INFLUENCE OF CULTURE ON INVESTOR SENTIMENT

(1)

INFLUENCE OF CULTURE ON INVESTOR SENTIMENT

A research based on the cultural dimension framework by Hofstede &

Minkov

by

Wim Beekhuis

UNIVERSITY OF GRONINGEN

Faculty of Economics and Business

(2)

ABSTRACT

This study investigates in 18 countries whether investor sentiment differs. In addition, the study explores if culture influences investor sentiment. Culture is based on a cultural dimension framework developed by Hofstede, Hofstede and Minkov (2010). Investor sentiment is based on Bandopadyaya and Jones (2006) who developed the Equity Market Sentiment Index, which is an indirect measure of investor sentiment. The results on an ANOVA test show no evidence of differences in investor sentiment between countries. In addition, the study does not find evidence that a relation exists between investor sentiment and cultural differences except for a very weak relationship with uncertainty avoidance.

Keywords: behavioral finance, investor sentiment, Hofstede cultural dimension framework, ANOVA, multivariate regression model

(3)

1. INTRODUCTION

For almost half a century the Efficient Market Hypothesis (EMH) is one of the leading views about the capital markets. The EMH asserts that financial markets are “informationally efficient”. That is, one cannot consistently achieve returns in excess of average market returns on a risk-adjusted basis, given that the information is publicly available at the time the investment is made. EMH is the result of three conditions in the market, i.e. rationality, independent deviations from rationality and arbitrage. EMH assumes efficient markets implying that no financial bubbles or sudden market crashes will arise.

However recent crises such as the internet bubble in early 2000 and global financial crisis in 2008, show that market prices were not trading at informationally efficient prices. Amongst the criticasters of EMH are behavioral finance scientists who include social, cognitive and emotional factors in the decision making of investors.

It is acknowledged that investors are exposed to bounded rationality, i.e. investors are intended rational but constrained by their cognitive capabilities which can weaken their ability in making optimal decisions (Simon, 1957). These irrational investors are mentioned as noise traders, i.e. investors who trade without all available information. Scientific literature generally assumes two types of investors in the market, i.e. institutional versus individual. The institutional investor can be defined as investors participating for a living while individual investors are investors whose primary line of business is outside the stock market (Brown and Cliff, 2004). The individual investors are in general referred as noise traders, creating noise on the stock market.

The measurement of noise trading is difficult as most models have sources of noise trading which are difficult to retrieve (Brown and Cliff, 2004). However it is common in scientific literature to use investor sentiment as a comparison for noise traders. Investor sentiment is the expectation of investors regarding the direction of the market. Brown and Cliff (2004) define investor sentiment as ‘…the expectations of market participant relative to a norm: a bullish (bearish) investor expects returns to be above (below) average, whatever “average” may be.’

(4)

distinguishes the members of one group or category of people from others’ (Hofstede, Hofstede and Minkov, 2010). This means that culture refers to national culture. Hofstede pioneered the cross-cultural differences at national level by developing a cultural dimension framework that compares and contrasts different countries on several dimension.

This framework was created by Hofstede in the time that he worked for IBM, i.e. from 1965 to 1971. In these years, Hofstede received entrance to an IBM questionnaire about the values of people in more than fifty countries around the world. Hofstede found four dimensions which could be used to define and classify the culture of a country. The IBM survey was extended by a fifth dimension after the Chinese Value Survey (CVS) questionnaire by Hofstede and Bond in 1991. The CVS questionnaire was a list of questions that had its origin in China instead of the Western countries. Based on the World Values Survey, which is a periodic survey covering more than 100 countries, Minkov (2007 and 2009) created another dimensions model. This model added a sixth dimension to the framework. All six dimensions are described by Hofstede et al. (2010) and include the following: Power Distance (PDI), Individualism versus Collectivism (IDV), Masculine versus Feminism (MAS), Uncertainty Avoidance (UAI), Long-term versus Short-term Orientation (LTO) and Indulgence versus Restraint (IND).

This paper investigates in 18 countries whether investor sentiment differs. In addition to this comparison, I will explore if culture influences investor sentiment. The measurement of culture will be based on the cultural dimension framework as described in Hofstede et al. (2010). Although investor sentiment is not directly caused by the cultural dimension framework, given the static score of culture in this framework, influences by culture are likely to exist. Therefore this paper will answer the following question: Does investor sentiment differ between countries? A research based on a cultural dimension framework by Hofstede, Hofstede and Minkov (2010).

(5)

not find evidence that a relation exists between investor sentiment and cultural differences except for a very weak relationship with uncertainty avoidance.

De Long et al. (1990) argue that irrational investors influence the expected pattern of share price development in terms of direction (increase versus decrease) and time pad (long- versus short-term movement). Given that existing scientific literature does not include in its measurement of investor sentiment the cultural background of investors, this paper makes a start on an exploration on the relation between culture and investor sentiment. Based on the lack of relationship between culture and investor sentiment in this research, the contribution of this paper is more to show new research initiatives to use another investor sentiment measure and advice a critical assessment on the measure of culture given that the cultural dimension framework has a static nature.

This paper is structured as follows. In the next section, a literature review will be provided including cultural dimension framework by Hofstede et al. (2010), investment sentiment and the hypotheses will be formulated. Section 3 will discuss the methodology and data. Analysis of the test results are presented in section 4. And finally section 5 provides conclusions of this paper.

2. LITERATURE REVIEW

Literature

Cultural dimension framework

The word culture has many different meanings. Definitions vary from excellence of taste in the fine arts and humanities to a set of shared attitudes, values, goals and practices that characterizes an institution, organization or group. Within this study, I use the meaning of Hofstede et al. (2010) who defines culture as “the collective programming of the mind that distinguishes the members of one group or category of people from others”. They focus their research on cross-cultural differences at national level by developing a cultural dimension framework that compares and contrasts different countries on six dimensions. Countries are ranked by means of an index score. Data is extracted from a survey by IBM in the period 1967 to 1973. These dimensions are the following:

(6)

unequally. Individuals are born with different intellectual, physical and nurture capability, and these differences might lead to unequal achievement. Some societies tolerate these inequalities, while others try to correct it.

2) Individualism versus collectivism dimension (IDV). Individualism includes societies in which the ties between individuals are loose: everyone is expected to look after him- or herself and his or her immediate family. While collectivism includes societies in which people from birth onward are integrated into strong, cohesive in-groups, which throughout people’s lifetime continue to protect them in exchange for unquestioning loyalty. Opposed to the individual society, emphasized is the identity of the group to which an individual belongs to. 3) Masculine versus feminine dimension (MAS). Wherein the masculine stereotype includes assertive, tough and focused on material success behaviour, while feminine stereotype includes modest, tender and concern on the quality of life behaviour. Masculine countries encourage people pursuing personal, material and visible achievements. Masculine behaviour is not restricted to men, but across the society. Feminine countries encourage values such as patience, compassion and empathy. The emotional gender roles overlap and feminine values are adopted by men.

