INVESTOR SENTIMENT EFFECT ON STOCK RETURNS IN SCANDINAVIAN STOCK MARKET

Transcription

1 INVESTOR SENTIMENT EFFECT ON STOCK RETURNS IN SCANDINAVIAN STOCK MARKET Žana Grigaliūnienė 1, Diana Cibulskienė 2 1 Siauliai University, Lithuania, zana@smf.su.lt 2 Siauliai University, Lithuania, cibulskiene@yahoo.de Abstract We study an investor sentiment effect on stock returns at aggregate level and cross-sectionally in Scandinavian stock market. We employ consumer confidence index (CCI) and economic sentiment indicator (ESI) as measures of investor sentiment. Consistently with prior research we predict that high sentiment has a negative effect on future stock returns. This effect is more pronounced for hard-to-value and hard-to-arbitrage stock returns. To account for cross-sectional effects we use portfolio returns sorted on the basis of book-tomarket ratio and dividend yield. And to capture sentiment effect at aggregate level we employ Scandinavian index portfolio returns. We run predictive regressions of stock returns on consumer confidence at different forecast horizons. To prove close relations of consumer confidence to overall economic activity we employ some macroeconomic variables that are motivated by the asset pricing theory. Keywords: investor sentiment, stock returns, consumer confidence index, and economic sentiment indicator. Introduction Stock returns predictability models are mostly proposed by neoclassical asset pricing theory that employs financial ratios (dividend-price ratio, book-to-market ratio, earnings-price ratio, and etc.) to explain stock return variability, or considers macroeconomic variables having explanatory power in forecasting stock returns. Thus, the proponents of asset pricing theory attempt to explain market anomalies developing traditional economical settings, whereas their opponents suggest a behavioral approach for explaining market anomalies, emphasizing the role of investor sentiment in asset pricing. Although scholars use various direct and indirect investor sentiment measures, their common finding is that high sentiment has a negative effect on future stock returns and vice versa (Brown & Cliff, 2005; Schmeling, 2009). Baker & Wurgler, 2006, 2007 and Lemmon & Portniaguina, 2006 observe even more pronounced effect on firms that are hard to price and thus difficult to arbitrage (small-size, rapidly growing, young, distressed, unprofitable, nondividend paying firms). Building on prior literature and empirical evidence in the U.S. stock market, we aim to analyze sentiment effect on stock returns in Scandinavian stock market at aggregate level and cross-sectionally. The special attention is given to observing differences in the effect of sentiment on returns of stock portfolios that are sorted according to their speculative appeal and their difficulty of arbitrage. We use portfolios sorted on book-to-market and dividend yield to account for cross-sectional effects and Scandinavian index portfolio returns to capture sentiment effect on stock returns at aggregate level. As a proxy for investor sentiment we use consumer confidence index and economic sentiment indicator of four Scandinavian countries that include Norway, Finland, Sweden, and Denmark. To prove the close relations of consumer confidence and economic sentiment to overall economic activity, we employ some macroeconomic variables that are motivated by the asset pricing theory. We include consumer price index (CPI), change in inflation rate, 3- month T-Bill rate, change in industrial production and gross domestic product (GDP) growth as controls into predictive regressions. We find significant negative relationships between stock returns and investor sentiment at different forecast horizons. These results are robust to the inclusion of other variables that have got an explanatory power. We contribute to the strand of behavioral finance literature that documents empirical evidence of sentiment effect on stock returns in different markets. Our findings are relevant, as they enrich the behavioral literature on return prediction both at cross-sectional and aggregate level. The paper is structured as follows. At first, we review the prior literature and discuss theoretical predictions. Then, we provide the data description and explain the empirical approach. Then we present empirical tests and conclude the findings. 929

