Investor Sentiment and Stock Returns: A Cultural and International View

Transcription

1 Investor Sentiment and Stock Returns: A Cultural and International View Tilburg University School of Economics and Management Faculty of Economics and Business Administration Department of Finance Master Thesis Finance Author: Job Andreas Eduardus Sinke ANR: Supervisor: Dr. J.C. Rodriguez Chairman: Prof. Dr. S.R.G. Ongena Date of graduation: 4 December, 2012

2 Investor Sentiment and Stock Returns: A Cultural and International View Job Sinke Master Thesis December 2012 Abstract: This paper examines if consumer confidence (a proxy for investor sentiment) affects expected returns internationally in 12 developed countries, looking at culture to be a possible determinant of cross-country differences. In line with earlier evidence I find that sentiment negatively forecasts aggregate stock market returns on average across countries. When investors are optimistic (pessimistic) future stock returns tend to be lower (higher). This relation holds for value stocks, growth stocks, small, mid and large cap stocks and for different forecasting horizons. Individual country results vary finding cross-country differences that can be explained by culture. This crosssectional perspective provides evidence that the impact of sentiment on stock returns is more pronounced in countries scoring high on uncertainty avoidance, masculinity, long-term orientation and for countries that score low on individualism and indulgence, the effect of power distance is ambiguous. Keywords: consumer confidence; investor sentiment; style/size differences; cultural dimensions; predictive regressions -2-

3 Contents 1. Introduction Theory and Hypothesis Development The sentiment-return relation Cross sectional differences between different styles or sizes of stocks Culture and the sentiment-return relation Data and Descriptive Statistics General data considerations Descriptive statistics Preliminary tests Predictive Regressions of Stock Returns on Consumer Confidence Methodology Results for panel regressions Results for individual countries Cross-sectional analysis Relation of results to earlier studies Discussion Conclusion References Appendices Appendix I. Different proxies for investor sentiment Appendix II. Details on consumer confidence surveys Appendix III. MSCI indices construction methodology Appendix IV. Discussion of Hofstede s his dimensions Appendix V. Tables

4 Tables Table 1. Countries used in the study Table 3. Descriptive statistics Table 4. Descriptive statistics Table 5. Correlations between cultural dimensions Table 6. Panel unit-root tests Table 7. Granger-causality tests Table 8. Consumer confidence correlations Table 9. Sentiment coefficient in long-horizon regressions Table 10. Sentiment significance in long-horizon regressions Table 11. Economic magnitude of the sentiment effect Table 12. R-squares of sentiment regressions Table 13. Return predictability for individual countries across horizons Table 14. Crisis joint significance Table 15. Sentiment coefficient in long-horizon regressions: behavioral panels Table 16. Sentiment significance in long-horizon regressions: behavioral panels Table 17. Cultural joint significance

5 1. Introduction The global stock market crash of 2007 and 2008 was an exceptional event. The MSCI World index of developed markets fell approximately 50 percent in dollar terms. Emerging markets fell even further. Individual country indices displayed falling returns, but with varying percentages. Such an event is often hard to explain by classical theories even ex post. The classical theories try to rationalize investor behavior, but in times of crisis this turns out to be difficult if not impossible. Behavioral finance rejects this neoclassical economic man or rational investor and recognizes investors to be normal. In this field the unbounded rationality is replaced by the observation of empirically established rationality defects. Or in other words behavioral finance is no longer based on the rationality of investors but accepts people, investors to behave different, normal, or irrational at times. Behavioral finance assumes this bounded rationality to be valid for most countries (Levinson and Peng, 2007). However, as seen in the MSCI World Individual country indices countries vary, cultures vary and cultural finance supports the diverging relevance of certain behavioral patterns between countries. Statman (2008) finds rationality to defect but in various magnitudes across countries. A stock market crash can therefore be explained by the behavioral approach. This field of research has recently attracted much interest, given the realization that investors might not behave rational after all. While the behavioral approach becomes more common, little attention has been paid to culture affecting finance. Guizo, Sapienza and Zingales (2006) document that culture may significantly affect economic outcome. One might wonder if finance is affected as well. More recently a field called cultural finance has established itself, which was named by Breuer and Quinten (2009). Existing economical and financial literature were reluctant to include cultural influences on financial questions for a long time. Values have been taken as given and culture is treated as a black box (Williamson, 2000). This reluctance to include culture comes from its definition and from methodological challenges when trying to operationalize culture. The absence of sufficient theoretical foundation can also explain the scarcity of research in this area. Recently culture is starting to gain interest of academics leading to papers with culture as the main subject (Guiso, Haliassos and Japelli, 2000; Chui and Kwok, 2008; Breuer and Quinten, 2009; Breuer and Salzmann, 2012). In different areas, culture is found to be able to explain country-specific differences. For example national culture turns out to be an important and effective factor for life insurance consumption (Chui and Kwok, 2008). Breuer and Salzmann (2012) link national culture to household finance using Schwartz s cultural dimensions. Breuer and Salzmann (2008) also find culture to influence corporate governance. Breuer and Quinten (2009) introduce the term cultural finance and make way for a new field of research. The fields of behavioral and cultural finance show -5-