4) Uncertainty avoidance dimension (UAI) is the extent to which the members of a culture feel threatened by ambiguous or unknown situations. People in countries that score lower on this dimension are less risk averse, more comfortable with uncertainty and more open to other different beliefs and opinions. People in countries that score higher on this dimension are more risk averse, less comfortable with uncertainty and urge to control the future.

5) Long-term versus short-term dimension (LTO). Countries with long-term orientation stands for the fostering of virtues oriented toward future rewards. Virtues include perseverance and thrift. Countries with short-term orientations stands for the fostering of virtues related to the past and present. Virtues include respect for tradition, preservation of “face” and fulfilling of social obligations.

6) Indulgence versus restraint dimension (IND). In which indulgence stands for a tendency to allow relatively free gratification of basic and natural human desires related to enjoying life and having fun. Restraint reflects a conviction that such gratification needs to be curbed and regulated by strict social norms.

(7)

in dimensions and matrices while instead culture is a heterogeneous concept. Final criticism from Baskerville is that IBM survey data are old and obsolete as culture develops over time. This argument is supported by McSweeney (2002). Other criticism from McSweeney (2002) include that the interpretation of a respondent on a survey is influenced by the culture and therefore not the rightful method. Another argument from McSweeney is that a study to subsidiaries of one company cannot provide information about entire national cultures.

Although the cultural dimension framework is indeed a simplified representation of the culture, I receive comfort to apply Hofstede's cultural dimension framework in this paper. Due to replications of the model focused on bankers, airline pilots, civil servants and other organizations with comparable results. In addition, the culture dimension framework is widely used within scientific literature. Quote from Chapman (1997) is the following: “Hofstede's work is used and admired at a very high level of generalization. Those who take country scores in the various dimensions as given realities, informing or confirming other research, do not typically inquire into the detail of the procedures through which specific empirical data were transmuted into generalization. Hofstede, of course, provides all the background one could wish for about these procedures, and that is another reason for admiring his work”.

Efficient Market Hypothesis

EMH is the result of three conditions in the market, i.e. rationality, independent deviations from rationality and arbitrage. Rationality applies to investors that make decisions based on reasoning and optimal goals. With independent deviation from rationality is expected that although not all investors apply rational decisions, their non-rational decisions are random and therefore cancel out in the aggregate without effecting prices. And finally arbitrage is defined as investors taking advantage of a price difference between two or more markets netting out price differences. The result of these three conditions is that prices reflect the fair value and that prices will adjust on information before the investor has time to act on it (Ross, Westerfield and Jaffe, 2005).

Irrational investors

(8)

practice. First explanation is overconfidence resulting in too narrow confidence intervals and poorly calibrated probabilities (Daniel, Hirshleifer and Subrahmanyam, 1998). Second explanation is optimism and wishful thinking as most people display unrealistic rosy views of their abilities and prospects. Third explanation is representativeness wherein people judge the probability or frequency of a hypothesis by considering how much the hypothesis resembles available data. Common biases herewith are base rate neglect and sample size neglect (Barberis, Shleifer and Vishny, 1998). Fourth explanation is conservatism as people over-emphasize base rates relative to sample evidence. Fifth explanation is belief perseverance as once people have formed an opinion, they cling to it too tightly and for too long. People are reluctant to search for evidence that contradicts their belief and if they find such evidence, they treat it with excessive scepticism. Sixth explanation is anchoring as people who have formed an estimation rely too much on an initial, possible arbitrary value. The result is that the adjustment on experimental data is insufficient. Final explanation is availability biases as people tend to search their memories for relevant information even if it is not similarly applicable.

Other explanation is given by Barberis, Huang and Santos (2001) who introduced the loss-aversion principle for investors. This principle is that investors suffer greater disutility from a wealth loss than utility from an equivalent wealth gain in absolute term. Empirically, the presence of irrationality is proved by research in which certain investor mood characteristics significantly correlates with stock market returns. Amongst others, next articles empirically support the irrationality. Saunders (1993) found negative stock returns on the NYSE with cloudy weather in New York City. Hirshsleifer and Shumway (2003) found positive daily returns after morning sunshine in the city of a country leading stock exchange. Edmans, Garcia and Norli (2007) found influences of sudden sport sentiment changes after soccer, cricket, basketball and rugby games on stock returns.

Noise traders

(9)

short-term, positions that arbitrageurs take against noise traders will not always result in return of stock price to its real value. Second assumption is that investors are subject to sentiment as they seem to respond to changes in expectations or sentiment which are not fully justified by information (Shleifer and Summers, 1990).

The measurement of noise trading is difficult as most models have sources of noise trading which are difficult to measure (Brown and Cliff, 2004). In addition, mispricing is hard to prove given that tests for mispricing should include a proper model of discounting. Unfortunately this depends on assumptions for the future (Barberis and Thaler, 2003). Therefore, I need to find other variables that represent information on noise trading.

It is common in scientific literature to use investor sentiment as comparison for noise trading (e.g. Brown, 1999). It addition it can be argued that irrationality of noise traders is caused by changes in sentiment. This argument is supported in the article by Shleifer and Summers (1990). Shleifer and Summers argue in this article that not all demand changes are rational and fully justified by information. Response to pseudo-signals provides investors the impression of change in future returns while these signals would not be included in a fully rational model. Examples of pseudo-signals are advice of brokers or financial gurus and demand shifts caused by ‘technical analysis’ (i.e. theory that calls for buying more stock when stocks have risen through a barrier and selling stock when stocks have fallen through a floor). These judgment biases, i.e. sentiment, in procession of information are correlated, leading into aggregate demand shifts.

This paper will use investor sentiment as an estimate for noise trading due to the close relation between noise trading and investor sentiment and the widespread use of investor sentiment as an estimate for noise trading in the literature.

Investor sentiment

(10)

information (Baker and Wurgler, 2006; Brown and Cliff, 2004), closed-end fund discount (Baker and Wurgler, 2006; Brown and Cliff, 2004; Neal and Wheatley, 1998), derivative trading activity (Brown and Cliff, 2004) and equity issues over total net issues (Baker and Wurgler, 2006). Given that direct investment sentiment measures are not public available, this research will use indirect measures of investor sentiment. The following three articles include these indirect measures.

Baker and Wurgler (2007) created an operational sentiment index to find the effects of sentiment on the market. Their method is focused on the question if certain stocks are more prone to movement of investor sentiment. Several proxies in an investor sentiment index were described. All proxies were filtered by the following arguments: data availability and largest focus on long-term horizon returns with which they suggest a contrarian trading strategy. Based on above filters, Baker and Wurgler came up with the following five proxies that predicts investor sentiment:

- Trading volume is based on the argument that if short-selling is costlier than opening and closing long positions (as it is in practice), irrational investors are likely to trade, and thus add liquidity, when they are optimistic and betting on rising stocks than when they are pessimistic and betting on falling stocks. Trading volume is measured by market turnover, the ratio of trading volume to the number of shares listed on the stock exchange.

- Dividend premium is based on the assumption that dividend-paying stocks resemble bonds in that their predictable income stream represents a salient characteristic of safety. The premium that investors are willing to pay for dividend-paying stocks are therefore inverse related to sentiment. Dividend premium is defined as the difference in the logs of the average market-to-book ratios of dividend payers and non-payers.