2 Prior Literature Review and Theoretical Predictions Traditional asset pricing theory does not treat investor sentiment seriously, but behavioral finance literature considers it as a significant factor that explains excess returns and conditional volatility shocks (Lee et al., 2002). The role of sentiment in stock markets was formally modeled by DeLong et al. (1990), Barberis et al., 1998; Daniel et al., 1998; and Hong & Stein, However, their models are difficult to test directly since they typically involve sources of sentiment, which are difficult to measure. Therefore, most of research papers focus on the empirical examination of the importance of sentiment rather than testing formal economic models. Sentiment literature gives special attention to sentiment measures, their validity and their predictive power. Brown & Cliff, 2004; Qui & Welch, 2006; and Baker & Wurgler, 2007 give a comprehensive review of direct and indirect sentiment measures widely used in a current body of research. Direct or survey-based sentiment measures include Investor's Intelligence survey (Brown and Cliff, 2004, 2005; Lee, Jiang, & Indro, 2002; Solt & Statman, 1988; Clarke & Statman, 1998) and the American Association of Individual Investors survey (Bandopadhyaya & Jones, 2006, 2008; Brown & Cliff, 2004; Clarke & Statman, 1998; Fisher & Statman, 2003; Baker & Stein, 2004; Baker & Wurgler 2006). However, the primary interest of the paper focuses on the sentiment measures that reflect the overall economic activity. Therefore, we employ Consumer confidence index (Bergman & Roychowdhury, 2008; Schmeling, 2009; Lemmon & Portniaguina, 2006; Qiu & Welch, 2006) and Economic sentiment indicator that are also attributed to direct sentiment measures. We do not use indirect sentiment measures as they imply potential information loss. A current body of literature focuses on the empirical examination of sentiment effect on stock returns (Brown & Cliff, 2004; He, Mian, & Sankaraguruswamy, 2007). Lee, Shleifer, & Thaler, 1991; Elton, Gruber, & Busse, 1998; Neal & Wheatley, 1998; Baker & Wurgler, 2006; Lemmon & Portniaguina, 2006 provide extensive evidence on a significant sentiment effect on cross-section of stock returns, but aggregate predictions remain less clear (Schmeling, 2009; Brown & Cliff, 2005 and etc.). Consistently with earlier empirical evidence we predict that optimistic/pessimistic sentiment implies subsequent lower/higher stock returns. This relation suggests a significantly negative estimate of β in the predictive regressions of stock returns on sentiment. We also predict that sentiment has stronger cross-sectional effects on portfolios that are hard to value and hard to arbitrage (e.g. growth vs. value, dividend paying vs. non-paying). However, empirical evidence is controversial. Glushkov, 2006 explores sentiment betas and finds evidence that hard-to-arbitrage and value stocks are more sensitive to sentiment as predicted. Baker & Wurgler, 2006; 2007 find evidence that shifts in sentiment affect returns on small stocks rather than large stocks. Brown & Cliff, 2005 argue that sentiment effect is largely concentrated in large-capitalization growth stocks. Lemmon & Portniaguina, 2006 find evidence that consumer confidence-based sentiment measure exhibits forecasting power for the returns on small stocks and no evidence that it forecast the time-series variation in returns to value and momentum strategies. Consistently with Lemmon & Portniaguina, 2006; Qiu & Welch, 2006 argue that changes in consumer confidence predict small stock returns. Contrary to conventional wisdom, Brown & Cliff, 2004 find no evidence that sentiment primarily affects small stocks. He, Mian, & Sankaraguruswamy, 2007 concludes that sentiment effect is stronger for stocks that prior literature documents to be prone to sentiment small, young, growth, and volatile firms. Data and Descriptive Statistics We use Københavns Fondsbørs Indeks, Helsinki Stock Exchange All-Share Index, Oslo Bors All- Share Index, OMX Stockholm 30 Index, and Scandinavian index portfolio returns common for Sweden, Finland, Norway and Denmark for the period between 1989 through We also employ Dow Jones EURO STOXX 50 to compare the results across Europe. To capture sentiment cross-sectional effects we use portfolio returns that are sorted on the basis of book-to-market ratio and dividend yield for Sweden, Finland, Norway and Denmark for the period between 1989 through 2009 from Ken French data library files. To proxy for sentiment we use Consumer Confidence Index (CCI) and Economic sentiment index (ESI). In order to find evidence that consumer confidence is related to overall economic activity, the research employs some macroeconomic variables that are motivated by asset pricing theory. Table 1 presents the descriptive statistics of consumer confidence index and economic sentiment indicator for Sweden, Finland, Norway and Denmark. 930

