the Baker & Wurgler index is capable of predicting stock returns for all portifolios, while consumer confidence is only able to predict returns on size portfolios. These findings suggest that the investor sentiment index of Baker & Wurgler is the most unbiased measurement of investor sentiment.

(10)

10

2.4 Recent bubbles and implications on sentiment

In this section the two most severe recessions of the last decades are briefly summarized and the role of market inefficiency and investor sentiment on these recessions is discussed. After this brief overview, the role a recession might play in the relationship between investor sentiment and stock returns is discussed.

The first important recession is caused by the late-1990’s technology bubble, commonly referred to as the dot com bubble. Investor sentiment was broadly high before its burst in march 2000. As Ljunggivist & Wilhelm (2003) point out, 80% of the IPO’s in 1990 and 2000 had negative earnings per share. Olef & Richardson (2001) conclude that the persistence and fall of the internet stock prices can be attributed to heterogeneous agents with varying degrees of belief about asset payoffs. Furthermore, he argues that pessimistic investors were overwhelmed by optimistic investors. Even before the crash occurred, Chan, Karceski & Lakonishok (2000) conclude that the relative stock-price performance of growth stock could have been better explained by a behavioral explanation in favor of a rational one.

The second important recession of the last decades is the more recent global financial crisis, which was indirectly caused by the real estate bubble in the sub-prime mortgage market between 2006 and 2007. The real-estate bubble expanded because of the widely held irrational belief that real-estate prices would continue to rise (Shiller, 2008). Demyanyk & Van Hemert (2008) show in their research that between 2001 and 2006 there has been a significant decrease in underwriting quality of mortgages. Milan and Sufi (2008) conclude that these expansions of credit to less qualified lenders cannot be attributed to fundamentals alone. They find that in regions with the highest denial rates of mortgages in the mid 1990’s experienced the largest growth in mortgages after 2000 even though they had relatively slower gains in income and employment. On the other hand, kojucharov et all (2008) conclude in their paper that the boom in the subprime mortgage market may have occurred rationally given the information flow during the early years of the boom.

Baker & Wurgler (2006) find that their composited investor sentiment index shows co-movement with the market (figure 1). With respect to the change in the investor sentiment index, they find that in light of major speculative episodes, the volatility of sentiment rises in a speculative episode. This suggests that the relative influence of fundamentals and sentiment on aggregate returns changes over time.

(11)

11

Figure 1: Investor sentiment index over time. The index is standardized with a unit variance. The unit variance determines the level of investor sentiment based on the variations in the six underlying proxies.

The index co-moves with the most speculative episodes in history including the technology crash at the end of the 1990’s (Baker & Wurgler, 2006).

Chung, Hung and Yeh (2012), use the suggestion that the relative influence of fundamentals and sentiment on aggregate returns changes over time. In their research they investigate the predictive ability of investor sentiment on the cross-section of stock returns across different economic states. They argue that a switch in economic regime challenges the predictive ability of sentiment on stock returns based on two notions. The first notion is that it is difficult to identify if a price change as a correction of mispricing is due to sentiment or if it is an adjustment with respect to an economic regime switch. The second notion is that there is investor uncertainty about the state of the economy, which leads to asymmetry in the predictive ability of sentiment across different economic states. When there is more uncertainty, investor beliefs are more influenced by news (Veronesi, 1999). They use a multivariate Markov-switching model to identify two economic states; a recession state where volatility is high and an expansion state where volatility is low. Their main findings are that investor sentiment performs predictive power on small stock, non-earning stock, growth stocks, and non-dividend paying stock only in an expansion state. Furthermore, in their research they also considered regime dummy variables in predictive regression. Their results with dummy variables suggest that their main findings continue to hold.

(12)

12

3. Methodology

This section is divided into three parts. First an overview of the collected data is given including motivation of the chosen data. Second, the methodology used in this research is discussed. Finally, testable hypothesis will be conducted.

3.1 Data collection and motivation

In order to test the Carhart four factor model including investor sentiment and the influence of the economic state, this research uses different data sources. The US stock market is represented by 25 portfolios formed on size and book to market ratio from the Fama & French database on a monthly basis. Fama & French have collected the data from the CRSP database. The monthly returns of the portfolios will be deducted by the risk free rate to capture the excess return of the portfolios. According to Baker & Wurgler (2006), the effect of sentiment on returns is higher for portfolios with equal-weighted returns than for portfolios with value-equal-weighted portfolios, because the effect of investor sentiment is stronger for stocks smaller in size. In this research portfolios with equally weighted returns are therefore used. The factors in the Carhart four factor model being; the market risk premium factor, size factor, value factor and momentum factor are also conducted from the Fama & French database on a monthly basis.

