The Effect of Investor Sentiment On Stock Return in Norway and Vietnam - Master Thesis

BI Norwegian Business School – Master Thesis
- The effect of investor sentiment on

stock returns in Norway and Vietnam -
Supervisor:
Ilan Cooper
Students:
Thuy Uyen Thanh Tran
Ngoc Yen Thi Nguyen
Campus:
BI Oslo
Date of submission:
31.08.2013
Study Programme:
Master of Science in Business and Economics: Major in Finance
This thesis is a part of the MSc programme at BI Norwegian Business School. The school takes no
responsibility for the methods used, results found and conclusions drawn.
Master thesis 02.09.2013
Table of Content
Abstract ............................................................................................................................ ii
1. Introduction ................................................................................................................ 1
2. Literature Review ...................................................................................................... 4
3. Empirical Method and Data.................................................................................... 9
A. Empirical method......................................................................................................................... 9
B. Returns and characteristics ...................................................................................................... 9
C. Potential sentiment proxies ................................................................................................... 14
D. A composite sentiment index ................................................................................................. 17
4. Empirical Tests...................................................................................................... 25
A. Sorting .......................................................................................................................................... 25
5. Conclusions ............................................................................................................. 40
6. Bibliography ............................................................................................................ 41
7. Appendices ............................................................................................................. 45
Appendix 1: Comparison of signs of the sentiment proxies ............................................ 45
Appendix 2: Comparison of the lead-lag relationship among sentiment proxies .. 45
Appendix 3: Comparisons of sentiment effect on firm characteristics ....................... 45
Page i
Abstract
This study investigates the effects of investor sentiment on stock returns in

the Norwegian and Vietnamese stock markets. The model introduced by Baker
and Wurgler (2006) has been utilized in which a composite sentiment index has
been constructed based on six proxies. Two additional proxies for investor
sentiment, VIX and CCI, have been added in order to improve the estimating
power of the sentiment index. Through establishing portfolios of different types of
stocks, we found that the sentiment effect on returns is stronger for stocks that are
hard to value and hard to arbitrage, i.e. small, high volatility, non-dividend-
paying, and value stocks. Sentiment negatively predicts these types of stocks’
returns, i.e. when sentiment is low (high), future stock returns tend to be higher
(lower). Particularly in Norway, when sentiment is high, subsequent returns are
relatively low for small firms and unprofitable firms. In Vietnam, when sentiment
is high, subsequent returns are relatively low for small firms and firms with highly
volatile stock returns. And vice-versa.
The results from a robustness test of the orthogonalized sentiment indices
for Norway and Vietnam shows that the sentiment indices for Norway are
sensitive to VIX whereas the sentiment indices in Vietnam show no pattern. This
implies that VIX plays an important role when constructing the sentiment index in
a developed stock market, i.e. Norway, than in an emerging stock market, i.e.
Vietnam. CCI as a sentiment proxy can also forecast stock returns in Norway,
however, its predictive power is not as strong as VIX.
Page ii
1. Introduction
Whether investor sentiment affects stock returns has been an important

topic in recent academic literature. Investor sentiment is the propensity of
individuals to trade based upon emotions and ‘noise’ instead of facts. Due to
sentiment, investors form expectations about future cash flows and investment
risks that are not justified (Swedroe 2012). Conventional theories are the classic
argument against sentiment effects because they presume that investors are
rational. The idea is mainly that rational traders seeking to exploit profit
opportunities caused by mispricing will eliminate sentiment effects. However,
sentiment effects become more likely if rational traders are not able to fully
exploit these opportunities (Stambaugh, Yu and Yuan 2010). Behavioral finance
theories therefore contest the rationality hypothesis by assuming that investors are
irrational, and that they are prone to exogenous sentiment waves. Investors may
have incorrect stochastic expectations, either with overly pessimism or optimism,
which results in an incorrect valuation of asset values, causing asset prices to
deviate from their intrinsic values. As economic fundamentals are revealed and
sentiment diminishes, the mispricing is corrected. A negative relation between
investor sentiment and future stock returns is a consequence of this mispricing
correction, i.e. when sentiment is high (low), future stock returns tend to be lower
(higher). This indicates that investor sentiment can have a predictive power on
stock returns (Dergiades 2012; Chung, Hung and Yeh 2012).
Previous empirical studies have found that investor sentiment can predict
stock returns. Although the different studies have utilized various proxies for
investor sentiment, their common finding is that high sentiment has a negative
effect on stock returns and vice versa, i.e. investor sentiment can forecast stock
returns negatively in the time series (Brown and Cliff 2005). Schmeling (2009)
employed a cross-sectional perspective and provided evidence from 18
industrialized countries which showed that investor’s sentiment acts on average as
a significant predictor for stock returns. Fisher and Statman (2000) and Baker and
Wurgler (2006) have shown that there exist profitable strategies that take
advantage of stock return movements induced by sentiment fluctuations. The
latter study found that stock characteristics such as firm volatility, age and size,
can affect sentiment’s predictive effect on return. Lemmon and Portniaguina
Page 1
(2006) also provided evidence of investor sentiment predicting the returns on

small size stocks.