3 Table 1. Descriptive Statistics on Investor Sentiment in Scandinavian Countries, Consumer confidence index (CCI) Economic sentiment indicator (ESI) Consumer confidence index (CCI) Economic sentiment indicator (ESI) Consumer confidence index (CCI) Economic sentiment indicator (ESI) Consumer confidence index (CCI) Economic sentiment indicator (ESI) *probability (p-value) in parenthesis Panel A: Sweden Mean Standard deviation Min Max Jarque- Bera ( ) ( ) Panel B: Finland Mean Standard deviation Min Max Jarque- Bera ( ) ( ) Panel C: Norway Mean Standard deviation Min Max Jarque- Bera 18, , (0.09) N/A N/A N/A N/A N/A Panel D: Denmark Mean Standard deviation Min Max Jarque- Bera ( ) ( ) Table 1 includes the mean, the standard deviation, the maximum and minimum values and a Jarque- Bera statistic for four Scandinavian countries. Jarque-Bera statistic allows us to test the normality of sentiment raw data. Under the null hypothesis of a normal distribution the Jarque-Bera statistic is distributed 2 as χ with 2 degrees of freedom. The reported p-value (in parenthesis) is the probability that the Jarque-Bera statistic exceeds (in absolute value) the observed value under the null hypothesis a small probability value leads to the rejection of the null hypothesis of a normal distribution. Overall, we conclude non-normal distribution of raw sentiment data for all panel countries, but for Sweden CCI we cannot reject the null of normal distribution at any significance level. We prove Sweden ESI and Norway CCI to be normally distributed at 8 % and at 9% significance level respectively. Prior research documents a strong correlation of sentiment proxies and macro variables that implies their relationships with overall economic activity. Asset pricing theory describes macro variables as good predictors of stock returns. Therefore, the research focus on finding close relations between sentiment indices and macro variables in order to prove the ability of sentiment to predict stock returns. We employ macro variables that are motivated by asset pricing theory and earlier empirical research: Consumer price index (CPI) used by Lemmon & Portniaguina, 2006; Schmeling, 2009; and Brown & Cliff, 2005; Change in industrial production (IPI), analogous indicator is used by Schmeling, 2009; Gross domestic product (GDP) growth applied in Chen, Roll, & Ross, 1986; Lemmon & Portniaguina, 2006; Short term T-Bill rate: 1-Months, 3-Months, 6-Months or 12-Months used by Brown and Cliff, 2005 (we used 3-Months term T-Bill rate: T_BILL_3). We download macro indicators from official statistics websites of Sweden, Norway, Finland and Denmark between 1995 through Note that the proxies for sentiment in Scandinavian countries move quite close that can be illustrated by the following figure. 931

4 FI_CCI NO_CCI DK_CCI EA_CONS SE_CCI DK_ESI FI_ESI EA_ESI SE_ESI Figure 1. Graphical comparison of sentiment proxies in Scandinavian countries Figure 1 reveals almost no difference between movements of consumer confidence index and economic sentiment indicator across countries. We also compare local ESI with ESI for European area and see almost no differences in movements. We find the same for CCI across countries in European area. There is no data for Norway ESI and there is insufficient data for CCI (it is computed on a quarterly basis and begins in 2000), therefore we can use ESI and CCI for European area to proxy for sentiment in Norway. Empirical Tests At first, the use of raw consumer confidence index and economic sentiment indicator may raise some statistical problems, since it may have a unit root. From the variety of unit root tests KPSS test could be suggested. The reasoning is the following, Augmented Dickey-Fuller (ADF), Phillips Perron (PP) tests too often suggest unit roots even if there is no clear evidence for them (sensitivity too structural changes, etc.). So it is good to formulate the test, which is consistent under null stationary hypothesis versus a wide range of non-stationary reasons. A test with such properties is the KPSS test. Sentiment proxy Table 2. Unit-root and stationary hypothesis tests Country p(adf) with KPSS with intercept Conclusion For ADF* intercept test 1% critical 5% critical 10% critical H 0 : I(1); value value value H 1 : I(0) CCI Sweden I(0) I(1) Finland I(1) I(1) Norway I(1) I(0) Denmark I(1) I(0) ESI Sweden I(1) I(0) Finland I(1) I(0) Norway n/a n/a n/a n/a n/a n/a n/a Denmark I(0) I(0) D(CCI) First difference in CCI D(ESI) First difference in ESI Conclusion for KPSS* H 0 : I(0); H 1 : I(1) Sweden I(0) I(0) Finland I(0) I(0) Norway n/a n/a n/a n/a n/a n/a n/a Denmark I(0) I(0) Sweden I(0) I(0) Finland I(0) I(0) Norway n/a n/a n/a n/a n/a n/a n/a Denmark I(0) I(0) *The conclusions are derived at 5% significance level 932