6 considerable similarities, they are both based on the notion of an irrational investor. The relation of sentiment and stock market returns has been greatly explored in existing literature; therefore the focus of this research lies in the cultural approach. This paper explores the sentiment-return relationship in a way that can be best classified in the cultural finance field of research and will attempt to identify culture as a cross-country determinant of variance in the sentiment-return relation. Earlier evidence of the sentiment-return relation is primarily focused on the U.S. stock market and is based on two assumptions: that investors are subject to sentiment (Delong, Shleifer, Summer and Waldman, 1990) and that there are limits to arbitrage (Shleifer and Vishny, 1997). This paper looks at multiple countries as well as the U.S. stock market, analyzing the international aggregate stock market. The stocks have been divided in size and style, as some stocks are harder to value and arbitrage then others, leading to different results (e.g. Baker and Wurgler, 2006; Lemmon and Portniaguina, 2006). Aggregate country level stock markets are obviously hard to value and hard to arbitrage. Macro data is known to be noisy and it is difficult to hedge away all idiosyncratic risk at country levels. Sentiment shocks will therefore affect stock markets on aggregate and not just different subgroups of stocks. The international approach taken by this paper allows for exploration of possible determinants of cross-country differences. In this paper I will look at cultural effects and test if the sentiment-return relation is more pronounced in countries pooled together based on cultural differences. The international dataset provides a natural out-of-sample test for earlier findings. Griffin, Ji and Martin (2004) or Ang et al. (2008) state the importance of testing market phenomena such as the sentiment effect out of sample. Pooling data across countries increases the power of tests which yield more reliable estimates (Ang and Bekaert, 2007). The question is no longer if there is a sentiment-return relation but how to explain the differences among countries. What drives the sentiment-return relationship? Is a more relevant question these days. I will examine the relationship between sentiment and returns in relation to Hofstede s cultural dimensions of power distance, individualism, masculinity, uncertainty avoidance, long-term orientation and indulgence versus restraint. As the main contribution of this paper I find that national culture has significant impact on the sentiment-return relationship after controlling for macro economical factors. Further I find that there is a significant impact of investor sentiment on aggregate stock returns across countries using predictive panel-fixed regressions for a panel of 12 countries from all over the world. I find differences in the magnitude of impact when looking at different style and size stock indices and culture is a cross-country determinant of the magnitude of impact. The impact of sentiment on stock returns is more pronounced in countries scoring high on uncertainty avoidance, masculinity, long-term orientation and countries that score low on -6-

7 individualism and indulgence, the effect of power distance is ambiguous. The found effects remain significant after controlling for other standard risk factors, or macro economical risk factors. The remainder of the paper is structured as follows. Section 2 introduces the theoretical background, focusing both on the sentiment-return relationship and the presumed effects national culture has on the relationship. Section 3 presents the dataset as well as descriptive statistics. Section 4 reports methodology, the results of the research and will discuss. Finally section 5 will conclude. 2. Theory and Hypothesis Development In the first part the sentiment-return relation will be explained, the second part will explain why there could be differences between types of stocks and which stocks should be the most affected. Third I will introduce culture as a factor determining the magnitude of the sentiment-return relation The sentiment-return relation The general finding of a sentiment-return relationship is not consistent with standard finance theory. The classic or standard finance theories are based on the assumption of a rational investor. The behavioral approach recognizes that investors are not rational but normal. This fact, leads to systematic biases in their beliefs inducing them to trade on non-fundamental information called sentiment. The classic theories state that stock prices reflect the discounted value of expected cashflows and that irrationalities among market participants are erased by arbitrageurs, with sentiment not playing any role in the framework. Rationality however finds it difficult to explain the stock market collapse in 1929 (Delong and Shleifer, 1991; White 1990). Maybe it was irrational exuberance (Shiller, 2000) driving prices above fundamental values. More recently the internet bubble or dot.com bubble showed that the rapid rise and fall of technology stocks was due to excessively bullish sentiments returning to more normal values in Several papers find it hard to reconcile these examples with rational pricing (Lamont and Thaler, 2001; Ofek and Richardson, 2001). Shiller (1987) surveys both individual and institutional investors inquiring about their behavior surrounding the 1987 crash. The survey shows that most investors interpret the crash as the outcome of other investors psychology rather than changes in fundamentals such as earnings or interest rates. Siegel (1992) finds that changes in neither interest rates nor changes in future earnings account for the dramatic valuation changes around the October crash Moreover he concludes that shifts in investor sentiment are correlated with market returns around the crash. Sentiment thus plays a role in explaining variance in stock markets. -7-