- IPO first-day returns: IPO’s sometimes earn such a remarkable return on their first trading day that it is difficult to find an explanation that does not involve investor enthusiasm. - Closed-end fund discount is the difference between the net asset value of a fund’s actual

security holdings and the fund’s market price. If closed-end equity funds are disproportionately held by retail investors, the average discount on closed-end equity funds may be a sentiment index, with the discount increasing when retail investors are bearish.

(11)

- Equity issues over total net issues is a broader measure of equity financing activity. This is the equity share of total equity and debt issues by all corporations. Summarized this proxy measures all equity offerings, not just IPO’s.

Brown and Cliff (2004) investigated investor sentiment and its relation to near-term stock market returns. They defined investor sentiment as the expectations of market participants relative to a norm: a bullish (bearish) investor expects returns to be above (below) average, whatever “average” may be. Given that Brown and Cliff (2004) researched the relation to near-term stock market, they employed a vector autoregression framework to assess different investor sentiment measures. These measures includes direct investor sentiment measures such as AAII and II and indirect investor sentiment measures such as the ratio of the number advancing issues to declining issues, percentage change in margin borrowing as reported by the Federal Reserve (as it represents investors using borrowed money to invest), expected volatility relative to current volatility, closed-end fund discount, net purchases of mutual funds, IPO first-day return and IPO volume. Brown and Cliff found strong relations between many indirect measures of investor sentiment. Changes in surveys and a composited measure of investor sentiment showed high correlation with market returns. Further analysis of the correlation showed that market returns clearly cause future changes in sentiment, while little evidence is found suggesting that sentiment causes subsequent market returns.

Bandopadhyaya and Jones (2006) developed an equity market sentiment index (EMSI) based on publicly available price data in an equity market. The index in this article combines investor sentiment with the market’s willingness to accept whatever risks are inherent in the market at a given time.

(12)

From above correlation between risk aversion and return on currency conclusion can be taken that risk aversion can be used in the short-term as predictor of investor sentiment. In the short run, risk appetite is the largest factor to influence the return. Investors take advantage on signals they spot in the market leading in correlation between risk aversion and market return. As such risk appetite is a measure of investor sentiment.

Instead of currencies, Bandopadhyaya and Jones (2006) used stock market price data for firms to calculate the equity market sentiment index (EMSI). In this article they demonstrate that the technique developed in Kumar and Persaud (2002) can be applied to an equity market setting by constructing the EMSI for a group of firms in the Massachusetts Bloomberg Index (MBI). Their method involves a comparison of news in Boston Globe affecting firms in Massachusetts and the economy of Massachusetts. They find that movements in EMSI correspond to positive and negative news events affecting firms in Massachusetts and also that changes in EMSI replicate changes in MBI.

This research is focused on comparing investor sentiment in different countries. The comparison is best when a single number for investor sentiment is used. As Brown and Cliff (2004) do not develop a single measure/index of investor sentiment, their methodology is not effective with my research objective. In addition, Baker and Wurgler (2007) constructed an investor sentiment index with several proxies which were best suitable for their research. Their methodology is not effective with my research objective given that not all proxies of Baker and Wurgler’s investor sentiment index are publicly available, i.e. retail held ownership of closed-end fund. However, some proxies from Baker and Wurgler are useful in my research, this is further described within the methodology paragraph. Given that Bandopadhyaya and Jones (2006) focused on developing a single investor sentiment index based on publicly available equity prices, I deem their sentiment measure most suitable to compare between the countries.

Hypothesis

The research will try to answer the following question:

Does investor sentiment differ between countries? A research based on a cultural dimension framework by Hofstede, Hofstede and Minkov (2010).

(13)

moderately risk seeking, neutral, moderately risk averse and highly risk averse. Hofstede and Minkov divided their cultural dimension framework in six dimensions. Therefore I want to structure my research with these six dimensions. Each hypothesis will focus on a single dimension.

PDI is the extent to which the less powerful members of institutions and organizations within a country expect and accept that power is distributed unequally. Greater power distance will mean that members of the society have a strong dependence on the power of others. Wildavsky and Dake (1990) distinguished the world in societies whose cultural values, perceptions and attitudes are shaped by either a market environment or a hierarchical and bureaucratic environment. Societies with a market environment view uncertainties as opportunities and are therefore more risk-taking. While a hierarchical and bureaucratic environment decides by standard operating procedures and are therefore more cautious and risk-averse. Underlying reason of above behaviour is that hierarchical societies try to prevent changes in their current structure leading into risk aversion. This results in the following hypothesis:

Hypothesis 1: Countries with greater power distance will have more risk-averse investor sentiment.

IDV dimension focuses either on loose ties between individuals or strong ties with groups throughout people’s lifetime. Individuals in individualistic societies have only strong ties with immediate family, while this is extended to other groups within the collectivistic society. Weber and Hsee (1998) argue that in collectivistic countries, families or in-group members will step in to help out a group member who encounters a large loss after selecting a risky option. Opposite from individualistic countries, collectivistic countries have an extra cushion to risky options. These results are supported by Fan and Xiao (2005), who found more risk taking in a collectivistic China than in an individualistic USA. As a result, I will test the following hypothesis:

Hypothesis 2: Individualistic countries will have more risk-averse investor sentiment.

(14)

research that provides empirical prove that women have lower risk tolerance than men (e.g. Hallahan, Faff and McKenzie, 2004; Carr and Steele, 2010). Explanation of this outcome includes biological differences between male and female, socialization processes and influencing stereotypes. I will test the following hypothesis:

Hypothesis 3: Femine countries will have more risk-averse investor sentiment.

UAI is the extent to which the members of a culture feel threatened by ambiguous or unknown situations. De Jong and Semenov (2002) argued that stock markets are less favourable in high uncertainty avoidance societies due to higher risks in the stock market. It implies that members of high uncertainty avoidance society are more risk averse than members of low uncertainty avoidance society. As a result, I will test the following hypothesis:

Hypothesis 4: Countries which are uncomfortable with uncertainty will have more risk-averse investor sentiment.

Research by Minkov included a dimension that divided countries into long-term versus short-term orientation. The long-short-term orientation countries are represented by values such as prestige, ordering relationships by status and observing this order, and having a sense of shame. Short-term orientation countries are represented by respect for tradition, and reciprocation of greeting, favors and gifts. Yaveroglu and Dontly (2002) and Nath and Murthy (2004) argue that long-term orientation countries are less prone to innovate as of their risk aversion. This is caused by their behaviour to prevent changes to their existing prestige and status by being risk averse. This is enforced by the opportunity that risk taking can result in failure, i.e. a sense of shame. This results in the following hypothesis:

Hypothesis 5: Long-term orientated countries will have more risk-averse investor sentiment.

