Do country differences in individualism affect the
impact of sentiment on the cross-section of stock
returns?
Lieneke Louisa Ross University of Groningen
S1895346 Faculty of Economics and Business
Grote Kromme Elleboog 7 Msc. Finance
9712 BJ Groningen January 2015
Do country differences in individualism affect the
impact of sentiment on the cross-section of stock
returns?
ABSTRACT
This thesis investigates if there is a cultural dimension in the way investor sentiment influences the cross-section of future stock returns. To this end, I compare stock returns of the Netherlands (a highly individualistic country) and Portugal (a highly collectivistic country) after periods of high and low sentiment. In line with theory, I find evidence of a higher sensitivity of investor sentiment on stock returns for the Netherlands and Portugal. The results are more pronounced for the Netherlands. Furthermore, results indicate that stock returns of individualistic countries are more sensitive to sentiment than collectivistic countries.
JEL classifications G14, G12
1. Introduction
The efficient market hypothesis (EMH), which is widely used in modern finance theory, assumes that investors always behave rationally, errors of investors are uncorrelated and that there is no limit to arbitrage in the stock market. However, recent incidences in history reveal that these assumptions do not always hold. For example, in the late 1990s tech stocks faced an enormous price inflation, which eventually led to a crash in 2000. Some academics suggest that this implosion of stock prices can be attributed to investor sentiment. That is, the general wave of optimism among investors about the future performance of tech stocks may have led to the overvaluation of the stocks. Another example is the recent financial crisis of 2008, where the collapse of several financial institutions led to a general lack of trust in the financial system. It can be argued that overall pessimism among investors, caused by the financial crisis, led to the subsequent devaluation of stocks.
The existence of investor sentiment and its influence on stock returns is supported in prior literature (see, e.g., Shiller, 2000; Baker and Wurgler, 2000). Baker and Wurgler (2006) find that in particular low capitalization, young, unprofitable, non-dividend-paying, growth stocks and stocks from firms in financial distress are more prone to sentiment. The future returns of these stocks are lower after a period of high sentiment and higher after a period of low sentiment. The rationale behind this negative relation between sentiment and the future returns of these stocks is that they are harder to value and more difficult to arbitrage. In addition to this finding, Schmeling (2009) suggests that this effect of sentiment on stock returns depends on country-specific cultural factors. He finds that countries that are culturally more sensitive for the overreaction of investors, are more prone to sentiment as well. Because features of a culture are fixed and not easy to change, the effect of sentiment should be therefore more persistent in sentiment sensitive countries. The question that arises is which cultural factor is able to explain the difference in sensitivity to sentiment between countries.
When investors are aware of the existence of sentiment in the stock market and its influence on the future returns of certain stocks, they are able to anticipate on it by adapting their portfolio allocation on the prevailing sentiment. For example, when investors experience a general wave of optimism in the stock market about certain stocks, they can go short in these stocks because the future returns are expected to be lower. The difference in sentiment sensitivity between countries can be important for investors that are active on foreign stock markets as well, because the fluctuation in future stock returns will be higher for countries that are more prone to sentiment. This means that the future stock returns may be higher in these countries, than countries that are more ‘immune’ for sentiment. This has implications for the optimal portfolio allocation of the cross-border investor.
In order to test if individualism explains the sensitivity to sentiment in a country, I compare the impact of sentiment on the future stock returns of the Netherlands, an individualistic country, with Portugal, a collectivistic country, between respectively 2001-2013 and 2004-2013. First, I use the principal component of three orthogonalized sentiment proxies of Baker and Wurgler (2006) to form a sentiment index for both countries separately, as well as a joint sentiment index, in order to identify periods of high and low sentiment. To test for the cross-section of stock returns conditional on beginning-of-the-year sentiment, I use an univariate comparison and a time-series regression. Subsequently, I test if the sensitivity to sentiment of the Netherlands differs from Portugal, by using an independent samples t-test, as well as a time-series regression.
In contrast to the expectation that a collectivistic culture is more sensitive to sentiment, I conclude that countries with an individualistic culture are more sensitive to sentiment than countries with a collectivistic culture. For the Netherlands, both methods used in this thesis indicate that young, growth and distressed stocks are significantly more sensitive to sentiment. For Portugal, I only find significant supporting evidence for dividend-paying stocks, which is not in line with theory. Despite the insignificance, results also indicate that young, unprofitable, non-dividend-paying firms and firms with a low level of tangible assets are more sensitive to sentiment for the Netherlands. For Portugal, there is also evidence of a higher sentiment sensitivity for stocks of small, (partly) growth, and (partly) distressed firms.
These results suggests that Dutch stocks are more sensitive to sentiment than Portuguese stocks. This is confirmed by the tests that compared the sensitivity of the two countries statistically. The returns of Dutch stocks are lower after a period of high sentiment than Portuguese stocks. Despite the insignificance of the results, the observations indicate that a culture of individualism is more sensitive to sentiment than a collectivistic culture.
A possible cultural explanation for a higher sensitivity to sentiment of the Netherlands is that the Dutch culture has a lower level of uncertainty avoidance and indulgence1 than the Portuguese culture. This indicates that people from the Netherlands are willing to take more risks and have a more
optimistic view in general. In line with this assumption, it is likely that Dutch investors have more risky stocks than Portuguese investors. Stocks that are sensitive to sentiment are more risky as well, thus the impact of sentiment on future stock returns is higher in the Netherlands than in Portugal.
The remainder of this thesis is structured as follows. The next section provides a review of prior literature and derives the hypotheses. Section 2 describes the data and methodology, and Section 3 presents the results of the tests. Finally, Section 4 concludes.
2. Literature
2.1 Investor sentiment and the cross-section of stock returns
De Long, Shleifer, Summer, and Waldmann (1990) were one of the first to formalize the role of investor sentiment. They make a distinction between rational investors that are not prone to sentiment and irrational investors that are sensitive to sentiment. Their results show that irrational investors are able to earn higher expected stock returns from their own destabilizing influence than rational investors, which leads to a large difference between market prices and their fundamental values. Subsequent research investigates the predictability of investor sentiment on stock returns, mainly in the US (see, e.g., Neal and Wheatley, 1998; Baker and Wurgler, 2000; Menzly, Santos, Veronesi, 2004). However, there is no clear consensus on the effect of investor sentiment. For example, Brown and Cliff (2004) are not able to find statistical supporting evidence on the sentiment effect in the US. They only find little evidence that there is a causal relation between sentiment and the stock market returns. Contrarily, Baker and Wurgler (2007) argue that it is not about the effect of investor sentiment on stock returns, but rather about how to measure investor sentiment correctly in order to determine its effect.
suggestion of Baker and Wurgler (2006) that investors select their stocks based on characteristics which correspond with the level of optimism or pessimism that they personally have at that moment. Thus, when irrational investors are overly optimistic they will select stocks in their portfolio that have a ‘positive image’ as well.