8 In the behavioral approach waves of irrational sentiment, times of overly optimistic or pessimistic expectations, can persist and affect asset prices for significant periods of time. Brown and Cliff (2004) describe sentiment to represent the expectations of market participants relative to a norm: an optimistic (pessimistic) investor expects returns to be above (below) average, regardless of what average may be. Baker and Wurgler (2007) broadly define investor sentiment, as a belief about future cash flows and investment risks that is not justified by facts in hand. Clearly sentiment is not a factor that can be described as rational. Several studies offer models establishing the relationship between investors sentiment and asset prices. Delong, Shleifer, Summers and Waldman (1990) show theoretically that correlated sentiment of noise traders affects equilibrium stock returns. Baker and Wurgler (2006) find that sentiment-based mispricing is founded both on an uninformed demand shock (noise trading) and on limits to arbitrage. Brown and Cliff (2005) argue that sentiment is most like, a very persistent effect. Trader noise or uninformed demand shocks may thus be correlated over time and give rise to strong persistent mispricing. Regarding trader noise Barber, Odean and Zhu (2008a) find that trading by individuals is highly correlated which is consistent with systematic noise trading that does not wash out in the aggregate. Limits to arbitrage deter investors from eliminating mispricing (Black, 1986; Shleifer and Vishny, 1997) since beforehand it is unclear how long buying or selling pressure from overly optimistic or pessimistic noise traders will persist. Eventually every mispricing is corrected so high levels of optimism are on average followed by low returns and low levels of optimism or even pessimism are followed by high returns on average. Arbitrage forces can eliminate profitable shortterm trading strategies but not for longer run mispricing (Brown and Cliff, 2005). Sentiment may drive asset prices away from fundamental values for extended periods of time but this is difficult to detect over short horizons (Summers, 1986). Betting against a sentimental investor is found to be costly and risky according to Shleifer and Vishny (1997). This leads to rational investors or arbitrageurs to be less aggressive in forcing prices to fundamentals as the standard theory would predict, resulting in limits to arbitrage. These limits come from short time horizons or from costs and risks of trading and short selling according to Delong, Shleifer, Summers and Waldman (1990). Market imperfections must lead to observed prices deviating from fundamental value (Grossman and Stiglitz, 1980). Thus it seems that sentiment can affect prices and that this effect is persistent. Investors are subject to sentiment (Delong, Shleifer, Summers and Waldman 1990), which seems intuitive given recent developments in the behavioral finance field. Many models find two types of investors; rational arbitrageurs who are free of sentiment and irrational investors prone to exogenous sentiment. These types are competing in the market and setting both prices and expected returns which lead to an aggregate affect of sentiment on stock market returns. -8-

9 Different models are explaining this bias that is called sentiment, using individual biases in investor psychology such as overconfidence, representativeness, conservatism and over- or under reactions. These psychological factors are said to be the source of the noise for traders. Daniel, Hirshleifer and Subrahmanyam (1998) find investors are overconfident about their private signals and they incorrectly attribute successful outcomes to their own abilities and blame bad outcomes to chance rather than their own mistakes, basing their research on self-attribution and overconfidence. Barberis, Shleifer and Vishny (1997) are using a model which has earnings as a random walk distribution but investors belief earnings to switch between a mean-reverting regime and growthregime, a belief that turns out to be wrong. Investors see a pattern where none exists which is called representativeness. The investors are slow to update their beliefs of the regime in the light of new evidence, a psychological effect known as conservatism. Hong and Stein (1999) find under reaction to lead to trading by momentum factors. Overreaction results when the momentum investors have gone too far. Shefrin (2008) relies on differences in opinion across investors. Differences in opinion, even when investors have the same information, can be large (Miller, 1997). Persistence in sentiment measures is consistent with conservatism and biased self-attribution (Brown and Cliff 2004). When aggregated these models of individual psychological factors make predictions about patterns in market wide investor sentiment, stock prices and volume. Earlier evidence (Brown and Cliff, 2005; Baker and Wurgler, 2006) find a negative sentimentreturn relation on the aggregate U.S. stock market level. Schmeling (2009) finds evidence for this relation on an international level. Baker, Wurgler and Yuan (2009) find that sentiment affects stock markets both globally and locally. This evidence leads to the first hypothesis: H1: International investor sentiment predicts future aggregate returns. Optimism (pessimism) leads to a negative (positive) significant relationship between returns and sentiment, these relations are robust after controlling for fundamental factors Cross sectional differences between different styles or sizes of stocks There is also evidence (Baker and Wurgler, 2006; Lemmon and Portniaguina, 2006) that sentiment affects the cross-section of returns differently for different investment styles such as value or growth stocks and small or large stocks. This can be explained by the characteristics of the different stocks, small or growth stocks are harder to arbitrage and harder to value than large value stocks with a long and stable earnings history. Stocks of low capitalization, younger, unprofitable, high-volatility, nondividend paying, growth companies or stocks in financial distress are likely to be disproportionally sensitive to waves of investor sentiment according to Baker and Wurgler (2007). Such stocks tend to be more costly to buy and sell short (D Avolio, 2002). Wurgler and Zhuravskaya (2002) also state that such stocks have a high degree of idiosyncratic variation in their returns making betting on them -9-