Final hypothesis in this research is related to Minkov’s IND. Indulgent societies are positive, optimistic and they remember positive emotions more easily, while restrained societies are cynical, pessimistic and they remember positive emotions less easily. Ben Mansour, Jouini, Marin, Napp and Robert (2008) found that optimism is positively correlated with risk aversion while pessimism is positively correlated with risk tolerance. This results in the following hypothesis:

(15)

3. RESEARCH METHOD & DATA DESCRIPTION Research Method

In this paper, measurement of investor sentiment is based on the methodology by Bandopadyaya and Jones (2006) to calculate EMSI. In their method each country will receive a daily index. The index is made of daily returns of each security in the country’s main equity index. For each security, the average standard deviation of the daily returns over the previous five days is also computed for each day of the sample period. The daily rate of return and the historic volatility are ranked. Then the Spearman rank correlation coefficient between the rank of the daily returns for each firm and the rank of the historic volatility of the returns for each firm is computed, multiplied by 100. EMSI is calculated with the following formula:

1 = ∑ − −

∑ − − × 100

where EMSI is Equity Market Sentiment Index, and are the rank of the daily return and the historical volatility for security i, respectively and and are the population mean return and historical volatility rankings, respectively.

The outcome of the index values ranges from -100 up to 100. The outcome will be divided in the following categories: < -30 the market is highly risk averse; between -10 and -30 the market is moderately risk-averse; between -10 and 10 the market is risk neutral; between 10 and 30 the market is moderately risk seeking and > 30 the market is highly risk seeking.

The methodology in this paper has three tests. First test is to the investor sentiment index via the ANOVA to test the equality of means between the countries. The ANOVA will show whether the investor sentiment is the same between the countries. The significance of the difference in investor sentiment are tested with the F-value test. In case of an insignificant score on the F-value, it is less likely that cultural differences are related to investor sentiment.

Second test is to test for the significance of cultural differences. I will use a multivariate regression model that estimates whether certain variables explain investor sentiment. The model that is estimated is as follows:

(16)

The dependent variable is EMSI and consists of the equities in the main index of a country. To prevent excessive sample size I will reduce each index randomly to 10 equities. EMSI is measured on a daily basis, however other independent variables are only available in monthly data (such as IPO volume). Therefore EMSI is averaged into a monthly measure. Independent variables in the model are chosen from proxies of investor sentiment described by Baker and Wurgler (2007). Publicly availability of the information is most important in the decision to include them in the model. is monthly trading volume in a major index in a country. Baker and Wurgler (2007) used the ratio of trading volume to the number of shares listed on the main stock index. It appears that the number of shares listed is not available on Bloomberg. Therefore I decided to use a detrended log trading volume as per Tetlock (2007). This measure detracts a 60-day backward moving average of log trading volume from the log trading volume. is IPO volume, the number of IPO issuances per month in a country. is IPO first-day return, the average return on the first-day of all IPO issuances per month in a country. refers to index score for each country on each dimension in the cultural dimension framework. Each dimension will provide another value for . This test measures if this index score influences the investor sentiment. The significance of the coefficients of the variables, especially for , are tested with the Student T-test. The multivariate regression model is finally controlled by two dummy variables, whereby _!,# represents country and _&,# represents year.

To avoid multicollinearity in the multivariate model, the independent variables are tested on their correlation. This is a necessity given that I question how various independent variables influence the dependent variable.

(17)

I will collect data for a period of eleven years. This will result in a sample size of 132 months, which is sufficient to exploit a multivariate regression model. Given all technological changes in the 21st century that increased participation of individual investor to the equity market, I will use data from 01/01/2000 – 31/12/2010.

Data Description

In this research, most data is retrieved from Bloomberg. On Bloomberg the following information was available: daily price information to calculate EMSI, daily market turnover for selected indices, amount of IPO issuances in a country and IPO first-day return. The index score of a country on each dimension in the cultural dimension framework is retrieved from Hofstede et al. (2010). Data and considerations on each variable are described below.

(18)

TABLE 1

Descriptive statistics of the monthly EMSI per country for the period 2000-10 calculated through the method described in Bandopadyaya and Jones (2006)

A u st ri a B el g iu m B ra zi l C a n a d a F in la n d F ra n ce G er m a n y G re ec e H o n g K o n g In d ia * J a p a n N et h er la n d s P o rt u g a l S p a in S u is se S w ed en U K U S A Mean (1.43) (1.40) (0.66) (0.72) (2.05) (2.02) (1.95) (1.19) (0.56) (1.62) (0.67) (1.72) (3.18) (2.22) (1.48) (1.34) (2.24) (1.05) Maximum 20.02 23.67 20.78 20.40 18.45 22.55 21.64 25.76 18.50 21.79 23.31 16.04 21.38 22.85 20.33 21.45 18.40 22.24 Minimum (24.32) (21.92) (26.60) (25.82) (23.96) (21.91) (27.40) (24.00) (26.53) (23.41) (25.98) (21.35) (25.29) (31.67) (21.31) (24.88) (22.41) (26.21) Standard deviation 8.64 9.15 9.07 8.71 8.43 7.83 8.99 9.54 8.79 9.49 9.72 7.84 9.37 9.27 9.01 8.57 7.89 8.63 Skewness (0.18) 0.11 (0.27) (0.21) (0.20) 0.04 0.01 0.34 (0.29) 0.09 (0.15) (0.08) 0.27 (0.04) 0.10 (0.16) 0.02 (0.41) Kurtosis (0.01) (0.30) 0.14 0.09 (0.10) (0.04) (0.03) 0.05 0.04 (0.30) 0.06 (0.46) 0.11 0.31 (0.15) 0.50 (0.05) 0.36

* Due to insufficient information on trading volume in India, dataset for India starts only from April 2000 until December 2010.

TABLE 2

Descriptive statistics of the detrended log monthly trading volume per country for the period 2000-2010 as per Tetlock (2007)

A u st ri a B el g iu m B ra zi l C a n a d a F in la n d F ra n ce G er m a n y G re ec e H o n g K o n g In d ia * J a p a n N et h er la n d s P o rt u g a l S p a in S u is se S w ed en U K U S A Mean 0.02 0.02 (0.04) 0.01 0.01 0.02 0.01 0.01 0.02 (0.01) 0.02 0.01 0.01 0.02 0.03 0.01 0.02 0.00 Maximum 0.75 0.40 0.85 0.35 0.47 0.43 0.45 0.67 0.56 1.20 0.56 0.43 0.57 0.41 0.64 0.51 0.69 0.51 Minimum (0.42) (0.51) (1.91) (0.28) (0.39) (0.52) (0.30) (0.50) (0.41) (1.00) (0.30) (0.55) (0.52) (0.48) (0.44) (0.49) (0.61) (0.34) Standard deviation 0.20 0.19 0.28 0.13 0.17 0.18 0.15 0.22 0.20 0.25 0.14 0.18 0.24 0.17 0.17 0.19 0.18 0.15 Skewness 0.37 (0.23) (2.76) 0.07 (0.08) (0.43) 0.03 0.03 0.30 0.02 0.57 (0.54) 0.08 (0.32) 0.66 0.06 (0.24) 0.37 Kurtosis 0.76 (0.06) 17.28 0.09 0.17 0.22 (0.22) (0.09) (0.14) 5.13 1.07 0.72 (0.63) 0.21 2.15 (0.21) 2.00 0.62

(19)

Sweden. The results are more or less comparable with Bandopadhyaya and Jones (2006) who researched the EMSI on the Massachusetts Bloomberg Index (MBI) from 2/7/2003 until 1/7/2004. In their research, average EMSI is risk neutral (4.20), standard deviation is 16.62, maximum EMSI is 48.09 and minimum EMSI is minus 35.44. Differences could be the result of averaging EMSI to a monthly measure instead of a daily measure by Bandopadhyaya and Jones (2006) or a shorter time period for Bandopadhyaya and Jones (2006). Previous research found also varying minimum and maximum figures for risk appetite (e.g. Baek, Bandopadhyaya and Du (2005) found minimum of minus 64.6 and maximum of 61.5; Pericoli and Sbracia (2009) found minimum of minus 5 and 16) and more or less similar mean (e.g. Baek et al. (2005) at 0.055 and Baek (2006) at minus 5.5).