Findings from the literature suggest that the formed index by Baker & Wurgler (2006) is the most unbiased measurement of investor sentiment. In Baker & Wurgler (2007) a second onorthalized sentiment index is constructed where the underlying proxies have first been orhogonalized with respect to a set of macroeconomic conditions, after criticism that the effect of the sentiment index in Baker & Wurgler (2006) is partly related to business cycle variations. In both Baker & Wurgler (2007) as Chung, Hung & Yeh (2012) this measure for investor sentiment is lagged for one period and is used to test the effect investor sentiment has on stock returns. This research follows previous research by using the lagged sentiment index from Baker & Wurgler (2007). The data for the sentiment index is publicity available on the website of professor Wurgler. In order to test the influence of the economic state on the relationship between investor sentiment and US stock market returns, the economic states have to be distinguished by a variable. This research uses the NBER business cycle status as a dummy variable for the economic state. This value can be either 1, representing a recession state, or can be 0, representing an expansion state. The time span of this research is 1997-2010. This time span captures the two most important financial bubbles of the last two decades with the implications of sentiment present in these bubbles.

(13)

13

3.2 Methodology factor model

In order to test the relationship between investor sentiment and aggregated US stock market returns and the role of the economic state on this relationship, a time series regression is performed on a multifactor model. The dependent variables are the excess returns over time from 25 portfolios formed on size and book to market ratio. The independent variables are investor sentiment lagged for one month and an interaction term between a dummy variable for the economic state and investor sentiment. The control variables included in the regression are: the market risk premium (Sharpe, 1964), the size effect and value effect (Fama & French, 1992) and a momentum effect (Jagadeesh and Titman, 1993). Equation (1) represents the model on which the time series-regression is performed. Variable descriptions are listed in table 2 of the appendix.

Rpit-Rft = αi+ βi(Rmkt – Rf)t + si SMBt + hi HMLt + mi MOMt +yi SENTt-1+y2i C*SENTt-1+ εit (1)

i=1,2,3,..,25

A T-tests is performed on all the individual factors in the model to check if these factors influence the excess returns of the 25 portfolios.

According to the conclusions drawn from the literature, two testable hypotheses are constructed; one for the effect of sentiment and one for the influence of the economic state. The first hypothesis is constructed to conclude which model explains the dataset better, the neoclassical model where sentiment does not play a significant role, or the behavioural model, where sentiment does play a significant role in explaining excess stock returns. More formally stated:

H0: Investor sentiment does not have an effect on stock returns (yi =0)

H1: Investor sentiment does have an effect on stock returns (yi>0)

The second hypothesis is constructed to conclude whether the state of the economy has a significant impact on the relationship between investor sentiment and stock market returns. According to the literature, expectations are that there is a significant relationship during an expansion state, while there is no significant relationship in a recession state. The variable distinguishing these two states is the dummy variable C. Where C=1 for a recession state and C=0 for an expansion state.

C=0 ( yi ) SENT

(14)

14

The coefficient on the interaction term between investor sentiment and the dummy variable for the economic state (y2)should be significant. More formally stated:

H0: The state of the economy does not have a significant effect on the relationship between

investor sentiment and aggregated returns (γ6=0)

H1: The state of the economy does have a significant effect on the relationship between investor

sentiment and aggregated returns (γ6≠0)

4. Results

In this section, the results of the regressions in this sample are analyzed. First, the data used in this research is analyzed by looking at summary statistics and correlations. Then the time-series regression is discussed and analyzed. At the end of this section, some concluding remarks are made.

4.1 Summary statistics

In this research 25 portfolios are used sorted on size and book to market ratio. These portfolios where originally formed by Fama & French (1992). Table 1 shows the summary statistics of the portfolios over the whole sample.

Table 1: Mean and standard deviation of the excess return of 25 portfolios formed on size and book to market ratio for the sample period 1997-2010. Stock smaller in size show higher volatility

and higher returns than stocks bigger in size. Also stocks with a low book to market ratio tend to have higher volatility but lower returns than stocks with high book to market ratios.