10 riskier. This higher volatility may lead to second-guessing by investors who provide funds for the arbitrageurs leading to withdrawals from those arbitrageurs when mispricing is the greatest (Shleifer and Visny, 1997). Kurov (2008) finds investor sentiment affects trader s behavior and market liquidity. Kumar and Lee (2006) show that retail investors, often characterized as noise traders, tend to overweight value stocks relative to growth stocks and that shifts in buy-sell imbalance of these retail investors are positively correlated with the returns of value stocks. This is a prime example of noise trader risk. Lee, Shleifer and Thaler (1991) identify noise traders with individual investors and show that small stocks are disproportionately held by individuals as opposed to institutions. Nagel (2005) finds a strong positive correlation between ownership by institutions and size. Chan and Chen (1991) identify that small firms are also associated with higher levels of financial distress risk. The key point is that the same securities that are difficult to value are difficult to arbitrage. This extra difficulty in arbitrage with these stocks should lead to higher sensitivity to investor sentiment, as the uncertainty of value means that the effects of overconfidence, self-attribution (Daniel, Hirschleifer and Subrahmanyam, 1998), representativeness and conservatism (Barberis, Shleifer and Vishny, 1998) are more pronounced. This evidence leads to the second hypothesis: H2: The effect of sentiment is more pronounced for stocks that are harder to arbitrage and harder to value. Small and growth stocks are influenced more heavily by investor sentiment compared to large and value stocks Culture and the sentiment-return relation Not all people across the globe behave in the same way. if, as argued by practitioners of behavioral finance, individuals have psychological biases that matter for finance, it would be surprising that individuals view of the world as determined by their culture does not matter for how they view and act in financial markets Stultz and Williamson (2003 p.347). A collective set of common experiences that people will share, will influence cognitive and emotional approaches to investing according to Statman (2008). Culture should thus lead to cross country differences among people, investors and behavior. Culture is however very broad and channels through which it can enter economic models are ubiquitous and vague, so that is it difficult to design testable, irrefutable hypothesis. Culture is as described by Traindis et al. (1986) a fuzzy, difficult-to-define construct. Guiso, Sapienza and Zingales (2006) define culture as those customary beliefs and values that ethnic, religious and social groups transmit fairly unchanged from generation to generation. Hofstede and Bond (1988) provide a more comprehensive definition of culture in the programming of the mind that distinguishes the members of one category of people from those of another. Culture is often composed in certain values, which shape behavior as well as one s perception of the world. Adler (1997) finds that -10-

11 culture influences our values, which affects our attitudes and then our behavior. This type of hierarchy has been empirically substantiated by Homer and Kahle (1988). Earlier literature has proven that culture can affect economics and finance. de Jong and Semenov (2002) focus on the stock market development of OECD (Organization of Economic Cooperation and Development) countries, finding that stock markets tend to be more developed in countries with lower levels of uncertainty avoidance and higher levels of masculinity. Chui, Lloyd and Kwok (2002) find that managers choose lower corporate leverage in countries with higher levels of conservatism and mastery using the cultural measures of Schwartz (1994). Guiso, Sapienza and Zingales (2003) have found religion to be associated with economic attitudes. Stulz and Williamson (2003) show that culture can affect predominant values, institutions and resource allocation in a country. Culture may lead to certain attitudes that are more conducive to certain outcomes (Guiso, Sapienza and Zingales, 2006). Culture can exert influence by affecting institutions in a country (Stulz and Williamson, 2003). Concluding that culture matters, is persistent (Guiso, Sapienza and Zingales, 2006) but also variant (Statman, 2008). Dimensions of culture can impact financial outcomes, through prior beliefs, values and preferences. The focus lies on dimensions of culture that are inherited by an individual rather than voluntarily accumulated. Becker (1996) finds that individuals have less control over their culture than over other social capital. Behavioral finance argues that imperfections in financial markets are due to a combination of psychological biases such as overconfidence, investor overreaction and various other human errors in reasoning and information processing. The behavior of investors is based on cognitive and emotional aspects of decision making. Psychology factors are however often quite vague and narrow, taking national culture as means of determining those values creates a more comprehensive approach, as it describes the mindset of people exhaustively. All of the psychological factors referred to by previous studies can be found as basic values in the cultural dimensions concept. Values with similar characteristics can be pooled in one cultural dimension. National culture thus presents a more systematic approach. Using Hofstede (1980) his traditional four dimensions in addition with two later added dimensions (2011) gives an indication if and how culture affects the sentiment-return relationship. For each of the dimensions theory can be related to the psychology factors in existing literature. Mainly overconfidence, self-attribution, representativeness, conservatism, herd-like behavior and persistence are part of the explanation why culture should affect the relationship. As argued above culture can affect economics and finance, the way in which I expect culture to affect the sentiment-return relationship is discussed in more detail below, deriving hypothesis from literature for all used dimensions. Individualism (IND): according to Hofstede (2001) measures the extent to which individuals are integrated into groups. In individualistic countries ties between individuals are lose, people are -11-