5 The table 2 summarizes the ADF and KPSS tests results for four panels. When we use the raw sentiment data, the ADF and KPSS test results are inconsistent. This table has a support for KPSS application. However, we find that raw sentiment is stationary for all countries, except for Sweden CCI and Finland CCI. To get the consistent of non-unit root and stationary time-series we use the first difference in CCI and ESI for all countries. After using first differences in sentiment measures, we may conclude that the sentiment proxies seem to be reasonable regressors for our further analysis. Table 3 correlates the sentiment measures and macro variables presented separately in panels for each country. Table 3. Sentiment Proxies Correlation with Macro Variables Panel A: Sweden SE_CCI SE_CPI SE_ESI SE_GDP SE_IPI SE_T_BILL_3 SE_CCI SE_CPI SE_ESI SE_GDP SE_ IPI SE_T_BILL_ Panel B: Finland FI_CCI FI_CPI FI_ESI FI_GDP FI_IPI FI_T_BILL_3 FI_CCI FI_CPI FI_ESI FI_GDP FI_IPI FI_T_BILL_ Panel C: Norway NO_CCI NO_CPI EA_ESI NO_GDP NO_IPI NO_T_BILL_3 NO_CCI NO_CPI NO_ESI NO_GDP NO_IPI NO_T_BILL_ Panel D: Denmark DK_CCI DK_CPI DK_ESI DK_GDP DK_IPI DK_T_BILL_3 DK_CCI DK_CPI DK_ESI DK_GDP DK_IPI DK_T_BILL_ The methodology used for computation of ESI implies its close relations with CCI, which is proved in the table 3. We identify strong positive correlation of CCI and ESI for all panels. Furthermore, we find weak positive or negative correlation between sentiment proxies and GDP that is consistent across the countries presented in the table 3. Theory implies negative relation of CPI and a sentiment, which is proved by empirical findings. Comparing with the other macroeconomic variables we have no strong simultaneous correlations between the variables. Next we test for the Granger (non)-causality, with null hypothesis X does not Granger cause Y (X Y) with up to 12 months leading information taken into account. Granger causality test results are shown in Table 4. Granger Causality tests allow us to check for a time series dependencies between returns and sentiment measures such as ESI and CCI for Sweden, Finland, and Denmark. We cannot apply test for Norway because its CCI is reported quarterly beginning with 2000, thus we are left with 40 observations. Following the Granger causality analysis we find no evidence that CCI impacts returns in any country, except for Sweden, where CCI affects zero dividend yield portfolio. We find strong and statistically significant effect of logarithmically transformed index returns on CCI and ESI, but for Denmark index returns impact on CCI is insignificant. Index returns Granger cause ESI in all countries at 1% and 5% significance level. Prior literature explores sentiment effect on firms that are hard to value and to arbitrage and reports a more pronounced sentiment effect 933

6 on small and large portfolio returns. However, evidence on sentiment effect on growth and value portfolio return is inconsistent. We find that growth portfolio returns Granger cause ESI and CCI in all countries at least at 10% significance level, but we do find the opposite effect, except for CCI (in Sweden) and ESI (in Denmark) cause the growth portfolio returns. We do not find consistent and significant Granger causality for value portfolio returns across countries, except for Denmark, where value portfolio returns Granger cause ESI. We get mixed results of causality tests across countries for high yield and zero yield portfolio returns. The results for Denmark are economically insignificant. For Finland we find that returns affect ESI and CCI at 5% significance level. For Sweden results are mixed, we identify significant two-way causality between high yield returns and ESI, and between zero yield portfolio returns and CCI. H 0, pairwise Index Growth portfolio return BEME low Table 4. Pairwise Granger Causality Tests Value portfolio return high BEME High D/P portfolio returns Zero D/P portfolio returns Panel A: Sweden CCI R 0,4606 0,1482 0,4629 0,7874 0,0415 ** R CCI 0,0032 *** 0,0163 ** 0,0974 * 0,2700 0,0800 * ESI R 0,7724 0,0830 ** 0,2312 0,0356 ** 0,7106 R ESI 0,00003 *** 0,1283 0,1343 0,0702 * 0,0856 * Panel B: Finland CCI R 0,7901 0,5127 0,5845 0,4200 0,8675 R CCI 0,0124 ** 0,0137 ** 0,1890 0,0444** 0,0417** ESI R 0,6275 0,2498 0,6599 0,9376 0,8850 R ESI 0,0227 ** 0,0148 ** 0,1750 0,0130** 0,0430** Panel C: Denmark CCI R 0,6466 0,0440 ** 0,4657 0,8472 0,3295 R CCI 0,6373 0,0818 * 0,1446 0,3256 0,8930 ESI R 0,8960 0,9062 0,5369 0,8607 0,1932 R ESI 0,0014 *** 0,0065 *** 0,0431 ** 0,3711 0,2094 This table reports Granger-causality