Stocks smaller in size show higher returns and higher volatility than stocks bigger in size. Stocks with a low book to market ratio tend to have higher volatility but lower returns than stocks with high book to market ratios. This is consistent with Fama & French (1992) that small stock outperform big stock and stock with higher book to market ratio outperforms stock with low book to market ratios. The size factor and the value factor are present in this sample.

Low 2 3 4 High Low 2 3 4 High

Small 0,50 0,87 1,16 1,05 1,41 10,44 7,97 6,54 5,74 6,91 2 0,40 0,83 1,01 0,78 0,89 8,80 6,89 6,12 6,27 7,54 3 0,44 0,79 0,83 0,83 1,26 8,30 6,41 5,81 6,01 6,68 4 0,70 0,72 0,69 0,82 0,62 7,19 5,89 6,15 5,94 6,88 Big 0,45 0,62 0,54 0,58 0,62 5,95 5,30 5,49 5,48 5,95 Size

Ratio Book equity to market equity (BE/ME)

Summary Statistics

(15)

15

Sentiment tends to fluctuate over time. From the literature and figure 1 can be concluded that sentiment tends to be more present in the recession caused by the technology bubble than in the recent global financial crisis. Therefore the two periods are split to analyze this notion. 1997 – 2004 represents the period of the technology bubble and its subsequent crisis. 2004-2010 represents the period of the global financial crisis caused by the US real estate bubble.

Table 2:Summary statistics for the investor sentiment index and the change in the investor

sentiment index constructed by Baker & Wurgler (2007). On average, there is no sentiment present in

the market in the period 2004-2010. However, sentiment is very volatile in this period. In the period 1997-2004, there is a lot of sentiment in the market and is highly volatile. For the period 1997-2004 there is relatively more sentiment in the market than in the period 2004-2010. Over the whole sample period 1997-2004, investor sentiment is present in the market and varies in a wide range.

Figure 3: Graphical presentation of the investor sentiment index. Most sentiment is present in the

market between July 1997 and January 2002 with a peak of sentiment between May 2000 and March 2001. After 2004, there is a moderate rise in sentiment. After September 2008 there is negative sentiment in the market.

(16)

16

Figure 4: Graphical presentation of the change in the investor sentiment index. Sentiment is most

volatile during the period July 1997 and March 2001. Between March 2001 and January 2007 there is moderate volatility. After January 2007, volatility rises again.

Over the whole sample, sentiment is present in the market. Most of this sentiment can be attributed to the period 1997-2004, where sentiment fluctuate from -0.90 to 2.50 and changes of investor sentiment fluctuates from -3.53 to 4.37. These results show that investors are prone to sentiment in this period and their vision on the market fluctuates in a wide range. During the period 2004-2010, on average there is no sentiment in the market. Sentiment in this period fluctuates in a small range from -0.49 to 0.45. Changes in sentiment however, range from -2.88 to 1.89. These results imply that the vision of investors on the market fluctuates a lot in this period. In conclusion, over the whole sample period, sentiment is present in the market and changes in a wide range during this period.

4.1.1 Correlations

Before the times series regression is performed, it is necessary to check for the correlations between the independent variables. Correlations between independent variables can cause multicollinearity. This implies that a statistical relationship exists between two or more independent variables that significantly affect the estimation of the model. If a correlation is close to one, the results from the time series regression could contain a bias. Table 2 shows the correlation matrix of investor sentiment, the interaction term between the economic state and investor sentiment and the control variables used in this research.

(17)

17

Table 3: Correlation Matrix of the independent variables. Correlation exists between sentiment and the

interaction term between sentiment and the dummy variable indicating the economic state. This correlation might imply some imperfect multicollinearity in the regression.

The results in table 2 show that investor sentiment is not strongly correlated with any of the control variables. Investor sentiment is however, correlated with the interaction term between investor sentiment, lagged for one month, and the dummy variable for the economic state. These results imply that there might be some imperfect multicollinearity present in this model.