12 expected to care not much about persons beyond their immediate family. Hirshleifer and Thakor (1992) find that in high individualistic countries the first priority of agents is to take care of their own interests. Collectivistic societies are integrated into strong groups. Chui, Titman and Wei (2008) relate this dimension to psychological factors such as over optimism, overconfidence and selfattribution. Park and Lemaire (2011) find countries with a high individual score to search for more security in the form of insurance, indicating a link between UAI and IND. Hofstede (2001) states that individualism affects the degree to which people display an independent behavior rather than a dependent behavior. Moreover he argues that children in collectivistic cultures build their identity for their social system. Countries with higher levels of collectivism indicate a tendency to herd-like behavior. While Chui, Titman and Wei (2008) clearly link collectivistic behavior to the tendency of investors to herd, to explain the sentiment-return relation individuals also need to overreact. Jackson (2004) provides evidence for Australia that people overreact and create a negative relation between returns and sentiment in this way. Therefore I expect countries with high levels of individualism to be less affected by sentiment. Power Distance (PDI): is found by Hofstede (2001) to resemble the extent to which less powerful members of organizations and institutions accept and expect power is distributed unequally. In cultures with high power distance, people take inequality as granted, tolerate the concentration of power and are more reluctant to give up independence (de Jong and Semenov 2002). In cultures with small power distance values like trust, equality and cooperation are more important. Chui and Kwok (2008) find in high power distance countries that individuals expect superiors to be mindful of their welfare and take actions to reduce their risk in return for their surrender of power and acceptation of authority. However this also happens in low power distance countries where political leaders take actions to reduce risk as well. Institutions thus play a larger role in high power distance countries and those countries are said to be more collectivistic, theoretically showing correlation between PDI and IND. Chui, Titman and Wei (2008) also use proxies for Stock market integrity, Power distance is related to those proxies. Corruption for example is more likely to occur in high power distance countries. Schmeling (2009) also uses anti-director rights as a proxy for market integrity. This can be linked to PDI in stating that high power countries have institutions protecting welfare as individuals expect this from their superiors and take actions to reduce risk, this should also include shareholder protection. Indices for corruption perception and anti-director rights have also been used by La Porte et al. (1998). Although this paper does not test these variables directly PDI includes some of these characteristics. Markets with higher institutional quality should have better developed flow of information and are consequently more efficient (Chui, Titman and Wei, 2008). House et al. (2004) find a significant difference in the way information is shared; in high power distance countries information is localized and in low power distance countries -12-

13 information is shared. This shows differences between human reasoning and information processing. Hofstede (2011) finds in high power distance countries subordinates expect to be told what to do, and in low power distance countries they expect to be consulted, showing correlation with individualism and the herd-like behavior. As a result of these findings I expect high power distance countries to be more affected although some evidence is mixed. ` Uncertainty avoidance (UAI): this dimension measures the likely degree of overreaction across countries. Hofstede (2001) documents that people in more uncertainty avoiding countries act and react more emotional compared to countries with low levels of uncertainty avoidance. People in those countries act more completive and thoughtful. People in high uncertainty avoidance countries are not comfortable with ambiguity and uncertainty and try to avoid such situations. In countries with high levels of uncertainty avoidance people prefer certainty, security, and predictability and are reluctant to accept risks (Riddle 1992; Offerman and Hellman 1997). Investors might need to deal with the possibility that they do not possess some information which might affect future price movement, particular in inefficient markets according to Lucey and Zhang (2009). This information asymmetry triggers uncertainty, and this uncertainty may influence attitudes and propensities of investors and people. House et al. (2004) states that people in countries with high levels of uncertainty avoidance show a stronger resistance to change. In this way conservatism can be related to this dimension. Hofstede (2001) finds uncertainty avoidance to be correlated with collectivism since the uncertainty avoidance index captures cross-country differences in the propensity of people to follow the same set of rules and thus behave in the same manner. Higher levels of uncertainty avoidance indicate both tendencies towards overreaction and to herd-like behavior, leading to the hypothesis that countries that score high on this index are influenced more heavily by investor sentiment. Masculinity (MAS): evaluates whether gender differences impact roles in activities. Hofstede (2001) describes masculinity as the opposite of femininity referring to the distribution of emotional roles between genders in countries. Masculine countries emphasize factors such as being very assertive and competitive and having a willingness to seek competitive outcomes (de Jong and Semenov, 2002). Portfolio managers in countries with high masculinity are likely to overreact and show overconfidence when they invest in shares, while they behave conservatively in countries with low masculinity (Lucey and Zhang, 2010). Barber and Odean (2001) clearly show the values overconfidence and self-attribution to be more pronounced in men than in women. In masculine societies, performing, achieving and making money are given paramount importance (Gleason et al., 2000). In feminine societies, helping others and the environment, having warm relationships and the quality of life are key values. Women seem to take less risk as they strive for helping others and quality of life or security, so the effect of overreaction should be lower with feminine countries. Chui -13-