tests for sentiment measures and returns. The lag length is chosen by minimizing SIC. Asterisks refer to the level of significance: *significance at 0.1, ** significance at 0.05, *** significance at CCI, ESI and index were logarithmic transformed Our results for Granger causality test with CCI and returns are partly consistent with Brown and Cliff, 2005; and Schmeling, 2009 findings. Schmeling, 2009 finds two-way causality such that sentiment (CCI) depends on previous returns and that returns depend on previous sentiment movements. We find one-way causality, such that CCI does not Granger cause returns, but returns Granger cause sentiment at 1% significance level. Literature determines the relation of sentiment to returns or to macro variables. Bad or good news about returns or macro developments result in excessive investors optimism or pessimism reflected in CCI or ESI. Schmeling, 2009 considers the finding that returns drive sentiment and that sentiment drive subsequent returns quite reasonable. Even though our empirical study documents one-way causality that runs from returns to sentiment it allows us assume that sentiment can predict the future stocks returns. Predictive regressions of stock returns on sentiment This section presents the further results on the sentiment returns relations. To run predictive regressions we employ equation (1): R it = a0 + a1 T t k + ε it (1) Where T is a sentiment proxy, k is the forecasting horizon (equal to 1); ε it is a disturbance term. Wishing to establish k dependence we regress portfolio returns on k lagged sentiment proxies, k = 1, 6, 12, 24 months. The regression is done according to equation (2): R = δ + δ T +Ψ γ + ζ (2) it, + k 0 1 t t t+ k 934

7 Where T t is sentiment proxy (ESI and CCI ); Ψ - macro variables collected into matrix (logarithms of CPI, IPI,, GDP (quarterly) and the level of 3-month T-Bill rate) to net out effects of commonly employed risk factors on returns; k is the forecasting horizon (equal to 1); ζ t+ 1 - disturbance term. We regress index, value, growth, high yield and zero yield portfolio returns for all Scandinavian countries on ESI and CCI computed for European area and provide forecasts in Tables 5, 6, 7, 8, 9. Table 5. Forecasts results for 1, 6, 12, 24 months. Scandinavian index, value, growth, high yield and zero yield portfolio returns Forecast horizon 1 month 6 months 12 months 24 months Panel A: Scandinavian indices returns CCI ESI CCI ESI CCI ESI CCI ESI Sentiment p-value *** ** *** *** *** *** *** *** Panel B: Scandinavian value portfolio returns Sentiment p-value *** *** *** *** ** *** ** Panel C: Scandinavian growth portfolio returns Sentiment , p-value ** *** *** ** *** *** *** *** Panel D: Scandinavian high yield portfolio returns Sentiment p-value *** *** ** *** * ** ** ** Panel E: Scandinavian zero yield portfolio returns Sentiment p-value *** *** *** *** *** *** ** *** Asterisks refer to the level of significance: * significance at 0.1, ** significance at 0.05, *** significance at Table 5 provides the results of pooled predictive regressions by combining time-series observations across four Scandinavian countries to get robust results. All slopes coefficients are negative, that is consistent with the hypothesis that high sentiment has a negative effect on future stock returns. Overall, we find that almost all slopes are economically and statistically significant, that proves sentiment to be a good predictor of future returns. However, regressing value portfolio returns on CCI lagged at 12 months, slope seems to be insignificant. High insignificance of control variables (3-month T-Bill rate, logarithmically transformed IPI, CPI, and GDP) suggests keeping a parsimonious model. Table 6. Forecasts results for 1, 6, 12, 24 months. Sweden index, value, growth, high yield and zero yield portfolio returns Forecast horizon 1 month 6 months 12 months 24 months Panel A: OMX Stockholm Index CCI ESI CCI ESI CCI ESI CCI ESI Sentiment p-value *** *** * *** * Panel B: value portfolio returns Sentiment p-value *** ** Panel C: growth portfolio returns Sentiment p-value * ** Panel D: high yield portfolio returns Sentiment p-value * Panel E: zero yield portfolio returns Sentiment p-value ** *** ** *** Asterisks refer to the level of significance: * significance at 0.1, ** significance at 0.05, *** significance at