4.2 Time series regression

4.2.1 Times series regression on portfolio returns

A time series regression is performed on the multifactor model including the lagged sentiment index and the interaction term between the lagged sentiment index and the dummy variable indicating the economic state. This regression is performed on 25 portfolios formed on size and book to market ratio. Investors use the book to market ratio to draw conclusions whether the stock can be classified as ‘’growth stock’’ or as “value stock”. Growth stocks are seen in the market as stocks that are more exposed to sentiment while value stocks are more exposed to fundamentals. Thus stocks smaller in size and with a lower book to market ratio are seen as speculative stock while stocks bigger in size and higher in book to market ratio are seen as safe stocks, which are less prone to speculation. Four groups can be distinguished. The first group consists of stocks that are small in size and have a low book to market. This group of 9 portfolios is situated in the top left corner of the table. The second group has stocks that are small in size and have a high book to market ratio. This group of 6 portfolios is situated in the top right corner. The third group consists of stock that are big in size and have a low book to market ratio. This group of 6 portfolios is situated in the bottom left corner. The fourth group consists of stock that are big in size and have a high book to market ratio. This group of 4 portfolios is situated in the bottom right corner of the table. Group 1 is the group that is most prone to speculation and hardest to arbitrage and group 4 is the group that is classified as most safe and most easy to arbitrage.

(18)

18

The results of the regression are shown in table 4.

Table 4: Times-series regression on 25 portfolios formed on size and book to market ratio for the sample period 1997-2004. Dependant variables are excess returns on 25 portfolios formed on size and book to market ratio.

Independent variables are the investor sentiment index lagged for one period and the interaction term between the sentiment index and the dummy variable indicating the economic state. These two variables are controlled for the market risk premium (Rmkt-Rf), the size factor (SMB), the value factor (HML) and the momentum factor (MOM). Significance is shown with asterisks: 1%***, 5%**, 10%*

Form table 3, multiple conclusions can be drawn about the effect of investor sentiment on returns and the influence of the economic state. In the first group of portfolios, investor sentiment significantly affects the subsequent returns of four out of nine portfolios and the economic state has a significant impact on these portfolios. On the smallest portfolios with the highest volatility however, investor sentiment does not have a significant effect on subsequent returns. This result contradicts the findings of both Baker & Wurgler (2007) and Lemmon & Portniaguina (2006) that investor sentiment has the most influence on the subsequent returns of small stock and high volatility stock. In the second group of portfolios, investor sentiment significantly affects only two out of six subsequent returns, while the state of the economy does not have a significant impact of any of those portfolios. This group of stock contains mostly “value stock”, but still relatively small in size, which explains why this group of portfolios show less significant results than the portfolios in group 1. In the third group of portfolios, investor sentiment significantly affects four out of six portfolios, while the economic state only affects two out of six portfolios. Most of the stocks in these portfolios can be classified as ‘’growth” stocks. Baker & Wurgler (2007) show that investor sentiment has a significant predictive effect the subsequent returns of extreme growth. However, most of the firms in this group of portfolios are also big in size, suggesting that the effect of investor sentiment on this group of portfolios should be lower than the first group. This is however not true for all the portfolios in this group. In the last group of portfolios, only one out of four portfolios is significantly affected by investor sentiment and the economic state. In this group of portfolios most stocks can be classified as safe stock that are hard to arbitrage. According to Baker & Wurgler (2006), these types of stock should be negatively related with investor sentiment. These results however, contradict with these findings since most of the portfolios that are bigger in size and higher in book to market ratio show a positive relationship.

(19)

19

The results from table 3 give some important overall implication. First of all, the results show that investor sentiment does not significantly affects the subsequent returns of the portfolios that are small in size and the portfolios that are low in book to market ratios. These findings are in contradiction with the previous researchers Baker & Wurgler (2007), Lemmon & Portniaguina (2006) and Chung, Hung & Yeh (2012). Another important implication is that the stocks that are seen in the market as safe and easy to arbitrage do not have a negative relationship with investor sentiment, as suggested by Baker & Wurgler (2006). The returns of the stocks that are more prone to speculation and difficult to arbitrage however, all show the expected positive relationship in line with the theory (Baker & Wurgler, 2006). Chung, Hung & Yeh (2012) conclude that the economic state significantly influences the predictive power of investor sentiment across different states. Brown & Cliff (2005) conclude that investor sentiment always has a significant influence on returns, regardless of the economic state. The results in table 2, are more in line with the findings of Chung, Hung & Yeh (2012). The control variables based on traditional finance theories all show a significant effect. More details in table 3 of the appendix.