14 and Kwok (2008) find feminine societies to purchase more insurance, as they care more about family needs, showing a link with uncertainty avoidance. Based on this information the effect of investor sentiment is more pronounced in countries with high masculinity. Long-term orientation (LTOWVS): Research done by Hofstede and Bond (1988) led to this added fifth dimension. Further research done by Minkov (2010) allowed extending the number of countries for this dimension using World Value Survey data. Hofstede (2001) finds the long-term orientation societies to find the values orientated to the future important, in particular saving, perseverance, persistence and adapting to changing circumstances. Short-term orientated societies relate to the past and present, such as traditions, preservation of face and fulfilling social obligations. Long-term orientation seems to be based more on synthetic thinking whereas the shortrun orientation is more analytical in thinking. Park and Lemaire (2011) describe this dimension as a variable that scores countries based on adherence to Confucian principles such as perseverance and thrift, respect of tradition and family values, and honor of parents and ancestors. Furthermore Hofstede (2011) finds that in short term orientated countries students attribute success to themselves and failure to luck, in the long term orientated countries success is attributed to effort and failure to a lack of effort, one can clearly see the self-attribution bias in these differences. Sentiment is a very persistent factor and the effects of sentiment can grow over time thus long term orientation should lead to more pronounced sentiment effects. Park and Lemaire (2011) find that Long-Term has a strong positive impact on life-insurance demand. The expected effect could be ambiguous or less clear because the self-attribution bias is based on short-term orientation and persistence on the long-term orientation. However the difference in thinking, synthetic to analytical, can create errors in reasoning and information processing tilting the expected effect that sentiment is more pronounced in long-term orientated countries. Indulgence versus restraint (IVR): this dimension has been added more recently following Minkov (2010) his research. Indulgence stands for a society that allows relatively free gratification of basic and natural human drives related to enjoying life and having fun. Restraint stands for a society that suppresses gratification of needs and regulated it by means of strict social norms (Hofstede 2011). According to Hofstede (2011) in indulgent societies positive emotions are more likely to be remembered opposed to negative. This can be explained as a form of self-attribution or at least uneven reaction to different signs, negative and positive. This implicates that under- and overreactions play a role for this dimension. This index is closely correlated with the traditional/secular-rational value of Inglehart (2006), which states importance of traditional values such as respect for authority and institutions. Childs often learn obedience and faith opposed to independence and determination in traditional countries. Power distance, institutions and authority are thus correlated with this index. And when looking at traditional values, conservatism, or -14-

15 reluctance to change can well be added to values characterized by traditional countries. In countries with high restraint levels, there are stricter norms, and there is more police found by Hofstede (2011). One can argue if these rules, institutions and increased focus on authority lead to better stock market integrity. I use this index as a rough proxy for market integrity and along with the other information the effect might not be as clear as with some of the other dimensions but I expect countries who are indulgent, or score high on this index to be more heavily influenced by sentiment as this stands for worse market integrity. H3: The impact of culture on the sentiment-returns relationship is significant and culture is a crosscountry determinant of the sentiment-return relation. Effects are more pronounced for countries that score high on UAI, MAS, ITOWVS and IVR, for countries that score low on IND and for PDI the effects is ambiguous. 3. Data and Descriptive Statistics 3.1. General data considerations This study takes a look at the effect of noise trader demand shocks on stock markets. The main focus of this paper is the cultural approach since the sentiment-return relationship is rather well established in earlier literature. The international aspect of the paper allows for country differences to become visible. The dataset is therefore build upon cultural differences between individual countries to find if culture is a cross-county determinant of the sentiment-return relation. The first task is to measure the effect of noise trader demand shocks on stock market returns. There is no consensus on the kind of proxy to use when measuring individual sentiment for a single country, so finding a reliable consistent proxy for multiple countries is certainly a difficult task. Earlier literature, although extensive, is not consistent in the use of proxies for sentiment. This paper will use consumer confidence as a proxy of investor sentiment 1. Consumer confidence is available for several industrialized developed countries and for reasonable periods of time. It is measured slightly different across countries but it seems to be the only consistent way to obtain a sentiment proxy that is largely comparable across countries and one that is not calculated from trading data itself. The surveys in the different countries are chosen to resemble each other and contain the same core questions looking at the similar horizons 2. In taking a cultural approach consumer confidence thus provides a natural proxy for sentiment, as it is international orientated and consistent across countries. 1 A short overview of alternative sentiment measures and evidence on consumer confidence as a proxy for investor sentiment can be found in appendix 1. 2 Appendix 2 gives details of the consumer surveys; it discusses similarities as well as differences between countries. -15-