Through the second channel, the limits to arbitrage are different across stocks, while the level of aggregate sentiment is similar for all stocks. In general, arbitrage is limited because of three problems: fundamental risk, noise trader risk and implementation costs (Ackert and Deaves, 2010). The first problem, fundamental risk, relates to unexpected new information arbitrageurs could not have anticipated on. For example, if an arbitrageur believes a stock is overvalued, he will short-sell it with the expectation that the price will be lower when he purchases the stock to close the position at a later point in time. When, contrarily to his expectation, there is positive news about the future outlook of these particular stocks, the stock price will rise and the investor will incur a loss. The exposure of arbitrageurs to this risk, may limit their trading behavior. A solution for this problem could be to use substitutes to prevent such losses. However, a perfect substitute is rare which causes arbitrage to be more risky and less likely to eliminate mispricing. The second problem, noise trader risk, relates to a further increase in the mispricing of a stock in the short run (Shleifer and Vishny, 1997). This may be driven by limits on short-sales. For individual investors, it is possible that the benefits of an arbitrage opportunity cannot be obtained, because they have a limited source of funds. For money managers, the benefits of an arbitrage opportunity may not be obtained, because of the short-run evaluation of their performance. A short-sale loss incurred to take advantageous of the arbitrage opportunity will be interpreted by other as bad performance of the money manager, which can cause a loss of its mandate. Furthermore, in some cases, it is not possible to short-sale certain stocks. The last problem that leads to limited arbitrage are the implementation costs that are incurred when trades are executed and transaction cost and market impact costs are being made. These costs are problematic when short-selling is not accessible for an investor and thus the cost of the transaction increases or when the security that needs to be sold short cannot be obtained (Ackert and Deaves, 2010). Stocks that are most sensitive to these limits of arbitrage are more likely to be mispriced.
that have a lower sentiment beta. That is, a portfolio that is long in low sentiment sensitive stocks and short in high sensitive stock is more profitable. Thus, the optimal investment decision is to hold a portfolio that has a low weight on sentiment sensitive stocks. In addition to Baker and Wurgler (2006), he finds that growth stocks are more sensitive to sentiment than distressed stocks and that profitable and unprofitable stocks of the same size have a similar sensitivity to sentiment.
The effect of sentiment on stocks that are hard to value and difficult to arbitrage is illustrated in Fig. 1. It summarizes this so-called seesaw impact of sentiment on the stocks. The x-axis represents the degree of difficulty to arbitrage the stocks and the y-axis represents the price of the stock. As can be derived from the figure, when sentiment is high, stock valuations are high for stocks that are speculative and hard to arbitrage. The opposite holds for periods of low sentiment, where the value of speculative, hard to value stocks is low.
Figure 1. Seesaw figure. The x-axis measures the level of difficulty to arbitrage the stock and the y-axis measures the price of the stock. Stocks that are speculative and harder to arbitrage will have a higher valuation when sentiment is high and a lower valuation when sentiment is low. P* represents the fundamental stock price.
Fama and French (1993, 1996) are proponents of the efficient market hypothesis and challenge the behavioral view by arguing that the higher returns of these sensitive stocks are a compensation for the higher systematic risk they face. They developed a three-factor model and examined if the cross-section of stocks returns can be explained by the size, leverage, past returns, dividend-yield, earnings-to-price ratios, and book-to-market ratios of the firms. They find that only size and the book-to-market ratio are explanatory characteristics of the cross-sectional variation in stock returns. The rationale behind the influence of these characteristics on stock returns is that size and the book-to-market ratio
1 2 3 4 5 6 7 8 9 10
Valuation Level P*
Safe, Easy to Risky, Difficult to Abitrage Abitrage
of a firm are proxies of distress. A distressed firm is more sensitive to changes in the business cycle than a financially stable firm.
Daniel and Titman (1997) investigated these findings of Fama and French (1993, 1996) and found that the higher returns of these stocks are not a compensation for higher systematic risk. The suggested co-movement of stocks by Fama and French (1993, 1996) is because stocks with similar factor sensitivities are distressed at the same time. For example, the stocks are from a similar industry or region. They conclude that firm characteristics determine expected stock returns. In addition, Baker and Wurgler (2006) control for the size and growth characteristics of the stocks in their research and still find a predictability effect of sentiment on the returns of high sentiment beta stocks. The first hypothesis in this thesis is therefore as follows:
H1: Sentiment has an impact on the future returns of hard to value and difficult to arbitrage stocks.
2.2 Cross-cultural differences in the impact of sentiment
Prior literature argues that the influence of sentiment on stock returns in a market depends on country-specific factors (see, e.g., Chang, Faff, and Hwang, 2012; Corredor, Ferrer, and Santamaria, 2013). Schmeling (2009) specifies this statement by suggesting that the influence of sentiment depends on country-specific cultural factors. He uses 18 industrialized countries in his research to investigate if the impact of sentiment is stronger on culturally more prone countries. According to the author, a country is more prone to sentiment when there is a higher herd-like overreaction of investors. Herd-like overreactions are the correlated decisions of irrational investors based on their (over)optimistic and (over)pessimistic expectations. The author proxies this herd-like overreaction by a combination of a high level of collectivism and uncertainty avoidance in a country. His results show that the effect of sentiment on stock returns of value and growth stocks is stronger for countries that have a higher herd-like overreaction. Chui, Titman, and Wei (2010) investigate if cross-country differences in culture have an impact on the stock returns of momentum strategies. They measure the difference in culture between countries by the individualism index of Hofstede (2001) and find that culture has an important impact on stock return patterns. The researchers argue that investors from a culture with a high level of individualism put more importance on their personal information and knowledge, instead of the consensus of their peers. Investors from a culture with a low level of individualism may place too much weight on the opinion of their peers and could therefore exhibit herd-like overreaction to the conventional wisdom.
integrated into groups and therefore more interdepend on their fellow group members (Hofstede, 2001). Furthermore, people in individualistic cultures are more confident about their own abilities and think more positive about themselves than people in collectivistic societies (Markus and Kitayama, 1991). In collectivistic cultures, people take responsibility for their peers as well, the interests of the group are more important than individual interests.
The level of individualism of a culture can be linked to investor sentiment. As is explained above, sentiment is the phenomenon of a large number of investors that value stocks in the same way. In collectivistic cultures, people base their decisions more on the opinion of others than in individualistic cultures, so it is likely that the valuation of stocks will be more similar for a larger amount of investors in countries with a collectivistic culture. That is, the sensitivity to sentiment is higher in individualistic cultures than collectivistic cultures. This leads to the following hypothesis:
H2: The impact of investor sentiment on future returns of hard to value and difficult to arbitrage stocks is more sensitive in collectivistic countries than individualistic countries.
3. Data and Methodology
3.1 Returns and characteristics
For the Netherlands, the sample includes all common stocks of all listed and formerly listed firms from the Netherlands from 2000 through 2013. For Portugal, the sample includes all common stocks of all listed and formerly listed firms from 2003 through 2013. The list of firms is collected from the Orbis database. The firm-level accounting and stock price data are from the Datastream database of Thomson Reuters. Following Fama and French (1992) accounting data are matched for fiscal year-ends in calendar year t-1 to monthly returns from calendar year t. Both countries are part of the pan-European stock exchange group Euronext, so both stock market are subject to the same regulation.