4.2.2 Times series regression on aggregate returns

Previous results show that there is a positive relationship between investor sentiment and subsequent stock market returns for almost all of the portfolios. It is however not agreed upon in literature if sentiment inhibits predictive power over aggregate returns. According to Baker & Wurgler (2006), the effect of investor sentiment on aggregate returns will be muted because stocks are not all moving in the same direction. A time-series regression over the whole sample is therefore performed on the average of the returns of the 25 portfolios to test whether this holds. Since the sample period includes two major crises, the time series regression is also performed on the two subsample periods. The results of these regressions are shown in table 5.

Table 5: Times series regression on aggregate US stock market returns. Times-series regression is performed on

aggregate returns of the whole sample (1997-2004) and two subsamples (1997-2004) and (2004-2010). Independent variables are the investor sentiment index lagged for one period and the interaction term between the sentiment index and the dummy variable indicating the economic state. These two variables are controlled for the market risk premium (Rmkt-Rf), the size factor (SMB), the value factor (HML) and the momentum factor (MOM). Significance is shown with asterisks: 1%***, 5%**, 10%*

Over the whole sample period (1997-2004), investor sentiment has a highly significant positive effect on subsequent returns over the whole market. Over the sample period, the portfolios sorted on size and book to market ratio contain on average more stock

(20)

20

that are prone to speculation and are harder to arbitrage than stocks that are seen as safe in the market that effect market returns through sentiment. The state of the economy has a significant role on this relationship, since the coefficient on the interaction term is highly significant. When the economy is in an expansion state (C=0), the total effect of investor sentiment on subsequent returns is 0.6. This leaves implications that in a high sentiment state, subsequent returns are not only explained by factors from traditional finance theories, such as the market risk premium, the size factor, the value factor and the momentum factor, but also explained by the level of sentiment from investors in the market. Behavioural assumptions are therefore favoured in asset pricing models during this economic state. When the economy is in a recession state (C=1), the effect of investor sentiment is -0.09. In the portfolios sorted on size and book to market ratio, on average there are more stock that are seen safe in the market that affect market returns through sentiment. Furthermore, in a recession state (C=1) the coefficient on sentiment is close to zero, which implies that that traditional factors, not sentiment, affect returns. The neoclassical assumption of the “Efficient market hypothesis” is favoured over the behavioural assumptions in this economic state.

Although there is still a small inverse effect (-0.09) of investor sentiment on market returns, these results are mostly in line with the conclusions drawn by Chung, Hung & Yeh (2012) that investor sentiment only explains returns in an expansion state and not in an recession state.

Since the sample period consists of two speculative bubbles that are differently affected by sentiment (figure 3 & 4), the sample is split into two subsamples to further explain the results of the whole sample. During the period of the dot.com bubble (1997-2004), investor sentiment significantly affects subsequent US market returns. Over the period 1997-2004 , the portfolios contain on average more stock that are prone to speculation and are harder to arbitrage than stocks that are seen as safe in the market that have an effect on market returns through sentiment. The state of the economy has no significant role on the relationship between investor sentiment and subsequent aggregate market returns, since the coefficient on the interaction term between investor sentiment and the economic state is not significant at a significance level of 10%. These findings are in favour of the conclusions drawn by Brown & Cliff (2005) and Charoenrook (2005), who conclude that investor sentiment significantly affects market returns regardless of the economic state.

During the period of the global financial crisis (2004-2010) there is no investor sentiment in the market. The estimated coefficient is not significant and close to zero. The state of the economy however has a strong significant effect on subsequent return. When the economy is in an expansion state (C=0), there is negative sentiment in the market, implying that mostly firms that are seen as safe in the market and are less prone to speculation affect market returns through sentiment. When the economy is in a recession state (C=1), market returns are not affected by investor sentiment and therefore not affected by any behavioural aspect. The neoclassical assumption of the “Efficient market hypothesis” is favoured over the behavioural assumptions in this

(21)

21

economic state. This is in line with the finding of Kojucharov et all (2008), who conclude in their paper that the boom in the subprime mortgage market may have occurred rationally given the information flow during the early years of the boom.

The implications of these findings are that the economic state has a significant role on the relationship between investor sentiment and US stock market returns in the period 1997-2010. When the economy is in an expansion state, behavioural models explain market returns better than neoclassical models. In an recession state however, neoclassical models explain market returns better than behavioural models. These implications could help policy makers to recognize the factors that determines returns in order to set policies to prevent financial bubbles. There are however limitations to these implications. First of all, the two major financial bubbles in this sample show conflicting results with respect to the relationship between investor sentiment and stock returns and the role of the economic state on this relationship. Secondly, in literature there is still debate about the correct measurement of investor sentiment. The measurement used in this research might therefore contain a bias.