Most studies on investor sentiment have mainly focused on the U.S.
market. Looking at a Scandinavian market, the Oslo Stock Exchange (OSE) has a
long operating history as the main trading market of Norwegian shares. Norway is
considered an industrialized country in which its stock market is regarded as long-
lasting, and stable with sophisticated investors. It is therefore worth investigating
if the Norwegian stock market is affected by investor sentiment. Additionally,
whether the sentiment index in Norway retains the expected appealing properties
and conforms to Baker and Wurgler’s (2006) findings is also studied.
Vietnam, however, is characterized as an emerging economy. Rapidly
growing markets are expected to have a larger number of unsophisticated
investors, and thereby more irrational, than developed markets. As a consequence,
the market mispricing should be more affected by this sentiment, and the role of
investor sentiment should be tremendous in these countries (Kling and Gao 2008).
In particular, the Vietnamese market has been characterized as an emerging
economy and a transitional economy because of its lack of earnings history,
unlimited growth potential, and unsophisticated investors. High volatility
characterizes the young stock market in Vietnam, which was officially established
on July 20, 2000 with a base index value of 100 (VNINDEX). In 2007, it
experienced a spectacular bubble when VNINDEX rocketed up to 1167.36 points.
Consequently, this bubble was followed by a stock market crash when it dropped
from 921 to 316 points during 2008 and down to 235.5 in 2009. The total market
capitalization value slumped from 27.5% of GDP in 2007 to 10.5% in 2008, and
recently 14.8% in 2011. As a result, Vietnam was ranked as the third-worst-
performing market worldwide with a loss of 27% in 2011(World Bank 2011).
This market performance gives rise to the question whether the Vietnamese stock
market are profoundly affected by investor sentiment.
The goal of this paper is to investigate the predictive ability of investor
sentiment on the cross-section of stock returns in Norway and Vietnam. Following
Baker and Wurgler (2006), who claims that there exist no perfect sentiment
measure, a composite index is constructed in order to capture the common
component of several sentiment proxies. Particularly, the index of sentiment
changes is the first principal component of the changes in these six variables:
closed-end fund discount, turnover, number of IPOs, average first-day return on
Page 2
IPOs, the equity share in the new issues, and the dividend premium. In addition,
two proxies are added to their sentiment index: the Chicago Board Options
Exchange Market Volatility Index (VIX) and Consumer confidence index (CCI).
The eight variables are orthogonalized with respect to macroeconomic conditions
in order to remove business cycle variation from the sentiment proxies.
This paper differs from previous research by including two more proxies,
i.e. VIX and CCI. Baker and Wurgler (2007) mention these potential sentiment
proxies, however, Baker and Wurgler’s (2006) composite index has not studied
the estimating power of VIX and CCI. Since previous studies have focused on the
U.S. stock market, it is important to test the robustness of findings from the U.S.
market for other markets that are characterized, e.g. by a different composition of
the investor population. This study addresses this issue by presenting out-of
sample evidence on investor sentiment impact on the Norwegian and Vietnamese
stock market. This paper therefore contributes to the literature by conducting a
comparison on how differently investor sentiment affects stock market returns in
Norway and Vietnam. This type of analysis seems interesting for several reasons.
Firstly, adopting an international perspective allows us to form new hypotheses
regarding the impact of investor sentiment on returns. This study will explore how
different a developed market, such as Norway, is affected by investor sentiment
than an emerging market, such as Vietnam. Secondly, utilizing Norwegian and
Vietnamese stock return data provides a natural out-of-sample test for previous
findings from the U.S. As a result, this paper will examine the predictive ability of
investor sentiment on the cross-section of stock returns in Norway and Vietnam.
Furthermore, several sentiment indices are constructed from some
composition of sentiment proxies, i.e. the first-stage index with all lead-lag
proxies, the parsimonious index, the orthogonalized index, the orthogonalized
index without VIX, the orthogonalized index without CCI, and Baker and
Wurgler’s index. This procedure examines which one of the sentiment indexes
can explain most of the total variance of the first-stage index, which includes all
lead and lag sentiment proxies. Especially, whether VIX is a reliable proxy when
constructing investor sentiment index, which have not received formal research
attention, is also investigated.
In order to examine whether sentiment has cross-sectional effects on future
stock returns, a sorting approach and a regression approach have been conducted.