In this thesis, I follow Baker and Wurgler (2006) and use nine characteristics that appear to have a higher sensitivity to sentiment. These characteristics are small, young, unprofitable, non-dividend-paying, distressed, extreme stocks and stocks of firms that have a low level of tangible assets. These characteristics are defined by the following variables: size (ME), age (Age), earnings-to-book ratio (E+/BE), dividend-to-earnings-to-book ratio (D/BE), property, plant, and equipment over assets (PPE/A), book-to-market ratio (BE/ME), external finance over assets (EF/A), and sales growth (GS). I classify these nine variables into four groups. The first group ‘Size and Age’ is formed by ME and
Age, the second group ‘Profitability and Dividend Policy’ is formed by E/BE and D/BE, the third
The monthly returns of the stocks are calculated as: 𝑅𝑡𝑅−𝑅𝑡−1
𝑡−1 . For the first group is size defined
as market equity, ME, from calendar year t, measured as the share price multiplied by the number of ordinary shares in issue. ME is matched to monthly returns from calendar year t. For Age, I use the year of incorporation from Orbis, with the year 2014 as a reference point. Thus, a firm established in 1955 has an age of (2014-1955) 59. When the firm is delisted during the sample period, the firm is excluded from the year of delisting on. For firms that are established after 2000 (2003) for the Netherlands (Portugal), the year of incorporation from Orbis is matched to the first appearance of the firm in Datastream where 2014 is used as reference point as well.
Profitability characteristics include E+/BE and the dummy variable E>0. The earnings-to-book ratio is defined for firms with positive earnings. Earnings is net income before extraordinary items and preferred and common dividends (item WC01551) and book equity is the book value per share (item WC05476) times outstanding common shares (item WC05301). The profitability dummy variable E>0 takes the value one for profitable firms and zero for unprofitable firms.
Dividend characteristics include D/BE and the dividend payer dummy variable D>0. Dividends to equity, D/BE, is dividends per share ex-date (item WC 05101) times outstanding common shares, divided by book equity as is described above. The dividend payer dummy takes the value one for firms with positive dividends per share by the ex-date and zero for no dividends per share by the ex-date.
In the paper of Baker and Wurgler (2006) it is suggested that asset tangibility can also proxy for the difficulty to value stocks. The researchers are the first to include asset intangibility of firms in the measurement of sentiment sensitivity stocks and find that firms with a low level of tangible assets are more prone to sentiment. I follow this paper and include the variable net property, plant and equipment (item WC02501) over assets (PPE/A) as an indicator for the level of tangibility in a firm.
3.2. Measuring investor sentiment
There is no consensus among researchers on how to measure sentiment. This is because sentiment is not directly observable, so there are different possibilities to measure sentiment. Prior literature suggests three different approaches to measure sentiment. First, a survey about investors’ responses to stock market movements and the aggregate economy can be used, such as the consumer confidence index of Directorate Generale for Economic and Financial Affairs (see, e.g., Fisher and Statman, 2000; Lemmon and Portniaguina, 2006). Second, market related sentiment proxies can be derived from selected market statistics (see, e.g., Brown and Cliff, 2004; Baker, Wurgler, and Yuan, 2012). At last, a combination of the two methods can be used (, 2009).
In this thesis I follow the sentiment index that is presented by Baker and Wurgler (2006). Many other studies used this sentiment index and argue that it is an adequate measurement of investor sentiment (see, e.g., Corredor, Ferrer, and Santamaria, 2013; Baker, Wurgler, and Yuan, 2012; Chang, Faff, and Hwang, 2012). I use a principal component analysis of three suggested proxies in order to identify periods of high and low sentiment. To ensure that the periods of sentiment found by the sentiment-index are correct, I orthogonalize for macroeconomic variables to avoid the influence of business cycle movements on the returns of stocks. The three proxies are the number of IPOs per year (NIPO), share turnover (TURN) and the dividend premium (P). The proxies are measured annually from 2000 till 2013 for the Netherlands and for Portugal from 2003 till 2013. For every year the index has a score of sentiment. A year is identified as a period of high sentiment when the annual score of sentiment is below average. When the score of sentiment is above average, it is a year of high sentiment.
The first proxy is the number of IPOs per year (NIPO). The number of IPOs each year are retrieved from the Orbis database and are matched with its first appearance in Datastream. Firms that have returns and IPO dates that do not correspond with each other are excluded from the sample. The returns are equally-weighted across firms. The rationale is that when there is a period of high sentiment, investors are optimistic and willing to overpay for stocks of IPO firms. Due to this higher public valuation of IPO firms by investors, it is more attractive for firms to go public when sentiment is high (Ritter and Welch, 2002). The simultaneous increase in investor sentiment and IPOs makes
NIPO an adequate proxy for investor sentiment.
market and a higher share turnover. This optimism of investors and its impact can be seen as an indicator of sentiment. The ratio of turnover is the difference between the natural log ratio at the particular year and its natural log average of the previous 5-year ratios, in order to find if there is a trend in turnover. For the Netherlands, the ratio displays a steep decline in 2001 and subsequently a positive trend from 2002 onwards. This can be caused by severe losses in the telecommunication and transportation industry which led to an average decrease of the value of stocks of 21.1%. Up until then, the year 2002 was the second-worst year in the Dutch stock market postwar history2. For Portugal, the ratio displays a decline in turnover, with a steep fall in 2007. A possible explanation for this can be the low growth of GDP per capita which was at its lowest in 2006 with 1.3%, which displays the poor performance of the Portuguese economy which was characterized by the Economist as ‘a new sick man of Europe’ 3
.
The third proxy is the dividend premium (P), which is the log difference between the average market-to-book ratio of dividend payers and non-dividend payers, measured for each year. According to Baker and Wurgler (2004a) the payment of dividends of a firm is related to the growth opportunities of the firm. If a firm pays out dividends, it is characterized as a firm with slower growth aspirations. When a firm does not pay out dividend it is seen as an (extreme) growth stock. In periods of low sentiment, investors are looking for more ‘safe’ investments and will therefore have a higher demand for dividend-paying stocks. Subsequently, in periods of high sentiment, the demand for growth stocks will be higher, so there will be an increase in demand for non-dividend-paying stocks (Baker and Wurlger, 2004b). Thus, an increase in demand for non-dividend-paying stocks increases with investor sentiment and therefore the dividend premium can be used as a proxy.
To form an index of these proxies I use a principal component analysis in order to separate the common sentiment factor from non-sentiment related factors. Previous studies4 support that the first common variation of the principal component analysis captures sentiment in an adequate fashion. In order to avoid the possibility that the sentiment index is influenced by rational sentiment, I first orthogonalize the index for macroeconomic factors as is conducted in studies such as Baker and Wurgler (2006), and Lemmon and Portniaguina (2006). Rational sentiment results from changes in the business cycle that rationally explain a change in investor sentiment. Cutler, Poterba, and Summers (1991) find that one-third of the variance in stock returns can be explained by these macroeconomic factors. The goal of this thesis is to construct an index that represents irrational sentiment, for example the presence of over-optimism in the market by investors that has no logical explanation.