(22)

22

5. Conclusion

This research examines the relationship between investor sentiment and US stock market returns and the influence of the economic state on this relationship for the period 1997-2004, thereby capturing the two most severe financial bubbles of the last decades that has led to severe global crises. A time-series regression on a multifactor model including investor sentiment and an interaction term between investor sentiment and the economic state is performed on 25 portfolios based on size and book to market ratio to examine this relationship.

Over the period 1997-2010, the economic state has a significant influence on the relationship between investor sentiment and aggregate US stock market returns. When the economy is in an expansion state, investor sentiment has a significant positive effect on subsequent aggregate US stock market returns. When the economy is in a recession state, there is a small negative relationship that is close to zero between investor sentiment and aggregate US stock market returns. These results are most similar to the results of Hung, Chung & Hey (2012).

The results imply that in an expansion state behavioural asset-pricing models are favoured over neoclassical models, while in a recession state neoclassical asset-pricing models are favoured over behavioural models. These implications could help policy makers and financial regulators to recognize the factors that determines market returns across different states of the economy to set regulatory policies accordingly.

There are however limitations to these implications. First of all, the two most important financial bubbles in this sample show conflicting results with respect to the relationship between investor sentiment and stock returns and the role of the economic state on this relationship. During the dot.com bubble (1997-2004) investor sentiment significantly affects US market returns, while in the period of the global financial crisis (2004-2010), investor sentiment does not significantly influences US market returns. Furthermore, the state of the economy does not significantly affect this relationship during the dot.com bubble, while it has a strong significant effect on the relationship in the period of the global financial crisis. These conflicting results show that the conclusions drawn from this research have to be interpreted with some caution. Another limitation is the measurement of investor sentiment. Although the investor sentiment index of Baker & Wurgler is interpret in the literature as the most unbiased measurement of investor sentiment, it could however still contain a bias. Finally, there might be some imperfect multicollinearity present in the model.

Whether these results would also hold in other global financial markets have to be concluded from further research on this topic.

(23)

23

References

Anderson, E., Ghysels, E., & Juergens J. (2005). Do Heterogeneous Beliefs Matter for Asset Pricing? Review of Financial Studies, 18, no. 3. 875-924.

Baker, M. & Wurgler, J. (2006). Investor Sentiment and the Cross-Section of Stock Returns. The Journal of Finance, LXI, no.4, 1645-1680.

Baker, M. & Wurgler, J. (2007). Investor Sentiment in the Stock Market. Journal of

Economic Perspectives, 21, no. 2, 129-151.

Banz, R. (1981). The relationship between return and market value of common stocks.

Journal of Financial Economics 9, 3-18.

Black, F. (1986). Noise. Journal of Finance 41, 529– 543.

Brown, G.W., & Cliff, M.T. (2004). Investor Sentiment and the Near-Term Stock Market.

Journal of Empirical Finance, 11(1), 1–27.

Brown, G. W., & Cliff, M. T. (2005). Investor Sentiment and Asset Valuation. Journal of

Business, vol. 78, no. 2.

Carhart, M. (1997). On persistence in mutual fund performance. The Journal of Finance 52, (57-82).

Chan, L.K.C., Karceski, J. & Lakonishok, J. (2000). New paradigm or same old hype in equity investing? Financial Analysts Journal 56, 23-36.

Charoenrook, A. (2005). Does Sentiment Matter? Working paper-3301937, University of New Orleans.

Chung, S.L., Hung, C.-H., & Yeh, C.-Y. (2012). When does investor sentiment predict stock returns? Journal of Empirical Finance, Vol. 19 , 217-240.

Cochrane, J.H. (2003). Stocks as money: Convenience yield and the tech-stock bubble. NBER Working paper 8987.

DeLong, Shleifer, J., Summers, A., Waldmann, R. (1990). Noise trader risk in financial markets. The Journal of Political Economy 98, 703–738.

Demyanyk, Y. & Van Hemert, O. (2008). Understanding the Subprime Mortgage Crisis. Working Paper, Federal Reserve Bank of St. Louis.

Fama, E. (1970). Efficient Capital Markets: a Review of Theory And Empirical Work. The

Journal of Finance, 25, 383-417.