Trading volume for each index (i.e. cumulative trading volume of each individual stock in a major index) is available at Bloomberg. Due to my method to detrend trading volume through detraction of a 60-days backward moving average, data is required from October 1999 until December 2010. India caused a problem as trading volume from the BSE Sensex Index is only available from January 3, 2000. As a result, I will start my tests for India starting from April 2000. All other countries continue to start from January 1, 2000. Please refer to Table 2 for descriptive statistics on the detrended log trading volume per country. All countries report an average log trading volume which is close to 0.01. Kurtosis is close to zero for all countries, except for Brazil, India, Suisse and UK. Their kurtosis is 17.28, 5.13, 2.15 and 2.00, respectively, which signals for a peak distribution. Results, especially USA figures, are comparable with Baker and Wurgler (2006) which detrended log trading volume on the NYSE exchange has an average of 0.00 and a kurtosis close to zero.

(20)

TABLE 3

Amount of IPO’s per year per country for the period 2000-2010

A u st ri a B el g iu m B ra zi l C a n a d a F in la n d F ra n ce G er m a n y G re ec e H o n g K o n g In d ia * J a p a n N et h er la n d s P o rt u g a l S p a in S u is se S w ed en U K U S A T o ta l 2000 7 7 1 53 15 127 37 46 88 97 203 9 3 6 22 28 249 397 1395 2001 6 1 0 120 0 52 13 19 87 25 172 0 1 2 5 6 92 84 685 2002 2 0 1 126 1 24 7 15 104 7 128 0 0 1 1 4 70 73 564 2003 3 0 0 109 0 13 0 14 59 6 125 0 1 0 0 0 65 76 471 2004 1 1 7 173 1 23 5 9 63 20 181 1 1 3 4 4 246 213 956 2005 4 12 9 255 2 38 24 5 67 48 160 2 0 1 8 6 328 194 1136 2006 6 17 24 295 4 64 74 2 60 77 193 13 1 10 7 13 296 201 1357 2007 7 16 59 386 2 50 52 3 83 103 121 8 2 11 10 19 244 240 1416 2008 1 11 4 277 0 36 10 9 29 41 48 2 1 1 4 12 67 54 607 2009 0 2 6 154 0 1 3 2 65 17 19 1 0 1 1 1 16 76 365 2010 1 1 11 260 0 8 16 0 94 67 22 0 0 9 2 20 74 197 782 Total 38 68 122 2208 25 436 241 124 799 508 1372 36 10 45 64 113 1747 1805 9761

(21)

in Canada (Carpentier and Suret, 2011). Based on Loughran et al. (2011), data in Table 3 is comparable with existing literature.

Based on the definition by Ritter (2011), IPO first-day return is retrieved by calculation of the percentage return from the offering price to the closing price in Bloomberg. I used two filters on retrieved data which are 1) IPO’s greater than or equal to € 10m and 2) IPO first-day return between -90% and 500%. I introduced the first condition because Bloomberg had in general insufficient trading information and trading volume to generate a quotation for smaller IPO’s. Including these small issuances would lead to biased results due to large amount of IPO’s with zero return on its first trading day. The second condition is introduced based on conclusions from further research to extraordinary IPO first-day return within Japan by use of Kaneko and Pettway’s Japanese IPO Database. Through this database I found that with some extraordinary first-day returns Bloomberg had mistakenly used a thousand separator instead of decimal sign. To prevent inclusion of these mistakes in the database, I will only use returns between -90% and 500%. Monthly data on IPO first-day return in the USA is retrieved from the database on IPO’s from Ritter, which is available on http://bear.warrington.ufl.edu/ritter/IPOs2010Statistics_11_1_11.pdf. As per Loughran and Ritter (2004) the following IPO’s were excluded from their database: best efforts offers, American depository receipts, closed-end funds; real estate investment trusts, banks, savings and loans, partnerships and IPO’s with an offer price below $ 5.00 per share. Although Loughran and Ritter had more stringent IPO requirements than I require in this research, I believe that his IPO first-day return database is a more reliable source for first-day return in USA than my database. Due to Ritter’s assessment of each IPO in detail which is impossible for me with 9,761 IPO’s in total.

(22)

TABLE 4

Descriptive statistics of the first-day return per month per country for the period 2000-2010

A u st ri a B el g iu m B ra zi l C a n a d a F in la n d F ra n ce G er m a n y G re ec e H o n g K o n g In d ia * J a p a n N et h er la n d s P o rt u g a l S p a in S u is se S w ed en U K U S A Months with IPO 24 16 46 113 14 54 57 32 113 81 105 17 8 25 10 6 120 121 Mean (%) (6.72) 15.58 (11.95) (0.86) 16.72 (4.88) 6.49 (10.90) 2.80 1.02 (13.00) 8.72 5.31 1.68 (3.10) (42.99) 15.81 15.15 Maximum (%) 27.35 119 180 177.75 242.86 30.19 346.87 58.94 180.12 148.33 123.75 119.72 31.25 29.79 65.41 (1.73) 300.88 116.20 Minimum (%) (75.57) 0.25 (76.88) (80.66) (34.00) (80.00) (79.20) (77.99) (75.42) (88.47) (88.82) (25.77) (33.36) (25.42) (49.54) (78.25) (24.96) (19.90) Standard deviation 0.24 0.29 0.41 0.23 0.68 0.22 0.53 0.33 0.34 0.40 0.35 0.35 0.21 0.16 0.31 0.39 0.31 0.17 Skewness (1.95) 3.44 1.93 3.46 3.20 (2.03) 4.95 (0.14) 1.76 0.48 1.16 2.60 (0.66) 0.13 0.91 0.11 6.54 3.12 Kurtosis 4.34 12.77 10.2 31.09 11.07 4.36 31.39 (0.03) 9.51 1.99 3.25 6.84 0.78 (0.74) 2.34 (3.05) 57.41 13.52

(23)

Comparison of IPO first-day return to existing literature is difficult as each research used a different set of data due to other selection criteria and time period. Baker and Wurgler (2006) had an average of 15.1% on IPO first-day return, while Brown and Cliff (2004) had an average of 16.2%. All IPO’s in my database had an average first-day return of 9.9%. This percentage is lower than in the mentioned literature, maybe due to another source of information (Bloomberg in this research). Comforting is that the countries with more months in which an IPO took place, i.e. the USA and UK, had a comparable IPO first-day return of 15.5%.