2http://www.cbs.nl/NR/rdonlyres/8C75B4F8-2186-4066-BF11-DA2EC4F440BB/0/pb02n001.pdf 3 http://www.economist.com/node/9009032
4
Table 1
Descriptive Statistics
The descriptive statistics are given for the Netherlands between 2000-2013 and Portugal between 2003-2013. Panel A describes the return variables, which are measured monthly. Panel B summarizes the size and age characteristics. Size is defined as market equity (ME), which is the share price multiplied by the number of ordinary shares in issue of calendar year t (item WC08001). Age is the amount of years since incorporation with the year 2014 as a reference point. Panel C gives the descriptives for the profitability characteristic. The earnings-to-book equity ratio (E+/BE) is all positive net income before extraordinary items and preferred and common dividends (WC01551) divided by the book value per share (item WC05476) times the number outstanding common shares (item WC05301). The profitability dummy (E) is one when firms have positive earnings. Panel D reports the dividend characteristics. Dividends to equity (D/BE) are the dividends per share ex-date (item WC05101) times outstanding common shares, divided by book equity. The dividend payer dummy (D) is one when the firm pays dividend per share. Panel E displays the tangibility characteristics. Plant, property and equipment (PPE/A) (item WC02501) and research and development (RD/A) (item WC02999) are both divided by total assets. For Portugal, RD/A is not included, because of insufficient availability of data. Panel F reports the growth/distressed stock characteristics. The book-to-market equity ratio (BE/ME) is the book value per share multiplied by the number of outstanding common shares divided by the market equity. External finance (EF) is the change in total assets minus the change in retained earnings (item WC03495) divided by total assets. Growth of sales (GS) is the change in net sales divided by prior-year net sales (item WC01001). Panels C through F use accounting data from the calendar year ending in t-1 and are matched with monthly returns from calendar year t.
The Netherlands Portugal
Characteristic N Mean Standard
Deviation Minimum Maximum N Mean
Standard
Deviation Minimum Maximum
A. Returns
Rt 35,447 0.84 49.47 -100 6900 11,826 1.06 23.04 -97.05 1,202.84
B. Size and Age
MEt-1 (€Mln) 1,476 1717 6,582 0.01 73,270 974 759 1,854 0.03 15,540
Aget (Years) 137 57.82 53.36 8 331 83 53 38.86 6 222
C. Profitability and Dividend Policy
E+/BEt-1 (%) 1,908 82.82 2,155.25 0.00 82,695.19 707 19.49 41.07 0.01 553.80 E>0t-1 1,905 0.78 0.41 0.00 1 707 0.50 0.50 0 1 D/BEt-1 (%) 1,476 38.40 378.62 0.00 12000 630 4.04 12.77 0 213.75 D>0 1,476 0.63 0.48 0.00 1 630 0.47 0.50 0 1 E. Tangibility PPE/At-1 (%) 1,962 26.03 26.40 0.00 99.52 631 30.27 20.83 0.01 84.42
F. Growth Opportunities and Distress
BE/MEt-1 533 1.57 7.17 -0.67 121.18 584 0.42 4.13 -54.49 32.31
EF/At-1 (%) 1,263 -0.67 165.32 -5,270.02 1,330.28 574 42.08 413.46 -5,449.04 3,230.56
In order to identify irrational investor sentiment, the three raw proxies (NIPO, TURN, and P) are regressed on the industrial production index, growth in consumer durables, nondurables, and services of each country, which are conducted from the Organisation for Economic Co-Operation and Development database. Furthermore, I include a dummy variable for times of recession for both countries according to the Federal Reserve Bank of St. Louis5. The residuals of the regressions are used as ‘new’ proxies for sentiment. Except for the three orthogonalized proxies, I include their one-year lags as well. The reason behind this is that there is a possibility that a period of high sentiment not immediately leads to an increase of IPOs, but that instead a year later the number of IPOs increases.
The orthogonalized proxies and its lags are used in the principal component analysis to identify the common variation that reflects sentiment. For each proxy, the highest correlation with the first principal component is used, so the direct proxy or its lag. Subsequently, a second principal component analysis is executed with only the three proxies with the highest correlations included, so that the total variance of the three proxies corresponds with the coefficients. For Portugal, instead of the lagged, the direct TURN variable is used because after conducting the second-stage principal component analysis, the sign of the correlation with the common factor of NIPO is not in line with the relation of the first-stage principal component analysis anymore. The correlation of the direct and lagged TURN proxy with the first principal component is respectively -0.404 and -0.445. The impact of the lead proxy is therefore approximately equal to the lagged proxy. The following two sentiment indices, for the Netherlands and Portugal, are formed:
SENTIMENTNEDt= 0.500NIPOt-1 - 0.624 Pt + 0.600 TURNt (1)
SENTIMENTPORt= 0.269NIPOt-1 - 0.626 Pt-1 + 0.732TURNt (2)
The first principal component has an explanatory power of 75% for the Netherlands and 47% for Portugal and the eigenvalues of both common factors are above 1.00. Thus, sentiment can be adequately defined by these proxies. A period of high (low) sentiment is defined as the years that have a score that is higher (lower) than average. For the Netherlands, the years 2000-2002, 2007, and 2012 are periods of high sentiment and the years 2003-2006, 2008-2011, and 2012 are years of low sentiment. For Portugal, in the period 2003-2007 there high sentiment and 2008-2013 there is low sentiment. To compare the difference in sentiment sensitivity between the Netherland and Portugal, I follow the same procedure as Baker, Wurgler, and Yuan (2012). I do a principal component analysis of the national sentimentindices of the Netherlands and Portugal for the period 2003-2013 and use the first common factor as an indiciation of common sentiment. This leads to the following joint sentiment index:
Table 2
Investor Sentiment Index
The results for the sentiment index for the Netherlands are provided in section A for the years 2000-2013 and for Portugal in section B for the years 2003-2013. The mean, standard deviation, minimum and maximum values of the three sentiment proxies are reported. The first proxy is the number of initial public offerings per year (NIPO). The second proxy is the dividend premium (P) which is the log difference between the average value-weighted market-to-book ratios of payers and non-payers. The third proxy is turnover (TURN) and is the difference between the natural log ratio at the particular year and its natural log average of the previous 5-year ratios. For the Netherlands, all three proxies are lagged for one year for the SENTIMENTNED index, for Portugal only the dividend premium is lagged for the SENTIMENTPOR index. The orthogonalized index, SENTIMENT˫NED and SENTIMENTPOR regresses the three raw proxies on the industrial production growth, durable, nondurable and services consumption growth and a dummy variable for recessions defined by OECD. The orthogonalizes proxies are the residuals from these regressions.
SENTIMENTJOINTt= 0.707SENTIMENT PORT
t - 0.707SENTIMENT NED
t (7)
The first principal component explains 51% and the eigenvalue is above 1.00. The weights show an exact opposite relation between the joint sentiment indices and the individual sentimentindices. Periods of Dutch sentiment are negatively related to the joint sentiment, while the Portuguese sentiment has a positive relation with joint sentiment. Furthermore, the correlation between the Dutch and Portuguese sentimentindex is -0.016. So, sentiment in the Netherlands is barley correlated with sentiment in Portugal. In the years 2003-2006, 2008, and 2010-2011 there is high joint investor sentiment and in 2007, 2009, and 2012-2013 there is low sentiment. Table 2 summarizes the results of all principal component analyses.