8 We run predictive regressions for each country. Table 6 summarizes the forecasts for Sweden at 1, 6, 12, and 24 forecasting horizon. We find a strong significant CCI effect on OMX Stockholm Index returns at 6, 12, and 24 months, ESI exhibits poor significance in forecasting of index returns. ESI cannot predict value, growth, high yield portfolio returns at any of forecasting horizons. CCI predicts value and high yield portfolio returns at 1- month horizon, growth portfolio returns at 24-month horizon respectively at 1%, 10%, and 5% significance level. ESI and CCI are able to predict zero yield portfolio returns at 1 and 6 months horizon. ESI seems to have higher predictive power than CCI, 1% and 5% significance level respectively. Table 7 summarizes the forecasts of index, value, growth, high and zero yield portfolio returns for Sweden for 1, 6, 12, and 24 forecasting horizon. Table 7. Forecasts results for 1, 6, 12, 24 months. Norway All-Share index, value, growth, high yield and zero yield portfolio returns Forecast horizon 1 month 6 months 12 months 24 months Panel A: Norway Oslo Bors All-Share Index CCI ESI CCI ESI CCI ESI CCI ESI Sentiment p-value *** *** *** *** Panel B: value portfolio returns Sentiment p-value *** *** ** *** ** * * Panel C: growth portfolio returns Sentiment p-value Panel D: high yield portfolio returns Sentiment p-value Panel E: zero yield portfolio returns Sentiment p-value ** ** * * ** * Asterisks refer to the level of significance: *significance at 0.1, ** significance at 0.05, *** significance at We find a strong significant CCI and ESI effect on Norway Oslo Bors All-Share Index returns at 6 and 12 months. To run predictive regressions for Norway we use CCI and ESI computed for European area, as there is no data available on Norway ESI and insufficient data on CCI that is compiled on the quarterly basis since Figure 1 proves ESI and CCI for European area to move in the same directions as the local ESI and CCI in the analyzed countries. Both ESI and CCI are insignificant in predicting index returns for 1 and 24 month horizon. Furthermore, ESI cannot predict value, growth, high yield portfolio returns for any of forecasting horizons. CCI predicts value and high yield portfolio returns for 1-month horizon, growth portfolio returns for 24-month horizon respectively at 1%, 10%, and 5% significance level. CCI proves to be a significant future stock returns predictor for all horizons, but ESI is insignificant for 1 and 12 month horizons and is of low significance for 6 and 24 month horizon. Table 8 summarizes the forecasts of index, value, growth, high and zero yield portfolio returns for Denmark for 1, 6, 12, and 24 forecasting horizon. Table 8. Forecasts results for 1, 6, 12, 24 months. Denmark share index, value, growth, high yield and zero yield portfolio returns Sentiment logarithm. transformed Forecast horizon 1 month 6 months 12 months 24 months Panel A: Københavns Fondsbørs Index CC CCI ESI CCI ESI I ESI CCI ESI

9 p-value Sentiment p-value ** * Panel B: value portfolio returns Sentiment p-value Panel C: growth portfolio returns Sentiment p-value Panel D: high yield portfolio returns Sentiment p-value Panel E: zero yield portfolio returns Sentiment p-value Asterisks refer to the level of significance: * significance at 0.1, ** significance at 0.05, *** significance at We find that CCI and ESI predict Københavns Fondsbørs Index returns for 6 month horizon at 5% and 10% significance level respectively. Overall, consistently with the prior empirical findings the slopes are negative, but are insignificant, therefore we argue that both sentiment measures cannot predict any other portfolio returns at any forecasting horizon in Denmark. Table 9 summarizes the forecasts of index, value, growth, high and zero yield portfolio returns for Finland for 1, 6, 12, and 24 forecasting horizon. Table 9. Forecasts results for 1, 6, 12, 24 months. Finland All-Share index, value, growth, high yield and zero yield portfolio returns Forecast horizon 1 month 6 months 12 months 24 months Panel A: Helsinki Stock Exchange All-Share Index CCI ESI CCI ESI CCI ESI CCI ESI Sentiment p-value ** ** * * * Panel B: value portfolio returns Sentiment p-value ** *** ** Panel C: growth portfolio returns Sentiment p-value * * * Panel D: high yield portfolio returns Sentiment p-value *** *** * ** * Panel E: zero yield portfolio returns Sentiment p-value *** *** *** *** *** *** ** Asterisks refer to the level of significance: * significance at 0.1, ** significance at 0.05, *** significance at We find a significant CCI and ESI effect on Helsinki Stock Exchange All-Share Index returns at 6 and 12 months horizon. CCI affects the returns for 24 month horizon, whereas ESI is insignificant. Both CCI and ESI affect value portfolio returns for 1 month horizon. And ESI predicts value portfolio returns at 5% significance level. Both sentiment measures seems to be insignificant or of low significance in predicting growth portfolio returns. However, they can predict high and zero yield portfolio returns for 1, 6, and 12 moth horizons. CCI proves to be insignificant for 12 and 24 month horizons, even though the coefficient is of the right sign (negative). Predictive regressions for long-short portfolios A good way to look for conditional characteristics effects is to use sentiment to forecast valueweighted portfolios that are long on stocks with high values of a characteristic and short on stocks with low 937