16 Consumer confidence data is collected for all sample countries. For all 7 European countries the data comes from the Directorate Generale for Economic and Financial Affiars (DG ECFIN) which conducts research for the European Union. This data is also by used Jansen and Nahuis (2003) and Schmeling (2009). Several high quality surveys exist for the U.S. this paper employs the Michigan Survey as was done before by Lemmon and Portniaguinia (2006). For the other countries data comes from datastream and the surveys are selected based on the comparability. The index of Japan is only available on quarterly frequency, converting this data to monthly frequency is done using last available values for months without data as in Baker and Wurgler (2006). I use data on stock returns and consumer confidence for 12 developed countries. Data availability and significant cultural differences dictated the sample of countries. For example Sweden is excluded because it has almost the same values as Denmark for the cultural dimensions, so cross country determination of variations would become less clear. Austria is excluded due to a limited time horizon for the sentiment data. I include Australia, Canada, New Zealand, Japan, the U.S. and 7 European countries. These markets cover the lion s share of international stock market capitalization; cover most liquid markets in the world being the U.S., Europe and Japan- thus providing a representative sample. For all countries stock data is collected based on MSCI country indices including size and style differences. The indices are chosen instead of the largest stock exchange in the individual countries for comparability reasons. MSCI is formed following the same methodology in every country 3. The MSCI indices are total returns including dividend and returns are in local currency to avoid currency and exchange rate effects. In the remainder of the paper these indices will be referred as MSCI (Aggregate market), SG (small-growth), SV (small-value), MG (mid-growth), MV (mid-value), LG (large-growth) and LV (large-value). The stock market data will be collected with a monthly frequency. With time horizon starting June 1994 ending in may Returns are however influenced by macro economics risk factors. Culter, Poterba and Summers (1991) establish that macro-economic variables explain approximately one-third of the variance in stock returns. To create an explanatory variable that is unrelated to fundamental risk factors some macro economics risk factors are added as control variables. I add four main macro economic factors, a selection based on earlier literature and data availability. Using these control factors the macro risk is netted out from the sentiment proxy. I acknowledge that this research might miss some important macro risk factors, but with the included set of control variables I feel comfortable that reasonable effort is made to clean the sentiment variable of this macro risk. This paper focuses on culture as a possible determinant of country specific differences, making controlling for macro risk factors less important. As missing some possible factors of macro risk will not lead to 3 A brief explanation of the used methodology of how the MSCI indices are constructed can be found in appendix

17 dramatic changes in cross country differences when pooling countries together based on Hofstede his cultural dimensions. The controls employed are the detrended short rate, the term spread, the annual change in industrial production and the annual CPI inflation. The short term rate is the 3-month interbank lending rate for individual countries which is stochastically detrended. Detrended short rate is also used by Campbell (1991); Hodrick (1992) and Brown and Cliff (2005). Term spread is calculated by the taking the difference between the yield of 10-year government bonds and the yield of 3-month treasury bills as done by Fama and French (1989). Sources for this control variable are national banks, European central bank and the Federal Reserve System. Inflation is measured by the change in CPI (consumer price index) and the source is international financial statistics (IFS). CPI is included based on Baker and Wurgler (2006); Lemmon and Portniguina (2006). CPI is measured as monthly year to year changes and is displayed in a percentage as done by Schmeling (2009). Industrial production is the monthly year to year change and is measured in a percentage. The source is IFS which is consistent across countries. Industrial production is included based on evidence by Schmeling (2009) and Zouaoui, Nouyrigat and Beer (2011). Concluding that netting out macro risk effects is broadly done in the literature and necessary for an explanatory sentiment variable that is unrelated to fundamental or macro risk factors. Existing literature also highlights dividend yield to be an important macro risk factor but when using MSCI indices it is hard and time-consuming to construct dividend yields. I chose to use MSCI for good comparability also across different style and size stocks but the downside is that dividend yield cannot be added as a control factor. Baker, Wurgler and Yuan (2009) find that dividend in some markets is uncommon and dividends do not appear to be viewed by local investors as indicating stability. For the U.S. dividends show stability and therefore influence optimism and sentiment, however in an international setting this does not have to be the case. Thus dividend yield could be less important in an international setting. The results for the sentimentreturn relation could still include some macro risk. However, when including dividend yield as a controlling factor this could still be the case. The data to explore cross country variance comes from dimensions of culture found by Hofstede. These are measured consistent across countries, the levels are used to pool countries in different panels. In this way culture serves as a possible determinant of cross-country variance. Chui, Titman and Wei (2008) use the dimensions individualism and uncertainty avoidance to motivate their herding and collectivism. Schmeling (2009) uses individualism and uncertainty avoidance as a possible determination of cross country variance. Zouaoui, Nouyrigat and Beer (2011) show that the impact of sentiment in countries culturally prone to herd-like behavior and overreaction is more pronounced based on Hofstede his dimensions. Extensive literatures in other fields also make use of Hofstede his dimensions to explain cross-country differences based on culture. -17-