Correlation with sentiment Correlation with components of sentiment
The
Netherlands Mean SD Min Max SENTIMENT
NED
t NIPOt-1 Pt TURNt-1
NIPOt-1 5.4 7.19 0 27 NIPOt-1 0.50 1.00
Pt 1.46 1.06 -1.74 2.13 Pt -0.60 -0.56 1.00
TURNt 0.51 1.67 -0.64 4.71 TURNt 0.57 0.46 -0.84 1.00
Portugal Mean SD Min Max SENTIMENTPORt NIPOt-1 Pt-1 TURNt
NIPOt-1 1.5 0.85 0 3 NIPOt-1 -0.04 1.00
Pt-1 0.77 2.01 -1.95 4.31 Pt-1 0.56 0.08 1.00
TURNt 0.16 0.34 -0.22 0.80 TURNt -0.40 0.22 -0.37 1.00
SENTIMENTJOINTt SENTIMENTNEDt SENTIMENTPORt
SENTIMENTNEDt -0.02 1.00
3.3 Methodology
3.3.1 Cross-section of stock returns
At first, I use an univariate comparison. I place each firm into a group based on the value of its firm characteristic. I make a distinction between three types of groups each year, the upper 30%, the middle 40% and the bottom 30%, where returns of firms are equally-weighted within each group. I choose this ranking, because the sample size for both countries is not large enough to form deciles as in Baker and Wurgler (2006). For the profitability and dividend-paying characteristics (E/BE and
D/BE) I make a distinction between stocks of firms that are unprofitable or pay no dividends and
stocks that have positive earnings or pay dividends. For each group I calulate the post ranking average monthly return. Cross-sectional differences between upper, middle, and bottom values of characteristics can now be distinguished. Additionally, I make a distinction between periods of high sentiment and low sentiment by using the previous calendar year-end sentiment as indicated by the sentiment index described above in order to identify time-series changes within the formed groups.
To test if the impact of investor sentiment is significantly conditioned on the characteristics of a firm, I use a time-series regression method. For this method I form groups based on firm characteristics in the same way as the univariate comparison method, but now I form long-short portfolios. I do this by subtracting the average monthly stock returns of the bottom group from the average monthly returns of the upper group. For example, the average monthly returns from the group with the youngest firms (bottom group) are subtracted from the average monthly returns from the group with the oldest firms (upper group). For each variable a portfolio is formed. Subsequently, I run an Ordinary Least Square (OLS) to regress sentiment on monthly returns of the various long-short portfolios to find a significant effect of sentiment on the returns of the conditional characteristics. The dependent variable is the average monthly return of the long-short portfolio and the main independent variable is the sentiment index per year. This leads to the following equations for the Netherlands and Portugal:
RNEDXit =High,t – R
NED
Xit = Low,t = c + φSENTIMENT
NED
t-1 + uit (3)
RPORXit =High,t – RPORXit = Low,t = c + φSENTIMENTPORt-1 + uit (4)
RNEDXit =High,t – R
NED
Xit = Low,t = c + φSENTIMENT
NED
t-1 + βRMRFt + γSMBt + ηHMLt
+ αUMDt + uit (5)
RPORXit =High,t – R
POR
Xit = Low,t = c + φSENTIMENT
POR
t-1 + βRMRFt + γSMBt + ηHMLt
+ αUMDt + uit (6)
The first factor is the value-weighted market excess return over the risk-free rate (RMRF). The monthly risk-free rate is conducted from the Organisation for Economic Cooperation and Development database. I subtracted this monthly risk-free rate from the monthly value-weighted stock return to obtain the excess return per month. The second factor is the difference in monthly returns between small and large stocks (SMB). The same groups of the univariate and regression method for the size characteristic are used for this factor. However, for this factor, the monthly returns from the largest stocks are subtracted from the monthly returns of the smallest stocks. The third factor is the difference in monthly returns between high and low book-to-market stocks (HML). The same groups of the univariate and regression method for the book-to-market ratio (BE/ME) are used for this factor as well. The difference between high and low book-to-market stocks is similar to the long-short portfolio formed for the regression method regarding the BE/ME characteristic. The fourth factor is the difference in monthly returns between high and low momentum stocks. Momentum is calculated as the cumulative raw return for the 11-month period from 12 through 2 months prior to the observation return. Following the univariate and regression method, I form three groups and use the highest and lowest group. I subtract the monthly returns of the low-momentum stocks from the high momentum stocks to obtain the difference in monthly stock return.
To limit the possibility that the results are driven by outliers, I winsorize the returns of my sample by eliminating 0.5% of the most extreme return observations on both sides of the distribution. Furthermore, I use the Newey-West procedure to correct for autocorrelation and heteroscedasticity.
3.3.2 Cross-cultural comparisons
According to the investor sentiment index of Hofstede (2001), the Netherlands has an individualistic culture with a score on individualism of 81/100. Compared with other large countries in Western and Southern Europe6, it is the most individualistic country. With regards to collectivism, only Portugal and Spain are characterized as a collectivistic culture within Western and Southern Europe. Moreover, they are the only collectivistic cultures within Europe. Portugal has the lowest
6 The United Nations Statistics Division considers Austria, Belgium ,France, Germany, Liechtenstein,
score on individualism of 27/100. Therefore, I use the Netherlands and Portugal in this thesis to compare the difference in sentiment sensitivity of stock returns. Fig. 2 depicts the individualism index for the three largest countries of both Western and Southern Europe.
Figure 2. Individualism index. The bars represent the score on individualism for each country, based on the individualism index of Hofstede7 and is defined as the interdependence of a society among its members. The maximum score is 100 and the minimum score 0.
In order to measure if there is a significant difference in the sensitivity to sentiment between the Netherlands and Portugal, I run two different tests. First, I do an independent samples t-test, where I use the same groups as for the comparison within the countries as is explained above. Thus, a distinction is made between upper, middle, and bottom groups, as well as between periods of high and low sentiment. The number that represents this ‘difference of the difference’ fluctuation in future stock returns is used for the t-test. I measure if the mean future stock return fluctuation of the Netherlands significantly differs from the mean future stock return fluctuation of Portugal by the independent samples t-test. I do an F-test in order to decide if the variances of the samples are equal or unequal and I winsorize future returns by deleting 0.5% of the most extreme return observations in order to eliminate the possibility that results are driven by outliers.
sentiment between the Netherlands and Portugal. The dependent variable is the average monthly return of the long-short portfolio and the main independent variable is the sentiment index per year. This leads to the following equation:
(RNEDXit =High,t – R
NED
Xit = Low,t) – (R
POR
Xit =High,t - R
POR
Xit = Low,t) = c + φSENTIMENT
JOINT
t-1 + uit
(4)
To limit the possibility that the results are driven by outliers, I winsorize the returns of my sample by eliminating 0.5% of the most extreme return observations on both sides of the distribution. Furthermore, I use the Newey-West procedure to correct for autocorrelation and heteroscedasticity. The results of the regressions are reported in Section B of Table 4.