Using monthly stock returns, the sorting approach forms 10 equally-weighted
Page 3
portfolios based on seven firm characteristics. Consistent with Baker and Wurgler
(2006), patterns in the average returns across deciles reveal that when sentiment is
low, future returns are higher for small stocks, high volatility stocks, non-
dividend-paying stocks, and value stocks. The regression approach is also
conducted in order to test whether the sentiment index can forecast the returns on
several long-short portfolios. By using the factors introduced by Fama-French
(1993) and an additional momentum factor, this approach controls for the size
effect, the growth opportunity effect and the momentum effect. The results from
the regression approach partially supports the significance of the patterns found in
the sorting approach. In Norway, when sentiment is high, subsequent returns are
relatively low for small firms and unprofitable firms. In Vietnam, when sentiment
is high, subsequent returns are relatively low for small firms and firms with
volatile stock returns. And vice-versa. Generally, the results are consistent with
the predictions that sentiment has a more profound effect on stocks that are hard
to value and difficult to arbitrage.
The remaining parts of this paper is structured as follows. Section 2
reviews the existing literature and derives testable hypotheses. Section 3 presents
the data and the empirical methods used. Section 4 describes the empirical tests.
Section 5 concludes.
2. Literature Review
The common findings of the sentiment-return relation opposes the premise

of standard finance theory which assumes that stock prices reflect the discounted
value of future cash-flows and that arbitrageurs eliminate irrationalities among
market participants. Classical finance theories neglect the role of sentiment since
investors are presumed to be rational, whereas behavioral finance proposes that
waves of irrational sentiment, i.e. excessive optimistic or pessimistic expectations,
can persist and impact stock prices (Schmeling 2009). Thus, a mispricing caused
by uninformed demand shocks may occur. This is consistent with the assumption
that sentiment can be considered as the propensity to speculate and hence reflects
investor’s optimism or pessimism. Particularly, the sentiment effect on returns
should be stronger if arbitrage is risky because of subjective valuations, high
volatility, thin trading, and short selling constraints. Specifically, some emerging
Page 4
economies are characterized by the lack of an earnings history and unsophisticated

investors. As a consequence, the impact of investor sentiment should be
tremendous in these countries (Kling and Gao 2008).
A number of scholars have provided empirical evidence that show that
there exist a negative sentiment-return relation in the U.S stock market, and that
proxies for investor sentiment can predict stock returns negatively in the time
series (e.g. Brown and Cliff, 2005, Lemmon and Portniaguina, 2006). Delong et
al. (1990) predicted that noise trader sentiment can persist in financial markets,
and they found that there exist a negative relationship between investor sentiment
and future stock returns held by noise traders as the mispricing is eventually
corrected. Recently, Baker, Wurgler and Yuan (2010) examined the effect of
global and local components of investor sentiment on major stock markets, and
whether sentiment spreads across markets. The study concluded that both global
and local components of sentiment could predict the returns on high sentiment-
beta portfolios, e.g. those containing high volatility stocks or small, distressed,
and growth company stocks.
Baker and Wurgler (2006) investigated how investor sentiment impacts
the cross-section of stock returns by constructing an investor sentiment index
based on the six measures; trading volume as measured by NYSE turnover;
dividend premium (the difference between the average market-to-book ratio of
dividend payers and non-payers); CEFD; number of IPOs; average first-day
returns on IPOs; and equity share in new issues. The authors developed a hard-to-
value and difficult-to-arbitrage hypothesis in order to explain the cross-sectional
effect of sentiment associated with firm characteristics, particularly for young,
small size, unprofitable, growth, distressed, and non-dividend-paying stocks.
Because of these stocks’ lack of earnings history, tangible assets and collateral,
they are more sensitive to subjective valuations and fluctuations in the propensity
of speculation. Additionally, these stocks are likely to have lower liquidity and
higher idiosyncratic risk, which means that they tend to be the riskiest and
costliest to arbitrage. Therefore, these stocks are more profoundly affected by
shifts in investor sentiment. The authors found that when beginning-of-period
proxies for sentiment are low (high), the following returns are relatively high
(low) for small, young, growth and distressed stocks. Building on these findings,
Grigaliuniene and Cibulskiene (2010) conducted a study on the sentiment-return
relation at an aggregate level and cross-sectionally in the Scandinavian stock
Page 5
market. Consistent with prior literature, the authors found that high sentiment has
a negative impact on future stock returns, in which this effect is stronger for hard-
to-value and hard-to-arbitrage stock returns (e.g. growth vs. value, dividend
paying vs. non-paying). However, results from different studies are controversial.
Baker and Wurgler (2007) reported that sentiment shifts impacts small stock
returns rather than large stocks. Glushkov (2006) investigated sentiment betas and
showed that value and hard-to-arbitrage stocks are more strongly affected by
sentiment. Brown and Cliff (2005) conclude that the sentiment impact is mostly
concentrated in large-capitalization growth stocks. On the contrary, Brown and
Cliff (2004) found limited evidence of sentiment impacting small stocks. Building
on previous literature and findings, particularly Baker and Wurgler (2006), this
paper will test whether a wave of investor sentiment has larger effects on
securities whose valuations are highly subjective and difficult to arbitrage in
Norway and Vietnam:
Hypothesis 1. The sentiment effect on returns is stronger for stocks that are hard
to value and hard to arbitrage, e.g. small, growth, and value stocks.