All scores on Hofstede’s dimensions used in this paper are retrieved from Hofstede et al. (2010). In this book, a survey by Hofstede on IBM employees in 76 countries was summarized into index scores. This survey was focused on four dimensions, i.e. PDI, IDV, MAS and UAI. The other two dimensions, LTO and IND, were constructed from the World Values Survey by Minkov in 93 different countries. The index scores were constructed through series of questions that were designed to find out people’s attitude towards each cultural framework dimension. Factor analysis was performed to analyse the average score per dimension in a country leading into an index score. Please refer to Table 5 for index score per country on each dimension according to Hofstede’s cultural framework.

Except for two, all scores on PDI ranges between 30 and 70. One exception is Austria which has an index score of 11 implying marginal inequality. The other exception is India with an index score of 77 implying more inequality between members of the society. Most Western European and Anglo World countries score relatively low on power distance.

Countries that scored high on PDI score generally low on IDV and vice versa. Most individualistic countries are UK and USA with index scores at 89 and 91, respectively. Most collectivistic countries are Hong Kong and Portugal with index scores at 25 and 27, respectively.

(24)

TABLE 5

Index score per country on each dimension according to cultural dimension framework (Hofstede et al. 2010)

PDI IDV MAS UAI LTO IND

Austria 11 55 79 70 60 63 Belgium 64* 75* 52* 95* 82 57 Brazil 69 37 49 76 44 59 Canada 39 80 52 48 36 68 Finland 33 63 26 59 38 57 France 68 71 43 86 63 48 Germany 35 67 66 65 83 40 Greece 60 35 57 112 45 50 Hong Kong 68 25 57 29 61 17 India 77 48 56 40 51 26 Japan 54 46 95 92 88 42 Netherlands 38 80 14 53 67 68 Portugal 63 27 31 104 28 33 Spain 57 51 42 86 48 44 Suisse 48* 67* 65* 63* 74 66 Sweden 31 71 5 29 53 78 UK 35 89 66 35 51 69 USA 40 91 62 46 26 68

* Index score is the average of two scores as bilingual countries, i.e. Belgium and Suisse, received a score per language area.

(25)

On LTO Asian and European countries score in generally high. While the Anglo, African and Middle East countries score in generally low. In my database of 18 countries most long-term oriented countries are Belgium (82), Germany (83) and Japan (88). Countries which are more oriented on the short-term are Portugal (28) and USA (26).

IND column is the last dimension in Table 5. As of Hofstede et al. (2010) indulgence is negatively correlated with long-term orientation. As a result, West European and Anglo countries are more indulgent while Asian countries are more restraint. Most indulgent countries are Sweden (78) and UK (69) and most restraint countries are Hong Kong (17) and India (26).

4. RESULTS

In this chapter I will describe the results on the ANOVA test, multivariate model and multicollinearity test. Based on these test results, I will draw conclusions on all hypotheses imposed in chapter Literature Review.

ANOVA

With an ANOVA test the equality of means will be tested between the 18 selected countries to see whether EMSI differs between selected 18 countries. Before I am able to perform an ANOVA test, I have to check whether the assumptions of this test technique are satisfied. First assumption of independence between groups is met given that each group is based on country origin. Second assumption is a normal distribution of the population. The complete EMSI population passed the Kolmogorov-Smirnov (K-S) test. EMSI population specified to countries showed normality as well, except for Portugal and the USA that failed on the K-S test marginally. Third assumption is homogeneity of variances which is tested by the Levene Statistic. This value is described in Table 6. For the full period, Levene Statistic is insignificant, which means that variances in population deemed to be homogene. Split into years, only 2004 and 2007 report a significant score. Based on above results, I am comfortable that the EMSI database satisfies with the requirements for an ANOVA test.

(26)

TABLE 6:

Levene Statistic, F-value and significance of the tests (between brackets) in the ANOVA test regarding the difference of EMSI between the selected 18 countries from 2000 to 2010.

Levene Statistic F-value

2000 0.730 (0.770) 1.232 (0.242) 2001 0.634 (0.862) 1.300 (0.195) 2002 1.133 (0.325) 1.787 (0.032)** 2003 1.388 (0.145) 0.276 (0.998) 2004 1.724 (0.041)** 1.061 (0.395) 2005 1.129 (0.329) 1.402 (0.138) 2006 0.404 (0.984) 0.849 (0.635) 2007 1.526 (0.089)* 1.010 (0.449) 2008 0.547 (0.926) 0.679 (0.821) 2009 0.986 (0.476) 0.444 (0.973) 2010 1.393 (0.143) 0.513 (0.945) Overall 0.828 (0.661) 0.784 (0.714)

With ***, **, * significant at the 1%, 5%, 10% significance levels, respectively.

Additional research to EMSI with the ANOVA test is based on region categorization by Hofstede et al. (2010). They divided the world in six regions: America Central/South, Europe South/Southeast, Europe North/Northwest + Anglo World, Europe Central/East + Ex-Sovjet, Muslim World + Middle East + Africa and finally Asia East/Southeast. The selected 18 countries originate from four regions. ANOVA test based on these regions instead of countries showed significant difference of means at 10% significance level (F-value of 2.411 and significance of 0.065). Based on this I can conclude that EMSI differs between America C/S, Europe S/SE, Europe N/NW + Anglo World and Asia E/SE.

(27)

The corresponding non-parametric test to ANOVA is the Kruskal-Wallis (K-W) test. This technique tests whether samples originate from the same distribution by comparing the median in different samples. The P-value on K-W test is 0.422, supporting the conclusion that EMSI is similar between the selected 18 countries. Winsorizing of EMSI up to 95% does not result into a different conclusion (P-value of 0.415).

ANOVA test on the independent variables trading volume, amount of IPO’s and IPO first-day return shows that strong significant evidence exists of differences between the selected countries on IPO amount and IPO first-day return. Test is not significant for trading volume. Please refer to Table 7 for results on these tests. Although it is not clear what caused the difference in amount of IPO’s and IPO first-day return, further regression analysis could show that partly is caused by cultural aspects. At least similarity in EMSI is not because of complete similar distribution of investor sentiment proxies.

TABLE 7:

Levene Statistic, F-value and significance of the tests (between brackets) in the ANOVA test regarding three investor sentiment proxies and selected 18 countries.

Levene Statistic F-value

Trading volume 6.212 (0.000)*** 0.762 (0.739) Amount of IPO 132.495 (0.000)*** 142.359 (0.000)*** IPO first-day return 15.239

(0.000)***

9.496 (0.000)*** With ***, **, * significant at the 1%, 5%, 10% significance levels, respectively.

The insignificance on the ANOVA test between countries leads to the conclusion that investor sentiment between countries is similar to each other. In addition it is questionable that I will find a relation between investor sentiment and scores on cultural dimensions which only differ per country in the multivariate regression model. The moderately weak evidence on differences in investor sentiment in regions does not change this expectation. The significant difference between countries on amount of IPO’s and IPO first-day return needs further examination by a regression analysis. This and the original multivariate regression model will be described in the following paragraph.