4. Results
4.1 Cross-section of stock returns
4.1.1 Univariate comparisons
Section A of Table 3 presents the results of the univariate comparisons for the Netherlands and Portugal. The first characteristic, size (ME), is not supported by the results for the Netherlands. This can be derived from the difference in future returns between upper and bottom groups and the difference in future returns between high and low periods of sentiment. The future return of small stocks is 0.54% lower after a period of high sentiment than after a period of low sentiment. For big stocks are the future returns after a period of high sentiment 1.40% lower than after a period of low sentiment. However, the age (Age) characteristic does support the prediction that young stocks are more prone to sentiment. The future returns of young stocks have a fluctuation of 2.05% between periods of sentiment, while middle-aged stocks have a fluctuation of 1.39% and old stocks have only a 0.96% difference in future returns. Theory also holds for unprofitable (E) and non-dividendpaying (D) stocks, as well as stocks from firms with a low level of tangible assets (PPE/A). The fluctuation of these stocks are all above 2%. The last two characteristics, growth and distress, are both measured by the same three variables, the book-to-market ratio (BE/ME), external finance over assets (EF/A), and growth of sales (GS). The higher sensitivity to sentiment of growth stocks is confirmed by the BE/ME and GS variable, while the sensitivity of distressed stocks is only supported by the GS variable. Thus, the support is somewhat more fragile for the distressed characteristic.
periods of over 4%, which is a strong support of a higher sensitivity of these stocks. However, young, unprofitable, and non-dividend-paying stocks do not find support by the results of Portugal. As regards with the age characteristic, the younger firms have a higher difference in future returns than the oldest stocks, but a lower fluctuation than middle-aged stocks. Thus, the expectation that young stocks are more sensitive to sentiment cannot be met. The impact of sentiment on growth stocks is supported by both book-to-market ratio and external finance variables. The sensitivity to sentiment of distressed stocks is strongly supported for Portugal, because all three variables used to measure this characteristic have a high fluctuation of 3% in future returns between sentiment periods.
In summary, both countries support the prediction that the future returns of growth and distressed stocks as well as stocks of firms with a low level of tangible assets are sensitive to sentiment. For the Netherlands, there is also supporting evidence for unprofitable, non-dividend-paying, and young stocks. Thus, only the sensitivity of small stocks to sentiment is not supported. For Portugal, the results also confirm the sensitivity to sentiment of small stocks. The lack of evidence for the sensitivity to sentiment of small stocks for the Netherlands and young, non-dividend-paying, and unprofitable stocks for Portugal can be explained by the difference in trading behavior between individual and institutional investors. The differences between individual and institutional investors lie in the time they spend on analysis, their sensitivity to attention-based trading, and their accessibility to different information sources. Individual investors spend less time on their invesment decisions, as well as on the analysis and have different information sources, which makes them more prone to investor sentiment (Kumar and Lee, 2006). It is possible that there is a higher presence of institutional investors in Portugal and therefore the impact of sentiment on the future returns on stocks is lower.
4.1.2 Time-series regressions
positive. Panel D and E report the results for the growth and distressed stocks by the three variables book-to-market value, external finance, and growth of sales. The influence of sentiment on the future returns of growth stocks is confirmed by the results in panel D. Both coefficients of sentiment have results that are significant and positive for BE/ME. After controlling for RMRF, SMB, HML and
UMD, a one-unit increase in sentiment is associated with a 0.44% increase in future returns of the
middle-short porfolio BE/ME. So, when sentiment is high, the future returns of growth stocks are low. The significant negative sentiment coefficient of the GS variable confirms the finding of the BE/ME variable. After controlling for co-movement, a one-unit increase in sentiment leads to a 0.27% decrease in the future returns of a long-short growth stock portfolio. The distressed characteristic is significantly supported as well by the EF/A variable at a 10%-significance level. The sentiment coefficient of the GS variable is also in line with prior literature. However, the BE/ME variable does not support this finding.
For Portugal, the majority of the relations between sentiment and future stock returns is not in line with theory. Only the future returns of small, (partly) growth, and (partly) distressed stocks have a negative relation with sentiment. The negative impact of sentiment on young stocks is also significantly supported. The rest of the future returns of stocks appear to have a positive relation with sentiment. This positive impact of sentiment is significant for non-dividend-paying stocks, so a one-unit increase in sentiment leads to a future return decrease of 1.08% on long-short dividend-paying portfolios. This positive relation between sentiment and growth and distressed stocks is supported by all three variables. Thus, the book-to-market ratio, external finance over assets, and growth of sales. The feature of hard to value and difficult to arbitrage stocks is that investors rely on their own personal belief in the future performance of the stocks when valuating stocks rather than to apply conventional valuation models. It could be that despite the general wave of pessimism during a period of low sentiment, Portuguese investors are more rational in their investment decisions. They do not follow each other in a general view of pessimism, but base their investment decisions on a more solid, rational foundation. The rational decision in a period of low sentiment would be to invest in sentiment sensitive stocks, because the future returns of stocks will be lower after a period of low sentiment than after a period of high sentiment. Subsequently, in a period of high sentiment, Portuguese investors behave more rationally and do not blindly follow each other and invest in more risky stocks. Alternatively, which is already explained above, this positive relation could also be the consequence of a higher presence of institutional investors in the Portuguese stock market. Prior literature suggests that institutional investors are more rational and less prone to sentiment than individual investors (Kumar and Lee, 2006). Thus, more institutional investors on the Portuguese stock market could explain this reverse sensitivity to sentiment.
argues that non-dividend-paying rather than dividend-paying stocks are more sensitive to sentiment. The sentiment coefficients of stocks from small, unprofitable, non-dividend-paying firms and firms with a low level of tangible assets are insignificant but in line with theory for the Netherlands. The sentiment coefficients of small, (partly) growth, and (partly) distressed are in line with the theory for Portugal. The two factors of Fama and French, the market risk premium and momentum appear to have a substantial influence on the predictability of stock returns. This indicates that there are other factors involved that have an influence on future stock returns as well, besides sentiment. This should be taken into consideration when interpreting the results. The observations of these coefficients are insightful, but not inconclusive because of the majority of insignficant results and the influence of the four co-movement factors on the future stock returns. As a small remark, I would like to point out that the results of the regressions of sentiment on long-short stock return portfolios of Baker and Wurgler (2006) are partly insignificant as well.
Overall, the observations of the univariate comparison are significantly supported by the results of the time-series regressions for young, growth, and distressed stocks for the Netherlands. For Portugal, only the higher sensitivity of dividend-paying stocks observed by the univariate comparison is significantly supported. This indicates that for the Netherlands, there is evidence that stocks that are hard to value and difficult to arbitrage are more sensitive to sentiment. However, the majority of the variables used to measure the stock characteristics, do not support the prediction of a higher sensitivity to investor sentiment. Therefore, I reject the first hypothesis that states that the impact of investor sentiment has an impact on future returns of hard to value and difficult to arbitrage stocks for the Netherlands. For Portugal, the observations of the univariate comparison are not significantly supported by the time-series regression, apart from the dividend-paying stocks. However, this is not in line with the prediction that non-dividend-paying stocks are more prone to investor sentiment. Therefore, I reject the first hypothesis that states that the impact of investor sentiment has an impact on future returns of hard to value and difficult to arbitrage stocks for Portugal.
4.2 Cross-country comparison
In general, the outcomes of the independent samples t-test indicate that there is no significant difference in sensitivity to sentiment between the Netherlands and Portugal. Only for the growth characteristic, measured by the BE/ME variable, there is a significant difference between the two countries. The results are reported in Section B in Table 3.
To examine the relation between sentiment and future stock returns between the two countries more precisely, I form a long-short portfolio that has a long position in Dutch stocks and a short positions in Portuguese stocks. The sentiment coefficient indicates if the return on the portfolios is profitable when sentiment is high. Theory argues that because of a high level of individualism in the Netherlands, the impact of sentiment is lower. That is, the future stock returns in the Netherlands are lower than the Portuguese stocks after a period of high sentiment. The results of the regressions show no significant relation between (joint) sentiment and the long-short portfolio. The results are reported in Section B in Table 4.