When testing for the impact of sentiment on stock returns, a fundamental question
that arises is how to measure the sentiment. Previous papers have utilized various
proxies where closed-end fund discounts (CEFD) have been a popular proxy, e.g.
Lee, Schleifer and Thaler (1991), found that market-wide sentiment affects the
differences between close-end fund prices and their net asset values. Ritter (1991)
used IPO stocks, and provided evidence of long-run reversals in returns on IPO
stocks. His evidence is in line with periodic waves of optimism that particularly
affect young growth stock prices. Other scholars used investor surveys data
(Brown and Cliff 2005), and micro trading data (Kumar and Lee 2006). Lemmon
and Portniaguina (2006) utilized consumer confidence indexes as a proxy for
sentiment. Another alternative measure of investor sentiment that has received
little attention in previous literature is the Chicago Board Options Exchange’s
market Volatility Index (VIX), which is also called the “investor fear gauge”. This
index expresses investor’s consensus view about expected future stock market
volatility. It is constructed from implied volatilities of S&P 500 index options, and
is used by traders as a sentiment indicator in which a high VIX indicates high fear.
Page 6
(Whaley 2000). Kurov (2010) used the VIX index as an alternative investor
sentiment measure, and found that investor sentiment plays a significant role in
the effect of monetary policy on the stock market. This paper will use the VIX
index as a proxy for investor sentiment, in which it is going to be added to Baker
and Wurgler’s (2006) sentiment index:
Hypothesis 2. VIX as a proxy for investor sentiment can forecast stock returns, in
which the sentiment-return relation is significantly negative.
However, overall the results from previous literature about investor sentiment are
by no means uniform. Brown and Cliff (2004) who used CEFD, found limited
proof to support the predictive power of sentiment on stock returns. Qiu and
Welch (2005) documented weak correlation between CEFD and sentiment,
however, the consumer confidence correlated well with investor sentiment. The
consumer confidence index has therefore received some attention in the literature
as a measure of sentiment. For example, Fisher and Statman (2003) reported that
consumer confidence has a positive correlation with other sentiment proxies such
as the sentiment measure compiled by the American Association of Individual
Investors (AAII). Doms and Morin (2004) reports that the measures of consumer
confidence include an irrational element because it responds to the tone and
volume of economics news reports rather than economic content. Schmeling
(2009) investigated the sentiment-return relation internationally by utilizing
consumer confidence as a proxy for investor sentiment in 18 industrialized
countries. In most of these 18 industrialized countries, the author found that when
consumer confidence is high, future stock returns tend to be lower and vice versa.
On average, sentiment negatively forecasted aggregate stock market returns across
these countries. The above findings highlights why consumer confidence is a
reasonable measure for investor sentiment. This paper will therefore build on
previous studies by utilizing consumer confidence as a proxy for investor
sentiment, in which it is going to be added to Baker and Wurgler’s (2006)
sentiment index:
Hypothesis 3. Consumer confidence as a proxy for investor sentiment can

forecast stock returns, in which the sentiment-return relation is significantly
negative.
Page 7
The differences between Asian and Western cultures have been an important topic
in previous literature. In social psychology, Hofstede (1980) suggested that in a
Western culture, when a catastrophic loss occurs, a person is expected to sustain
the adverse outcomes of his decisions on his own, while in an Asian culture, his
family and friends will intervene to support him. This is an example of the
difference on how a Western and Asian culture is affected by a disastrous loss,
which indicates that the risk attitude and the propensity to speculation and
arbitrage might also differ (Lin 2010). Furthermore, the accumulated studies on
sentiment focus mainly on the U.S market or other developed markets, which
leads to the question whether this relation holds outside these developed markets.
There exist important exceptions, most notably in Asia. Although momentum
profits is large and significant in the U.S and most European countries, it has been
found that momentum profits is absent in Japan and the rest of Asia
(Rouwenhorst, 1998). Therefore, it is important to test the robustness of the
findings from the U.S. market for other markets that are characterized, for
example, as emerging markets. Rapidly growing markets such as emerging
markets are expected to have a larger number of unsophisticated investors, and
thereby more irrational, than developed markets. As a consequence, the market
mispricing should be more affected by this sentiment, and investor sentiment
should have a greater effect on an emerging market than a developed market
(Kling and Gao 2008). This paper will test whether the return-sentiment relation
holds outside the U.S. by including out-of sample evidence of sentiment effect on
the Norwegian and Vietnamese market. These two markets are of special interest
because Norway is an industrialized, developed European country, whereas
Vietnam is a transitional, emerging Asian economy.