Multivariate regression model

(28)

TABLE 8:

Coefficients and T-test values (between brackets) of the multivariate regression model without data mining techniques

)*: Constant -0.347 (-0.329) -0.544 (-0.228) -0.315 (-0.317) -0.758 (-0.311) -0.217 (-0.118) 0.640 (0.504) ,-: Volume 1.540 (1.645)* 1.540 (1.645)* 1.540 (1.645)* 1.540 (1.645)* 1.540 (1.645)* 1.540 (1.645)* ,.: Amount of IPO 0.014 (0.379) 0.014 (0.379) 0.014 (0.379) 0.014 (0.379) 0.014 (0.379) 0.014 (0.379) ,/: IPO first-day return 0.499 (0.589) 0.499 (0.589) 0.499 (0.589) 0.499 (0.589) 0.499 (0.589) 0.499 (0.589) ,0: PDI -0.004 (-0.248) ,0: IDV 0.003 (0.085) ,0: MAS -0.001 (-0.067) ,0: UAI 0.005 (0.206) ,0: LTO -0.003 (-0.085) ,0: IND -0.016 (-0.694) F-statistic 4.392*** 4.392*** 4.392*** 4.392*** 4.392*** 4.392*** R-squared 0.053 0.053 0.053 0.053 0.053 0.053 Number of observations 2,372 2,372 2,372 2,372 2,372 2,372

Each column refers to a different multivariate regression model including a single dimension from Hofstede’s cultural framework, while each row refers to the coefficient of the independent variables or other statistics from the regression.

PDI, IDV, MAS, UAI, LTO and IND represents Power Distance, Individualism versus Collectivism, Masculine versus Feminine, Uncertainty Avoidance, LTO versus STO and Indulgent versus Restraint, respectively.

(29)

relationship between the set of independent variables and the dependent variable EMSI. However, based on R-square, overall predictability of the multivariate model is low at a score of 5.3%. This means that very weak correlation between the set of independent variables and EMSI exists. The coefficient on volume is significant at 10%, while the coefficients on amount of IPO’s, IPO first-day return and all cultural dimensions are insignificant.

Given these results, I have exploited data mining techniques to verify the results of the original multivariate regression method. First technique was to winsorize EMSI, volume and IPO first-day return up to 5% and 95%, in order to deal with the outliers in the database. As Table 9 shows, winsorizing did not show material differences from Table 8. Second data mining technique was to replace the continuous index scores from the dimension in Hofstede et al. (2010) into decile ranking. Results of this regression analysis are shown in Table 10, but no significant changes compared to Table 8 are found. Figure 1 shows the mean plot of EMSI along all deciles defined by each cultural dimension. As Figure 1 shows, there is a non-linear relationship between EMSI and each cultural dimension. Therefore, final data mining technique was replacing the cultural scores on the dimensions by natural logarithm. Results of this regression analysis are shown in Table 11.

The results on the multivariate regression model after combined implementation of all the previous described data mining techniques are described in Table 12. F-statistic of the model remained significant at <1% level. Predictability of the multivariate model remained low at a score of 5.2%.

The coefficients on the control variables IPO amount, IPO first-day return and volume are not significantly different from zero in the model. Volume is closest to significance at a level of 15%. Please note that while coefficients of control variables do not significantly differ from zero, reported coefficients are positive. Increased investor sentiment, i.e. increased risk seeking behavior, is in line with existing scientific literature in which higher trading volume, issuance of IPO’s and IPO first-day returns signals an optimistic investor sentiment (Baker and Wurgler (2007); Brown and Cliff (2004). Finally, the coefficient on UAI improved to significance (at 7.4%) reporting a coefficient of 1.071. Improvement is due to replacement of index scores into decile ranking. Although significant at <10% level, I deem this as weak evidence as significance does not come close to more stringent significance levels.

(30)

TABLE 9:

Coefficients and T-test values (between brackets) of the multivariate regression model whereby EMSI, volume and IPO first-day return are winsorized up to 5% and 95%.

)*: Constant -0.302 (-0.316) -0.678 (-0.311) -0.338 (-0.376) -0.191 (-0.087) 0.007 (0.004) 0.688 (0.599) ,-: Volume -1.449 (-1.441) -1.449 (-1.441) -1.449 (-1.441) -1.449 (-1.441) -1.449 (-1.441) -1.449 (-1.441) ,.: Amount of IPO 0.011 (0.326) 0.011 (0.326) 0.011 (0.326) 0.011 (0.326) 0.011 (0.326) 0.011 (0.326) ,/: IPO first-day return 0.760 (0.419) 0.760 (0.419) 0.760 (0.419) 0.760 (0.419) 0.760 (0.419) 0.760 (0.419) ,0: PDI -0.005 (-0.360) ,0: IDV 0.006 (0.194) ,0: MAS 0.000 (-0.022) ,0: UAI -0.002 (-0.105) ,0: LTO -0.006 (-0.194) ,0: IND -0.017 (-0.779) F-statistic 4.284*** 4.284*** 4.284*** 4.284*** 4.284*** 4.284*** R-squared 0.052 0.052 0.052 0.052 0.052 0.052 Number of observations 2,372 2,372 2,372 2,372 2,372 2,372

(31)

TABLE 10:

Coefficients and T-test values (between brackets) of the multivariate regression model with all index scores on cultural framework dimensions transformed in decile.

)*: Constant 0.591 (0.468) -1.145 (-0.643) -0.400 (-0.392) 0.699 (0.385) -0.491 (-0.270) -0.769 (-0.582) ,-: Volume -1.540 (-1.645)* -1.540 (-1.645)* -1.540 (-1.645)* -1.540 (-1.645)* -1.540 (-1.645)* -1.540 (-1.645)* ,.: Amount of IPO 0.014 (0.379) 0.014 (0.379) 0.014 (0.379) 0.014 (0.379) 0.014 (0.379) 0.014 (0.379) ,/: IPO first-day return 0.499 (0.589) 0.499 (0.589) 0.499 (0.589) 0.499 (0.589) 0.499 (0.589) 0.499 (0.589) ,0: PDI -0.098 (-0.645) ,0: IDV 0.188 (0.694) ,0: MAS 0.008 (0.067) ,0: UAI -0.218 (-0.935) ,0: LTO 0.025 (0.085) ,0: IND 0.126 (0.694) F-statistic 4.392*** 4.392*** 4.392*** 4.392*** 4.392*** 4.392*** R-squared 0.053 0.053 0.053 0.053 0.053 0.053 Number of observations 2,372 2,372 2,372 2,372 2,372 2,372

(32)

FIGURE 1

(33)

TABLE 11:

Coefficients and T-test values (between brackets) of the multivariate regression model with all index scores on cultural framework dimensions transformed in natural logarithm.