18 Hofstede conducted a comprehensive study of how values in the workplace are influenced by culture based on employee value scores collected by IBM between 1967 and 1973 covering 70 countries and including about 88,000 respondents. First using only the largest countries which had the most respondents and later extending the analysis. In 2010 the 76 countries are listed from research partly based on replications and extensions to the IBM study. Hofstede s (1980a) work has been criticized for: reducing culture to overly simplistic dimension conceptualization, failing to capture culture over time and ignoring within-country cultural heterogeneity (Sivakumar and Nakata, 2001). Despite this criticism, researchers have favored using his framework due to its clarity and parsimony 4. This study looks at different countries and will try to find if culture affects the sentimentreturn relationship. Since the field of research called cultural finance is rather new using Hofstede s his dimensions as determinants of culture seem a good starting point. Further research can be done using different values and perhaps using a more complex measure of culture. Breuer and Quinten (2009) find that there is extensive literature using Hofstede s his dimensions both in economics and finance. Using Hofstede s dimensions thus seem a natural metric to use for the cross-country differences in culture. Table 1 provides the used countries, the number of participants for the consumer confidence index and the codes used for the countries. Table 2 provides an overview and a description of the variables used in the study including: variable names, used abbreviations, measures and sources Descriptive statistics Tables 3 and 4 provide descriptive statistics for the stock market data and the sentiment proxy. The tables provide country specific mean returns and standard deviations for stock markets and for sentiment the first-order autocorrelations are also provided. The statistics are shown for two timeseries separately; table 3 runs from June 1994 until December 2006 and table 4 runs from January 2007 until May The periods are divided because of the economic crisis and this crisis can distort any findings and results, a first indication of any influence of the crisis can be seen by comparing the tables. The crisis became apparent in the summer of 2007, literature however concludes that macro variables have some predictive power therefore the period is split at the beginning of 2007 since macro variables may reflect the crisis before it became visible in falling stock markets. Table 3 represents the normal period. Value stocks have higher mean returns than growth stocks and lower standard deviations for most countries, a fact documented by literature by the 4 Appendix 4 contains a discussion of the advantages and the disadvantages of using Hofstede s (2011) cultural dimensions. -18-

19 value premium (Fama and French, 1998). The mean returns for small stocks are generally higher than for large stocks, evidence for the size premium (Fama and French, 1993). Overall small stocks have higher returns and standard deviations opposed to large stocks and value stocks have higher returns and lower standard deviations compared to growth stocks. Table 4 shows the crisis period. When comparing the table with the normal period some interesting differences can be seen. All returns are lower and standard deviations are generally higher. Some countries seem to be more affected by the crisis than others. For example Spain, Italy, France and the U.S. are affected the most and the Scandinavian countries, Australia and Germany seem less effected whereas New Zealand is almost not affected. This can be explained by the fact that the crisis predominantly hits Europe and the U.S., while Scandinavia, Australia and New Zealand are less influenced. The premiums found in the normal period are now less clear, change or even disappear. On average all returns decline for every country, some even realizing negative returns over this period. The crisis thus influences at least one side of the sentiment-return relationship. When looking at investor sentiment once again the effects of the crisis become clear. In the normal period the consumer confidence proxy means are higher on average than in the crisis period on average. This seems natural as during a crisis people are less optimistic or even pessimistic. Optimism should lead to overvaluation and lower future returns and pessimism should lead to undervaluation and higher future returns. The crisis period shows consumers to be pessimistic or at least be less optimistic so that the relation could differ. The effects of the crisis on the sentiment return relationship can be significant as clearly both sentiment and returns are affected. The descriptive statistics for the consumer confidence indices show a high degree of serial correlation in time-series. First-order autocorrelations are high and uniformly above 80%. Sentiment is quite variable and it shows strong persistence. The persistence can be explained to be a good characteristic for sentiment, as it suggests that investors are not too unsteady or inconstant and waves of optimism and pessimism may reinforce themselves. In the empirical analysis this serial correlations will be taken care of Preliminary tests The high autocorrelations in tables 3 and 4 suggest it to be interesting to test if the consumer confidence indices are unit-root non-stationary. These panel unit-root tests are displayed in table 6. The table includes: (a) tests for a common unit root (Levin, Lin, Chu, 2002), (b) individual unit-roots in the 12 consumer confidence series (Im, Pesaran and Shin) and (Philip-Perron-Fisher). The results show that consumer confidence, a proxy for sentiment, is data that is stationary, but highly persistent. Persistence is found to be a good characteristic as explained earlier. That sentiment turns out to be stationary is also statistically comforting. A stationary process is a stochastic process whose -19-