Page 8
3. Empirical Method and Data

A. Empirical method
Employing Baker and Wurgler’s (2006) empirical method, the cross-

sectional impact of investor sentiment on stock returns was captured based on the
following model:
[ ]
where i represents firms, t represents time, x represents a vector of firms’

characteristics, and T represents a sentiment proxy. The coefficient captures
the generic effect of investor sentiment, captures the generic effect of firm
characteristics on stock returns whereas captures sentiment-driven mispricing
in cross-sectional patterns. Therefore, the null hypothesis, i.e. , suggests
that the non-zero effect on stock returns only exists for compensation of
systematic risk. On the contrary, if the null hypothesis is rejected, i.e. ,
systematic patterns of correction for mispricing might be expected. A cross-
sectional approach is taken into consideration due to the fact that the causes of
mispricing on stocks vary across sections, namely the stock fundamental
characteristics.
B. Returns and characteristics
The data concerning monthly stock returns and characteristics on the firm
and security level: size and total risks, profitability, dividend policy, tangibility,
growth opportunities and distress, are from the DataStream database. The sample
consists of all common stocks in the Oslo Bors All-Share Index (OSLO SE OBX)
and Ho Chi Minh Stock Exchange (HOSE) from January 01, 1991, for the former,
and from July 28, 2000, for the latter, up to July 01, 2013. Regardless of the lack
of information in Vietnam’s subsample before 2005, this database is still utilized
in order to ensure data consistency across countries and to avoid several biases
from analyses, e.g. survivorship bias. For simple calculation, the firm-level data in
the previous year (t-1) is matched to the monthly returns in the current calendar
year (t).
Page 9
Table I provides descriptive statistics for all returns and characteristic

variables following Baker and Wurgler’s (2006) definitions. Panel A shows the
returns variables. Returns (R) are computed from monthly changes in the Total
return index (RI) which includes dividend yield. Momentum (MOM) is calculated
as the accumulation of 11 monthly returns from 12 to 2 months prior to the given
month.
The firm and security characteristics data are summarized in the different
panels. Panel B reports the size and total risk characteristics. Size is calculated as
the log of market equity (ME). Market equity (W08001) is defined as the multiple
of the stock price and its number of common shares outstanding. Total risk (σ) is
computed as the annualized standard deviation in monthly returns for the 12
month period, from January to December each year. Total standard deviation over
the first 6 months of 2013 for all stocks on OBX and HOSE are also annually
estimated.
Panel C presents the profitability variables. The ratio of earnings to book
equity, i.e. return on equity (E/BE) is defined for firms with positive earnings.
Earnings (E) are measured as the net income before extraordinary items/preferred
dividends (WC01551), plus deferred income taxes and investment tax credit on
income statements (WC04101), and minus preferred dividend requirements
(WC01701). Book equity (BE) is computed as total shareholders’ equity
(WC03995) plus deferred taxes (WC03263) on balance sheets. The dummy
variable for profitability (E>0) is set value to one for profitable firms and zero for
unprofitable firms.
Panel D shows dividend characteristics, which are consist of the ratio of
dividends to equity (D/BE). Dividends (D) are defined as the multiplicity of
dividends per share (DPS) and the number of shares outstanding (W05301). The
dummy variable for dividend policy (D>0) is set to the value of one for firms
which pay positive dividends and zero for firms which pay no dividends.
Panel E summarizes characteristics of asset tangibility, i.e. PPE/A and
RD/A. The former (PPE/A) is the proportion of gross plant, property, and
equipment (WC02301) whereas the latter is the proportion of research and
development expense (WC01201) in total assets (WC02999). However, the data
concerning R&D are relatively insufficient among concerned variables.
Characteristics of growth opportunities and distress or both are represented
in Panel F. The book-to-market ratio (BE/ME) is calculated as the book equity
Page 10
over market equity for the 12 month period prior to the current observation. The
characteristic of external finance (EF/A) is measured as the ratio of external
finance (WC04500), which consist of company financing from outside sources, to
the total assets in the previous year. Following Baker and Wurgler (2004) sales
growth is measured as the percentage change of net sales or revenues (WC01001)
over the year.
The subsample means of the returns and characteristics variables are also
calculated in order to give an overview of trends over time.
Page 11
Table IA
Summary Statistics, Norway, 1991-2013
This table presents the descriptive statistics for all returns and characteristics variables. Panel A shows the returns variables. Returns (R) are monthly returns computed as monthly
changes in the Total return index (RI) which includes dividend yield. Momentum (MOM) is calculated as the cumulative return for the 11-month period between 12 and 2 months
prior to t. Panel B reports the size and total risk characteristics. Size is calculated as the log of market equity (ME). Market equity (ME) is defined as stock price times common
shares outstanding. Total risk (σ) is computed as the annualized standard deviation in monthly returns for the 12 month period, from January to December each year. The return on
equity (E/BE) is defined for firms with positive earnings. Earnings (E) are calculated as income before extraordinary items plus income statement deferred taxes minus preferred
dividends. Book equity (BE) is calculated as shareholder’s equity plus balance sheet deferred taxes. The profitability dummy variable (E>0) equals one for profitable firms and zero
for unprofitable firms. Panel D summarizes dividend variables, which includes dividends-to-equity (D/BE). Dividends (D) are defined as dividends per share times shares
outstanding. The dividend dummy variable (D>0) is equal to one for firms with positive dividends and zero for non-paying dividend firms. Panel E summarizes measures of
tangibility. Plant, property, and equipment (PPE) and research and development (RD) are scaled by total assets (A). Panel F summarizes growth opportunities and distress variables.