)*: Constant 1.049 (0.594) -1,483 (-1.344) -0.392 (-0.413) -2.153 (-2.270) 0.276 (0.262) -0.320 (-0.337) ,-: Volume -1.540 (-1.645)* -1.540 (-1.645)* -1.540 (-1.645)* -1.540 (-1.645)* -1.540 (-1.645)* -1.540 (-1.645)* ,.: Amount of IPO 0.014 (0.379) 0.014 (0.379) 0.014 (0.379) 0.014 (0.379) 0.014 (0.379) 0.014 (0.379) ,/: IPO first-day return 0.499 (0.589) 0.499 (0.589) 0.499 (0.589) 0.499 (0.589) 0.499 (0.589) 0.499 (0.589) ,0: PDI -0.626 (-0.694) ,0: IDV 0.787 (0.935) ,0: MAS 0.031 (0.067) ,0: UAI 1.094 (1.652)* ,0: LTO -0.482 (-0.592) ,0: IND -0.065 (-0.067) F-statistic 4.392*** 4.392*** 4.392*** 4.392*** 4.392*** 4.392*** R-squared 0.053 0.053 0.053 0.053 0.053 0.053 Number of observations 2,372 2,372 2,372 2,372 2,372 2,372

(34)

TABLE 12:

Coefficients and T-test values (between brackets) of the multivariate regression model based on section 3 and data mining techniques described in section 4

PDI IDV MAS UAI LTO IND All cultural

dimensions

All cultural dimensions (excl. IDV) 1*: Constant 1.103 (0.691) -0.152 (-0.150) -0.360 (-0.419) -2.084 (-2.428)** 0.261 (0.271) -0.339 (-0.394) 1.262 (0.672) 1,875 (1.037) 2-: Volume 1.449 (1.441) 1.449 (1.441) 1.449 (1.441) 1.449 (1.441) 1.449 (1.441) 1.449 (1.441) 1.405 (1.399) 1.432 (1.427) 2.: IPO amount 0.011 (0.326) 0.011 (0.326) 0.011 (0.326) 0.011 (0.326) 0.011 (0.326) 0.011 (0.326) 0.033 (1.215) 0.019 (0.773) 2/: IPO first-day return 0.760 (0.419) 0.760 (0.419) 0.760 (0.419) 0.760 (0.419) 0.760 (0.419) 0.760 (0.419) 0.122 (0.070) -0.253 (-0.149) 20: PDI -0.636 (-0.779) -0.636 (-1.218) -0.774 (-1.519) 20: IDV -0.150 (-0.194) 0.496 (1,211) 20: MAS 0.009 (0.022) -0.203 (-0.690) -0.381 (-1.141) 20: UAI 1.071 (1.787)* 0.270 (1.064) 0.182 (0.747) 20: LTO -0.448 (-0.604) -0.300 (-1.007) -0.167 (0.604) 20: IND -0.019 (-0.022) -0.829 (-1.685)* -0.526 (-1.242) F-statistic 4.284*** 4.284*** 4.284*** 4.284*** 4.284*** 4.284*** 6.336*** 6.606*** R-squared 0.052 0.052 0.052 0.052 0.052 0.052 0.049 0.048 Number of observations 2,372 2,372 2,372 2,372 2,372 2,372 2,372 2,372

Each column refers to a different multivariate regression model including a single dimension from Hofstede’s cultural framework (except for the last two columns, which are added to find the effect of multicollinearity from IDV), while each row refers to the coefficient of the independent variables or other statistics from the regression.

PDI, IDV, MAS, UAI, LTO and IND represents Power Distance, Individualism versus Collectivism, Masculine versus Feminine, Uncertainty Avoidance, LTO versus STO and Indulgent versus Restraint respectively.

(35)

Given that even after data mining techniques the results of the multivariate regression model does not support a relationship between investor sentiment and cultural dimensions, I will explore if the non-existence of a relationship is supported by marginal deviated multivariate regression models or disturbing influences of certain variables.

Table 13 shows the result of a multivariate regression model excluding the dummy variable country as there is a possibility that the country dummy already captures the cross-national cultural difference. Table 13 does not show significant improvement compared to Table 12 (significance on UAI disappeared). Due to the significant result on the ANOVA test whereby the population was divided by region instead of countries and the significance on only 2002 figures, I will exploit two additional multivariate regression models which remain ceteris paribus except for one variable or data input. In the first model, dummy variables for country are substituted into region. In the second model, whole period from 2000 to 2010 is replaced by 2002 figures. Both models continue to show a relationship between EMSI and set of independent variables (F-value is significant at a <1% level and <10% level, respectively) and low predictability (approximately 5%). All coefficients on Hofstede’s cultural dimensions are insignificant, therefore results on these multivariate regression models continue to support the results from the original model in which no relation exists between investor sentiment and cultural framework dimensions. Even the moderate significance on the dimension UAI has disappeared.

(36)

TABLE 13:

Coefficients and T-test values (between brackets) of the multivariate regression model without dummy variable on country.

)*: Constant -0.322 (-0.368) -0.829 (-1.290) -0.355 (-0.526) -1.020 (-1.599) -0.525 (-0.828) -0.497 (-0.730) ,-: Volume -1.446 (-1.441) -1.446 (-1.441) -1.446 (-1.441) -1.446 (-1.441) -1.446 (-1.441) -1.446 (-1.441) ,.: Amount of IPO 0.032 (1.453) 0.032 (1.453) 0.032 (1.453) 0.032 (1.453) 0.032 (1.453) 0.032 (1.453) ,/: IPO first-day return -0.143 (-0.085) -0.143 (-0.085) -0.143 (-0.085) -0.143 (-0.085) -0.143 (-0.085) -0.143 (-0.085) ,0: PDI -0.206 (-0.553) ,0: IDV 0.106 (0.422) ,0: MAS -0.214 (-0.896) ,0: UAI 0.235 (1.056) ,0: LTO -0.138 (-0.568) ,0: IND -0.147 (-0.509) F-statistic 8.191*** 8.191*** 8.191*** 8.191*** 8.191*** 8.191*** R-squared 0.046 0.046 0.046 0.046 0.046 0.046 Number of observations 2,372 2,372 2,372 2,372 2,372 2,372

(37)

TABLE 14:

Coefficients and T-test values (between brackets) of the multivariate regression model with amount of IPO as dependent variable.

)*: Constant -1,715 (-1,476) 31.911 (11.131)*** 3.673 (4.555)*** 5.478 (2.025)** 67.539 (24.330)*** 20.913 (12.256)*** ,-: Volume -1.282 (-2.365)** -1.282 (-2.365)** -1.282 (-2.365)** -1.282 (-2.365)** -1.282 (-2.365)** -1.282 (-2.365)** ,.: EMSI 0.005 (0.379) 0.005 (0.379) 0.005 (0.379) 0.005 (0.379) 0.005 (0.379) 0.005 (0.379) ,/: IPO first-day return 0.917 (1.872)* 0.917 (1.872)* 0.917 (1.872)* 0.917 (1.872)* 0.917 (1.872)* 0.917 (1.872)* ,0: PDI 1.879 (5.886)*** ,0: IDV -7.267 (-9.276)*** ,0: MAS -0.202 (-0.903) ,0: UAI -0.633 (-0.903) ,0: LTO -15.814 (-21.265)*** ,0: IND -4.374 (-9.276)*** F-statistic 99.834*** 99.834*** 99.834*** 99.834*** 99.834*** 99.834*** R-squared 0.561 0.561 0.561 0.561 0.561 0.561 Number of observations 2,372 2,372 2,372 2,372 2,372 2,372