10 values. By controlling for macro variables in portfolio forecasting regressions, we can examine the extent to which the conditional predictability patterns are orthogonalized to economic conditions. The research uses the high-minus-low portfolio for netting out the conditional effects. Table 10 summarizes the results of predictive regressions with a long-short portfolio as a dependent variable and a sentiment as an independent variable. Thus, we run regressions of the type with and without controlling for macro economic indicators: Sentiment proxy R growth R = c + d sentiment 1 + β Ψ + ε (3) R value high Rzero = c + d sentimentt 1 t t t it + β Ψ + ε Table 10. Time-series regressions of Portfolio returns, CCI CCI* ESI ESI* d p(d) d p(d) d p(d) d p(d) Panel A: Sweden Growth-value * high zero yield *** * Panel B: Denmark Growth-value * high zero yield Panel C: Finland Growth-value ** high zero yield *** * Panel D: Norway Growth-value *** ** high zero yield Note: * - controlling for CPI, IPI, 3-month T-Bill, GDP (quarterly). Asterisks refer to the level of significance: * significance at 0.1, ** significance at 0.05, *** significance at CCI with controlling for macro indictors is insignificant across countries. When using alone CCI is significant for Sweden portfolio, which is long on growth and short on value portfolio at 10% of significance level; for Finland portfolio, which is long on high yield and short on zero yield portfolio returns at 1% significance level; for Norway portfolio, which is long on growth and short on value portfolio at 1% significance level, it is also significant at 5% when long-short portfolio is regressed on ESI. For Denmark, we have a significant ESI impact on portfolio, which is long on growth and short on value returns, after controlling for 3-month T-Bill rate, market return, and logarithmically transformed macro indicators (IPI, CPI, GDP). However, only ESI and Denmark market return are significant. When just ESI and market risk premium included into regression, all regression coefficients are significant at least at 5% significance level. We find that CCI has an explanatory power for portfolio, which is long on high yield and short on zero yield portfolio return for Finland at 1% significance level. After controlling for macro indicators CCI becomes insignificant. However, in a model with ESI and controls we find that for growth minus value portfolio ESI seems to be significant at 5% significance level. We find that for high minus zero yield portfolio ESI is insignificant, however, when the model includes controls, ESI is significant at 5% significance level, and GDP and 3-month T-Bill rate are significant at 1% significance level. We find the same for for CCI, that is significant at 1% and GDP and 3-month T-bill are also significant. For Norway we find a significant effect of CCI and ESI on growth minus value portfolio returns, at 1% and 5% significance level respectively. After controlling for macro variables both ESI and CCI are insignificant. However, in a model, where CCI is included, 3-month T-Bill rate and logarithmically transformed IPI are significant. Neither ESI nor CCI seems to be poor predictors of high and zero yield portfolio returns. Conclusions In accordance with the classical finance theory, investor sentiment does not play any role in the crosssection of stock returns. However, many research challenged that point of view. We build the research on theoretical arguments and prior empirical findings by Baker and Wurgler, 2006; 2007; Brown and Cliff, it (4) 938