Overall, the results of the independent samples t-test and the time-series regressions do not find a significant difference in the sensitivity to sentiment between the Netherlands, an individualistic country, and Portugal, a collectivistic country. Therefore, I reject the second hypothesis.
However, the results of the time-series regression show that the sentiment coefficient is mainly negative for the variables, which indicates that Portuguese stocks are less sensitive to sentiment than Dutch stocks. Thus, the long-short portfolio is for the most variables not profitable. A possible cultural explanation for the higher effect of sentiment in the Netherlands than Portugal could be that the Dutch society is more willing to take risks and has a more optimistic view in general than the Portuguese society. The Hofstede Centre8 finds that the Portuguese have a very high preference for uncertainty avoidance and that they have a tendency to pessimism, compared with the Netherlands. Fig. 3 and 4 display the different scores on uncertainty avoidance and indulgence for the same countries as used for the individualism score. Indulgence is defined as the extent to which people try to control their desires and impulses. A low score on indulgence means that people are more reluctant to fulfill their desires. One of the features of this indulgent society is that people have a tendency towards pessimism. This avoidance of uncertainty and general view of pessimism can be linked to the investment behavior of Portuguese investors. Despite a general optimistic view during a period of high sentiment, the investors could still be reluctant to invest in more risky stocks, such as young or non-dividend-paying stocks. For example, young stocks do not have a substantial earnings history yet, so the decision to invest in these stocks is more difficult because there is less information on past firm performance available than older firms. Thus, the investor is better able to estimate the risks involved in investing in an aged stock than in a young stock. Another example is the preference for dividend-paying stocks, because a certain payment is guaranteed, while for a non-dividend-paying stocks there is a possibility that there is no capital gain involved.
8
Figure 3. Uncertainty Avoidance Score. The bars represent the score on individualism of each country, based on research of the Hofstede Centre9 and is defined as the extent to which the members of a culture feel threatened by ambiguous or unknown situations and have developed beliefs and institutions that avoid these situations. The maximum score is 100 and the minimum score 0.
Figure 4. Indulgence Score. The bars represent the score on indulgence of each country, based on research of the Hofstede Centre9 and is defined as the extent to which people try to control their desires and impulses. The maximum score is 100 and the minimum score 0.
Table 3
Future Returns by Firm Characteristics and Sentiment Periods
For each year, two or three equally-weighted portfolios are formed, based on the value of the firm characteristic for both the Netherlands and Portugal. The characteristics are firm size (ME), age (Age), earnings-to-book ratio (E/BE), dividend-book ratio (D/BE), fixed assets (PPE/A), book-to-market ratio (BE/ME), external finance over assets (EF/A), and sales growth (GS). A firm can be characteristized as bottom (lowest 30%), middle (middle 40%) or upper (highest 30%). For the E/BE and D/BE variable, the bottom groups are the future returns of unprofitable/non-dividend-paying stocks and the upper groups are the profitable/unprofitable/non-dividend-paying stocks. Average future monthly returns of periods of high sentiment are seperated from periods of low sentiment per portfolio. The periods of sentiment are classified by the sentiment index, which is SENTIMENTNED for the Netherlands and SENTIMENTPOR for Portugal. Periods of high sentiment are for the Netherlands 2000-2002, 2007, and 2012 and for Portugal 2003-2007. The t-statistic of the independent samples t-test of the average fluctuation in future returns between sentiment periods and upper and bottom groups between the Netherlands and Portugal are reported in the last colomn. Ꞌ indicates if the sensitivity to sentiment between upper andbottom groups over periods of sentiment is in line with theory and ***, **, * represent the two-sided statistical significance of the statistics at a 1%, 5% and 10% level.
Section A
The Netherlands Portugal
Section B
Cross-country comparison
Groups Comparisons Groups Comparisons The Netherlands - Portugal
Characteristic Sentiment Bottom Medium Upper
Upper – Bottom Upper – Middle Middle – Low
Bottom Medium Upper
Upper – Bottom Upper – Middle Middle – Low Upper - Bottom ME High 0.20 -0.25 -0.71 -0.91 1.08 -0.05 -0.20 -1.67 Low 0.74 1.15 0.70 -0.04 5.15 0.77 1.44 -4.03 0.70 Difference -0.54 -1.40 -1.41 -0.87 4.07Ꞌ 0.82 1.64 -2.44 Age High -0.98 -0.34 -0.03 0.94 -0.65 0.64 0.86 1.51 Low 1.07 1.05 0.93 -0.14 1.45 3.49 2.69 1.23 0.87 Difference -2.05Ꞌ -1.39 -0.96 1.08 2.10 2.85 1.82 -0.28 E/BE High -1.20 -0.42 0.78 0.68 0.18 -0.50 Low 1.18 1.57 0.40 2.22 2.52 0.23 0.20 Difference -2.38Ꞌ -1.99 0.38 -1.54 -2.34 -0.80 D/BE High -0.88 -0.21 0.67 2.35 2.83 0.48 Low 1.42 1.55 0.13 0.41 0.47 0.06 0.08 Difference -2.30Ꞌ -1.75 0.55 -1.94 -2.35 -0.41 PPE/A High -0.81 -0.40 -0.42 0.39 -0.99 0.66 0.82 1.81 Low 1.61 1.45 1.45 -0.16 3.09 3.41 2.80 -0.29 1.24 Difference -2.42Ꞌ -1.85 -1.87 0.55 4.08Ꞌ 2.75 1.98 -2.10
Middle - Bottom Upper - Middle
Table 4
Time-Series Regressions of Portfolio Returns
Per characteristic long-short portfolios are created and the future average monthly returns are regressed on SENTIMENTNED for the Netherlands and SENTIMENTPOR for Portugal of the prior year. Additionally, a second regression on the average future monthly returns of the long-short portfolios together with two Fama and French factors (HML and SML), the market risk premium (RMRF) and a momentum factor (UMD) are executed. For the Netherlands the sample includes monthly returns from 2001 to 2013 and for Portugal from 2004 to 2013. The long-short portfolios are based on the following firm characteristics: firm size (ME), age (Age), profitability (E), dividends (D), tangibility (PPE), growth and distress (BE/ME, EF/A and GS). A firm can be characteristized as bottom (lowest 30%), middle (middle 40%) or upper (highest 30%). Average monthly returns are matched to the prior-year SENTIMENT. SENTIMENT is orthogonalized for industrial production growth, durable, nondurable and services consumption growth and periods of recession. The first and third set show the univariate regression results and the second and fourth set have RMRF, SMB, HML and UMD included as control variables. SMB (HML) is not included as control variable when it is a dependent variable. The fifth set displays the coefficient of the time-series regression of SENTIMENTJOINT on the monthly returns a of portfolio that has a long position in Dutch stocks and a short position in Portuguese stocks. SENTIMENTJOINT includes monthly returns from 2004-2013. The coefficient of the SENTIMENT variable is reported as φ and ***, **, * represent the statistical significance of the statistics at a 1%, 5% and 10% level.