The book-to-market ratio (BE/ME) is calculated as the book equity over market equity for the 12 months prior to t. External finance (EF) represents company financing from outside
sources. It includes the issuance and retirement of stock and debt. Sales growth (GS) is computed as the percentage change in net sales over the year. In Panels C through F,
accounting data from the fiscal year ending in t − 1 are matched to monthly returns in calendar year t.
Full Sample Subsample Means
N Mean Median SD Min Max 1990s 2000s 2010-2013
Panel A: Returns
Rt(%) 50111 0.99 0 18.72 -98.72 1400 1.41 0.82 0.68
MOMt-1(%) 42413 10.33 9.37 62.44 -396.92 1270 19.79 12.24 -5.31
Panel b: Size and Total Risk
MEt-1(1000NOK) 46328 4356.65 625.53 23394.13 0.83 539000.00 1875.15 4810.68 6808.37
σt−1 (%) 47598 46.97 37.72 43.08 0 1422.97 43.61 47.35 50.88
Panel C: Profitability
E+/BEt−1 (%) 48030 -3.18 7.05 465.16 -19102.59 9767.00 5.89 0.14 -33.37
E > 0t−1 53970 0.66 1 0.47 0 1 0.73 0.64 0.58
Panel D: Dividend Policy
D/BEt−1 (%) 42495 4.09 0.28 17.29 -136.12 531.35 2.99 4.72 4.13
D > 0t−1 44585 0.51 1 0.50 0 1 0.59 0.47 0.43
Panel E: Tangibility
PPE/At−1 (%) 36120 121.25 67.82 1937.18 0 105182.50 103.15 136.66 98.79
RD/At−1 (%) 9968 8.40 1.49 23.57 0 504.42 7.86 9.79 5.93
Panel F: Growth Opportunities and Distress
BE/MEt−1 41623 1.42 0.84 3.08 -79.58 71.26 1.65 1.27 1.47
EF/At−1 (%) 46813 454.11 2.47 24648.92 -1887.49 1534837.00 25.36 827.01 12.52
GSt−1 (%) 51521 11.73 0 791.88 -161.04 118421.60 2.66 13.64 21.09
Page 12
Table IB
Summary Statistics, Vietnam, 2005-2013
This table presents the descriptive statistics for all returns and characteristics variables. Panel A shows the returns variables. Returns (R) are monthly returns computed as monthly
changes in the Total return index (RI) which includes dividend yield. Momentum (MOM) is calculated as the cumulative return for the 11-month period between 12 and 2 months
prior to t. Panel B reports the size and total risk characteristics. Size is calculated as the log of market equity (ME). Market equity (ME) is defined as stock price times common
shares outstanding. Total risk (σ) is computed as the annualized standard deviation in monthly returns for the 12 month period, from January to December each year. The return on
equity (E/BE) is defined for firms with positive earnings. Earnings (E) are calculated as income before extraordinary items plus income statement deferred taxes minus preferred
dividends. Book equity (BE) is calculated as shareholder’s equity plus balance sheet deferred taxes. The profitability dummy variable (E>0) equals one for profitable firms and zero
for unprofitable firms. Panel D summarizes dividend variables, which includes dividends-to-equity (D/BE). Dividends (D) are defined as dividends per share times shares
outstanding. The dividend dummy variable (D>0) is equal to one for firms with positive dividends and zero for non-paying dividend firms. Panel E summarizes measures of
tangibility. Plant, property, and equipment (PPE) and research and development (RD) are scaled by total assets (A). Panel F summarizes growth opportunities and distress variables.
The book-to-market ratio (BE/ME) is calculated as the book equity over market equity for the 12 months prior to t. External finance (EF) represents company financing from outside
sources. It includes the issuance and retirement of stock and debt. Sales growth (GS) is computed as the percentage change in net sales over the year. In Panels C through F,
accounting data from the fiscal year ending in t − 1 are matched to monthly returns in calendar year t.