11 2004, 2005; and Schmeling, However, our research findings support their inferences only to some extent. Running the pooled least squares regressions, regressing Scandinavian countries indices, value, growth, high yield, and zero yield portfolio returns on CCI and ESI, we find both measures are statistically significant predictors of stock returns at aggregate level. Aiming to prove investor sentiment effect on stock returns at cross-section, we find that in most of cases the relations between sentiment and stock returns are negative, but are not always statistically significant across countries. Predictive regressions of portfolio returns on k lagged sentiment proxies demonstrate that the results are significant for value stock portfolio returns for all forecasting horizons for Norway; for 1 and 6 month forecasting horizons for Finland; for 1 and 6 months for Sweden in case when CCI is used; for Denmark results are insignificant. Longer periods contain almost no information about portfolios returns. Overall performance for the 1 month forecasting period is better for CCI models. For growth portfolio we find no significant result for Norway, Denmark. However, for Finland we prove ESI and CCI to be poor predictors of growth portfolio returns. We find that CCI predicts growth portfolio returns at 6 and 24 month forecasting horizons at 10% and 5% significance level. Neither CCI nor ESI cannot predict high yield portfolio returns in Finland, Norway and Denmark. However, we find mixed results for ESI and CCI effect on high yield portfolio returns in Finland. ESI and CCI have an explanatory power I predicting zero-yield portfolio returns in all Scandinavian counties except for Denmark. The research uses the long-short portfolios for netting out the conditional effects, and the results are mixed. We get the significant results when the model includes only a sentiment proxy. When macro indicators are included as controls, in most of cases the model becomes insignificant. References 1. Baker, M., & Stein, J. C. (2004). Market Liquidity As a Sentiment Indicator. Journal of Financial Markets, No.7, Baker, M., & Wurgler, J. (2006). Investors Sentiment and the Cross-Section of Stock Returns. Journal of Finance, No. 61, Baker, M., & Wurgler, J. (2007). Investors Sentiment in the Stock Market. Journal of Economic Perspectives, Vol. 21. No. 2, Bandopadhyaya, A., & Jones, A. L. (2008). Measures of Investor Sentiment: A Comparative Analysis Put-Call Ratio Vs. Volatility Index. Journal of Business & Economics Research, August, Vol. 6, No. 8, Bandopadhyaya, A., & Jones, A.L. (2006). Measuring Investor Sentiment in Equity Markets. Journal of Asset Management, Vol. 7, No. 3/4, Barberis, N., & Shleifer, A., & Vishny R. (1998). A model of Investor Sentiment. Journal of Financial Economics, No. 49, Bergman, N. K., & Roychowdhur, S. (2008). Investor Sentiment and Corporate Disclosure. Journal of Accounting Research, Vol. 46, No. 5, Brown, G.W., & Cliff, M.T. (2004). Investor Sentiment and the Near-term Stock Market. Journal of Empirical Finance Brown, G.W., & Cliff, M.T. (2005). Investor Sentiment and Asset Valuation. Journal of Business, Vol. 78, No. 2, Chen, N., & Roll, R., & Ross, S. (1986). Economic Forces and the Stock market. Journal of Business, No. 59 (3), Clarke, R.G., & Statman, M. (1998). Bullish or Bearish? Financial Analyst Journal, May/June, Daniel, K., & Hirshleifer, D., & Subrahmanyam, A. (1998). Investor Psychology and Security Market Under- and Overreactions. Journal of Finance, No. 53, DeLong, J.B., & Shleifer, V., & Summers, L.H., & Waldmann, R.J. (1990). Noise Trader Risk in Financial Markets. Journal of Political Economy, No. 98, Elton, E.J., & Gruber, M.J., & Busse,.A. (1998). Do Investors Care about Sentiment? Journal of Business, No. 71, Fama, E., & French, K. (1992). The Cross-Section of Expected Stock Returns. The Journal of Finance, Vol. XLVII, No. 2, Fischer, K.L., & Statman, M. (2003). Investor Sentiment and Stock Returns. Financial Analyst Journal, March/April,

12 17. Glushkov, D. (2006). Sentiment Beta. Available at SSRN: He, W., & Mian, G. M., & Sankaraguruswamy, A. (2007). Market Sentiment, Investor Size and Reaction to Firm- Specific News. Working paper. 19. Hong, H., & Stein, J.C. (1999). A Unified Theory of Underreaction, Momentum Trading, and Overreaction in Asset Markets. The Journal of Finance, Vol. LIV, No. 6, Lee, C., & Shleifer, A., & Thaler, R. (1991). Investor Sentiment and the Closed-end Fund Puzzle. Journal of Finance, No. 46, Lee, W.Y., & Jiang, C.X., & Indro, D.C. (2002). Stock Market Volatility, Excess Returns, and the Role of Investor Sentiment. The Journal of Banking & Finance, No. 26, Lemmon, M., & Portniaguina, E. (2006). Consumer Confidence and Asset Prices: Some Empirical Evidence. Review of Financial Studies 19, Neal, R., & Wheatley S.M. (1998). Do Measures of Investor Sentiment Predict Returns? Journal of Financial and Quantitative Analysis, Vol. 333, No. 4, Qiu, L., & Welch, I., (2006). Investor Sentiment Measures. Working paper. 25. Schmeling, (2009). Investor Sentiment and Stock Returns: Some International Evidence. Journal of Empirical Finance 16 (2009), Solt, M.E., & Statman, M. (1988). How useful is the Sentiment Index. Financial Analyst Journal, Vol. 44, No. 5,