Section A
SENTIMENTNEDt-1 SENTIMENT
NED t-1 Controlling for RMRF, SMB, HML, UMD
SENTIMENTPORt-1 SENTIMENT
POR t-1 Controlling for RMRF, SMB, HML, UMD Section B SENTIMENTJOINTt-1 φ φ φ φ φ
Panel A. Size and Age
ME SMB 0.018 -0.003 -1.197 -1.149* Ned - Por -0.121
Age Upper – Bottom 0.091 0.258** 0.767* -0.149 Ned - Por -0.031
Panel B. Profitability and Dividend Policy
E > 0 - < 0 0.065 0.096 -1.109 -1.307 Ned - Por -0.285
D > 0 - = 0 0.076 0.168 -0.238 -1.084* Ned - Por -0.010
Panel C. Tangibility
PPE/A Upper – Bottom -0.020 0.077 0.576 -0.018 Ned - Por -0.015
Panel D. Growth Opportunities
BE/ME Middle – Bottom 0.429*** 0.438*** 0.196 0.435 Ned - Por -0.002
EF/A Upper – Middle 0.019 0.051 0.238 0.061 Ned - Por -0.036
GS Upper – Middle -0.135** -0.268** 0.449 -0.228 Ned - Por 0.286
Panel E. Distress
BE/ME Upper – Middle 0.054 0.033 0.671 -0.267 Ned - Por 0.193
EF/A Middle – Bottom 0.017 0.121* 0.238 -0.043 Ned - Por 0.040
5. Conclusions
In this thesis, I contribute to the growing literature of behavioral finance, by testing the effect of sentiment on stock returns in the Netherlands, an individualistic country, and Portugal, a collectivistic country. The comparison of these two countries allows me to test the impact of individualism as a cultural dimension on the occurance of sentiment. The main finding of this thesis is that Dutch stocks are more prone to sentiment than Portuguese stocks. Therefore, I conclude that the level of individualism in a country does not explain the sensivitity to sentiment of future stock returns. I test two related hypotheses in this study. The first hypothesis posits that investor sentiment affects stocks with certain characteristics more than others. The second hypothesis postulates that countries with a higher level of individualism are less sensitive to sentiment than collectivistic countries. Evidence on the validity of these hypotheses are important for investors because its decision-making regarding the optimal portfolio allocation can be influenced by the predictability of stock returns. When investors are aware of the period of sentiment and they know which stocks are more sensitive to sentiment, they can anticipate on this possible profitability opportunity.
Prior to the testing of the hypotheses, I form a sentiment index for both countries by using the first principal component of three sentiment proxies similar to Baker and Wurgler (2006), to identify periods of low and high sentiment for both countries. In addition, I use the first principal component of the two national sentiment indices to construct a joint sentiment index.
I test the first hypothesis by identifying if small, young, high volatility, unprofitable, non-dividend-paying, extreme growth and distressed stocks are more sensitive to sentiment for both countries. Theory suggests that stocks with these characteristics are difficult to value and harder to arbitrage because there is a greater disagreement among investors how to valuate these stocks and therefore mispricing is more likely to occur. When sentiment is high, investors are optimistic about the performance of stocks and have the tendency to overvalue these stocks. The future returns should therefore be lower for sentiment sensitive stocks when sentiment is high, because of a higher demand by investors. I use an univariate comparison and run an OLS in order to test if the theory holds for Dutch and Portuguese stocks.
by the presence of more rational investors in the Portuguese stock market. These rational investors know that after a period of low sentiment, stock returns will be higher than after a period of high sentiment. Thus, when there is a general wave of optimism, the majority of the investors will not follow this sentiment and invest in stocks of which they expect that they will be more profitable in the (near) future. The impact of sentiment is not as powerful as is expected in Portugal.
In order to test the second hypothesis that a country with a high level of individualism is less sensitive to sentiment than a collectivistic country statistically, I do an independent samples t-test in order to test if there is a significant difference in the fluctuation of future stock returns between periods of sentiment between the two countries. Furthermore, I run a regression of the joint sentiment index on a portfolio which has a long position in Dutch stocks and a short position in Portuguese stocks. The assumption is that the future returns of Portuguese stocks are lower than Dutch stocks, because of a higher sensitivity to sentiment.
The results of the independent samples t-test do not indicate that there is a significant difference between the sensitivity to sentiment of Dutch and Portuguese stocks. The results of the time-series regressions are insignificant as well. Therefore, I reject the second hypothesis that collectivistic countries are more prone to sentiment than individualistic countries. However, the results of the regressions show evidence that Portuguese stocks are less sensitive to sentiment than Dutch stocks. The expectation that a lower level of individualism leads to a higher level of sensitivity is not supported. A possible cultural explanation for this result is that Portuguese investors have a higher level of uncertainty avoidance than the Dutch investors and that they have a more pessimistic view in general. Thus, despite a period of high sentiment, they are still reluctant to invest in risky stocks.
The choice of proxies to form the sentiment index are of a high importance. Despite prior literature that supports the sentiment index of Baker and Wurgler (2006), there is a change in the sentiment effect on future stock returns when correcting for excess returns, small-minus-big stock portfolios, high-minus-low book-to-market ratio, and momentum. In consideration of the sensitivity of the results to the choice of the sentiment proxies, this should be taken into consideration when interpreting the results. In order to obtain a more robust sentimentindex, a consumer confidence index can possibly be included (see, e.g., Schmeling, 2009). Furthermore, the sample period is rather small, only 9 years for Portugal and 13 years for the Netherlands, and both stock markets are relatively small compared with for example the US stock market. In addition, only two countries in thesis are used to compare.
6. References
Ackert, L., Deaves, R., 2010. Behavioral Finance: Psychology, Decision-Making and Markets. South-Western Cengage Learning, Mason.
Baker, M., & Wurgler, J., 2000. The equity share in new issues and aggregate stock returns. The Journal of Finance 55, 2219-2257.
Baker, M., Wurgler, J., 2004a. A catering theory of dividends. The Journal of Finance 59, 1125-1165.
Baker, M., Wurgler, J., 2004b. Appearing and disappearing dividends: The link to catering incentives. Journal of Financial Economics 73, 271-288.
Baker, M., Wurgler, J., 2006. Investor sentiment and the cross‐section of stock returns. The Journal of Finance 61, 1645-1680.
Baker, M., Wurgler, J., 2007. Investor sentiment in the stock market. Unpublished working paper. National Bureau of Economic Research, Cambridge.
Baker, M., Wurgler, J., Yuan, Y., 2012. Global, local, and contagious investor sentiment. Journal of Financial Economics 104, 272-287.
Baker, M., Stein, J. C., 2004. Market liquidity as a sentiment indicator. Journal of Financial Markets 7, 271-299.
Brown, G. W., Cliff, M. T., 2004. Investor sentiment and the near-term stock market. Journal of Empirical Finance 11, 1-27.
Chang, C., Faff, R. W., Hwang, C. Y., 2012. Local and global sentiment effects, and the role of legal, trading and information environments (Available at SRRN: http://ssrn.com/abstract=1787700 or http://dx.doi.org/10.2139/ssrn.1787700)
Corredor, P., Ferrer, E., Santamaria, R., 2013. Investor sentiment effect in stock markets: Stock characteristics or country-specific factors? International Review of Economics and Finance 27, 572-591.
Chui, A.C.W., Titman, S., Wei, K.C.J., 2010. Individualism and momentum around the world. The Journal of Finance 65, 361-392.
Cutler, D. M., Poterba, J. M., Summers, L. H. 1991. Speculative dynamics. The Review of Economic Studies 58, 529-546.