Full Sample Subsample Means
N Mean Median SD Min Max 2005-2009 2010-2013
Panel A: Returns
Rt(%) 15744 -0.19 -1.76 16.37 -67.82 389.28 0.72 -0.51
MOMt-1(%) 8680 -9.37 -16.16 62.85 -278.83 289.64 -47.19 -6.25
Panel b: Size and Total Risk
MEt-1(1000NOK) 14900 1870 359 6080 11.02 7340 1330 2090
σt−1 (%) 12215 49.14 43.03 27.08 2.96 387.94 64.92 45.71
Panel C: Profitability
E+/BEt−1 (%) 19099 26.51 16.82 96.74 -231.05 3281.05 33.22 19.88
E > 0t−1 21767 0.96 1.00 0.21 0 1.00 0.97 0.95
Panel D: Dividend Policy
D/BEt−1 (%) 12873 6.19 5.75 11.57 -786.70 68.49 4.81 6.84
D > 0t−1 11904 0.73 1.00 0.44 0.00 1.00 0.53 0.78
Panel E: Tangibility
PPE/At−1 (%) 18533 52.22 42.36 43.10 0.18 358.92 56.99 47.52
RD/At−1 (%) 84 1.39 0.45 1.56 0 4.35 1.35 1.45
Panel F: Growth Opportunities and Distress
BE/MEt−1 12366 1.17 0.95 0.88 -1.47 6.66 0.91 1.30
EF/At−1 (%) 19075 27.64 -0.44 644.65 -42.93 25661.58 21.21 33.98
GSt−1 (%) 21418 4.11 0.00 162.06 -100.00 22865.75 3.75 4.38
Page 13
C. Potential sentiment proxies
In previous literature, investor sentiment can be investigated by two

approaches; explicit sentiment proxies based on investor surveys, and implicit
sentiment proxies based on market variables. The latter approach has attracted
much attention, in which the overall sentiment is derived from market statistics,
e.g. price movements, trading patterns, etc. However, the forecasting power of
each individual sentiment proxy as a sentiment index is quite poor due to its own
idiosyncratic component. Following a similar methodology introduced by Baker
and Wurgler (2006), a composite sentiment index is constructed on the common
variation basis of 6 proxies including the closed-end mutual fund discount
(CEFD), share turnover (TURN), the number of IPOs (NIPO), the average first-
day return on IPOs (RIPO), the share of equity issues (ES), and the dividend
premium (PD-ND).
Firstly, the closed-end mutual fund discount (CEFD), is defined as the
year-end, value-weighted average discount on closed-end mutual funds. It is
measured as the ratio of the difference between a fund’s net asset value (NAV)
and its market price to its NAV. CEFD takes positive values if funds are trading at
a discount and vice versa. Previous studies, e.g. Zweig (1973) and Delong et al.
(1990), argue that the average CEFD might be a sentiment index which captures
investor expectations, i.e. the more bearish the retail investors are, the higher the
discount is, as a compensation for the buyers (Baker and Wurgler 2007). CEFD is
expected to have a negative relationship with the sentiment factor. There exist 4
closed-end funds in the Norwegian sample and 5 funds in the Vietnamese sample.
Secondly, share turnover (TURN) is the ratio of total value of shares traded
during the period divided by the average market capitalization for the period.
Average market capitalization is calculated as the average of the end-of-period
values for the current period and the previous period. Turnover, or liquidity
generally, might capture sentiment due to the fact that irrational investors prefer
betting on raising stocks in a market with short-sales constraints when they are
optimistic than pessimistic and therefore add liquidity (Baker and Stein 2004).
Consistent with prior literature, the relationship between turnover and market
returns is expected to be negative (Jones 2001). TURN is defined as the natural log
of turnover which is obtained from Thomson Reuters DataStream.
Page 14
Thirdly, the number and the average first-day return on IPOs (NIPO,
RIPO) are obtained and calculated from Oslo Bors Information (OBI) and HOSE.
NIPO and RIPO are annually measured. Baker and Wurgler (2006) suggest that
NIPO, i.e. IPO volume, which represents the underlying demand for IPOs
increases when the sentiment is high. However, it has a characteristic of high
fluctuation and large sensitivity to investor sentiment. RIPO is calculated as the
difference between the first trading price and the offer price divided by the offer
price. RIPO, which is computed as an equal-weighted return in the observed
month, is expected to be positively related to investor sentiment. However, data
on RIPO is not accessible in Vietnam due to lack of information.
Fourthly, the share of equity issues (ES) in total equity and debt issues by
all firms, or more generally equity financing activity, may measure investor
sentiment. Baker and Wurgler (2000) find that equity might be overvalued due to
investor sentiment. Therefore, there is more equity issuance than debt issuance in
order to reduce the cost of capital when sentiment is high, i.e. high values of the
equity share forecast low stock market returns. ES is measured as the proportion
of aggregate equity issuance in aggregate equity and debt issuance published
annually by Oslo Bors and HOSE.