Fama, E.F., French, K.R. (1996). Multifactor Explanations of Asset Pricing Anomalies. The

(24)

24

Fama, E.F., French, K.R. (1992). The Cross-Section of Expected Stock Returns. The Journal

of Finance 47, 427-465.

Fama, E. F., MacBeth,J. D. (1973). Risk, Return and Equilibrium: Empirical Tests. The

Journal of Political Economy 81, 607-636.

Jegadeesh, N., Titman, S. (1993). Returns to Buying Winners and Selling Losers: Implications for Stock Market Efficiency. The Journal of Finance 48, 65-91.

Kojucharov, N., Clyde, F., Martin, R.F., Lili, X. (2008). The Subprime Mortgage Crisis: Irrational Exuberance or Rational Error?

Lemmon, M., & Portniaguina, E. (2006). Consumer Confidence and Asset Prices: Some Empirical Evidence. The Review of Financial Studies, v19, n4.

Lintner, J. (1965). The valuation of risk assets and the selection of risky investments in stock portfolios and capital budgets. The review of economics and statistics 47, 13-37. Ljunqvist, A. & Wilhelm, W. (2003). IPO pricing in the dot-com bubble. Journal of Finance 58, 723-752.

Mian, A. & Sufi, A. (2008) The Consequences of Mortgage Credit Expansion: Evidence from the 2007 Mortgage Default Crisis. NBER Working Paper, 13936.

Ofek, E. & Richardson, M. (2001). Dotcom Mania: The Rise and Fall of Internet Stock Prices. NBER Working Paper No. 8630.

Sharpe, W.F. (1964), Capital Asset prices: A Theory of Market Equilibrium under Conditions of Risk. The Journal of Finance 19, 425-442.

Shiller, R.,J. (2008). The subprime solution: How Today’s Global Financial Crisis Happened, and What to Do about it. Princeton University Press, 2012.

Shefrin, H. (2010). Behavioralizing Finance. Leavey School of Business Research Paper, no. 10-01.

Shleifer, A., Vishny, R. (1997). The limits of Arbitrage. The Journal of Finance, 53. Qiu, L. & Welch, I. (2006). Investor Sentiment Measures. NBER Working Series 10794. Zhang, C. (2008). Defining, Modeling, and Measuring Investor Sentiment. University of California, Berkeley.

(25)

25

Appendix

Table 1: Sentiment proxies used to construct the index

The table represents the six correlated proxies used to construct an index for investors sentiment. Correlations are determined by taking the first order conditions and filtering out the common investor sentiment component. The index is corrected for any time lags between the different proxies. The index is standardized with a mean of zero and a unit variance, to make it more compatible for regressions.

Proxy: Defined as:

Closed-end fund Average difference between the net asset values of

closed-end stocks fund shares and their market prices

Trading volume

Share of equity issues

Dividend premium Log of average Market to book ratio of dividend payers and

non dividend payers

Initial public offerings Number of initial public offerings

Return on initial public offerings Average first day returns

Table 2: Variable definitions

Definitions of the variables used in equation (1). The definitions are based on the descriptions of other researchers.

Variable: Description:

Rpt-Rft Excess return of portfolio i (Sharpe, 1963)

αi Abnormal return (Sharpe, 1963)

Rmkt – Rf Market risk (Sharpe, 1963)

SMBt Size factor “Small Minus Big” (Fama & French, 1992)

HMLt Value factor based on market capitalization “High Minus

Low” (Fama & French, 1992)

MOMt Momentum factor (Jagadeesh and Titman, 1993)

SENTt-1

Ct

Onorthalized sentiment index, lagged for one period (Baker & Wurgler, 2007)

Dummy variable indicating a recession state (C=1) and an expansion state (C=0). (National Bureau of Economic Research)

C*SENTt-1 Interaction term between the investor sentiment index,

lagged for one period and the dummy variable indicating the economic state

(26)

26 Table 3: Times-series regression on 25 portfolios formed on size and book to market ratio for the sample period 1997-2004.

Dependant variables are excess returns on 25 portfolios formed on size and book to market ratio. Independent variables are the investor sentiment index lagged for one period and the interaction term between the sentiment index and the dummy variable indicating the economic state. These two variables are controlled for the market risk premium (Rmkt-Rf), the size factor (SMB), the value factor (HML) and the momentum factor (MOM). Significance is shown with asterisks: 1%***, 5%**, 10%*

Investor sentiment and stock market returns : the influence of the state of the economy