Finally, the dividend premium (PD-ND) is defined as the log difference
between the average market-to-book ratios of the payers and nonpayers. The
dividend premium represents a firm’s propensity to pay dividends and can serve
as a proxy for a characteristic of safety, i.e. those firms which are larger, more
profitable, but with lower growth opportunities (Baker and Wurgler 2007). An
inverse relationship is expected between PD-ND and sentiment investor. PD-ND is
calculated at the end of the year after sorting all securities on OBX and HOSE into
payers and nonpayers using raw data from DataStream.
Additionally, in recent literature, multiple potential sentiment proxies may
be considered in order to increase the forecasting power of these common proxies
as well as tackle data insufficiency. Baker and Wurgler (2007) suggest both
explicit and implicit sentiment proxies including investor surveys, investor mood,
retail investor trades, mutual fund flows, option implied volatility, and insider
trading. Among these proxies, Option Implied Volatility and Consumer
Confidence Index are employed to establish a composite sentiment index.
Firstly, Option Implied Volatility (VIX) might reflect investor sentiments
generally due to the fact that the greater the forecasted volatility might be, the
Page 15
higher the expected option price should be (Baker and Wurgler 2007). This
implied volatility might be characterized by VIX, which is a measure of Standard
and Poor’s 100 index option volatility on the Chicago board of exchange. VIX is
considered as “investor fear gauge” because it is likely to increase sharply when
markets decline steeply during financial stress. VIX is simply obtained from
Yahoo Finance.
Secondly, Consumer Confidence Index (CCI) is employed as the only
explicit sentiment proxy as proposed by Lemmon and Portniaguina (2006), Qui
and Welch (2006). This metric is defined as the degree of consumers’ optimism
on the economic state that was presented in their saving and spending activities.
There are several reasons why CCI should be included to capture investor
sentiment. Firstly, there exist data for consumer confidence internationally, in
both developed and developing countries. Secondly, this metric can be collected
easily for reasonable periods of time as time-series data. Thirdly, although CCI is
measured slightly different across countries, it seems to be the most consistent
comparability of sentiment data. CCI is not obtained directly from trading data but
from the behavior of respondents through surveys on their expectations about their
financial situation as well as the whole economy. Hence, CCI contains an
irrational element which is needed in order to investigate investor sentiment
(Doms and Morin 2004). Moreover, previous literature report no or weak
correlation of closed-end fund discounts with investor sentiment (Brown and Cliff
(2004), Qiu and Welch (2005). However, other studies showed evidence of the
correlation between CCI and the other sentiment proxies (Fisher and Statman
2003). Consistent with previous findings, CCI is expected to be inversely related
with stock returns (Schmeling 2009). Data on CCI are obtained from DataStream
(NWCNFCONQ) for the period of 10 years (Q3 1992 - Q4 2012) in Norway and
from the database of The Nielsen Company, available from Q1 2006 to Q4 2012.
Page 16
D. A composite sentiment index
Prior research show that each of the above-mentioned proxies might serve
as a sentiment index. However, beside the sentiment component each proxy may
capture, there also exist other components, which are unrelated to the sentiment
factor, i.e. idiosyncratic components. Therefore, following a similar methodology
proposed by Baker and Wurgler (2006), a composite sentiment index is formed
based on the common variation in the chosen proxies. Principal Components
Analysis (PCA) is utilized in order to isolate the sentiment component which is
common among those proxies.
Table II presents the summary statistics of sentiment proxies which are
employed to extract the sentiment component for each country. As for Norway, 8
proxies are analyzed including CEFD, TURN, NIPO, RIPO, ES, PD-ND, VIX, and
CCI. As for Vietnam, a sentiment index is constructed from 6 proxies including
CEFD, TURN, NIPO, ES, VIX and CCI. The data on dividend premium in
Vietnam is available; however, it is excluded due to the fact that it has no
correlation with other proxies.
Baker and Wurgler (2006) find that the determination of relative timing of
those proxies is of concern due to a non-contemporaneous relationship between
these proxies and investor sentiment. Some proxies may not reflect the fluctuation
of sentiment simultaneously but reveal the sentiment earlier or later than the
others, i.e. a lead-lag relationship. In prior research, ES and NIPO, which are
related to firm supply responses are supposed to lag behind CEFD, TURN, RIPO,
PD-ND, which are related to investor behavior (Ibbotson and Jaffe 1975, Lowry and
Schwert 2002). There is no clear evidence that CCI and VIX have a lead or lag
relationship with other proxies in prior research. However, VIX and CCI are
expected to reflect the sentiment simultaneously because they are based on
investor behavior. Baker and Wurgler (2006) find that TURN, RIPO, and PD-ND
take longer to fully reveal the sentiment compared to CEFD, NIPO, and ES.
Through conducting a PCA, the first principal component of 8 proxies in
Norway, i.e. CEFD, TURN, NIPO, RIPO, ES, PD-ND, VIX, CCI, and their one-year
lags, i.e. CEFDt-1, TURNt-1, NIPO t-1, RIPO t-1, ES t-1, PD-ND t-1, VIX t-1, CCI t-1 are
estimated for Norway. This process results in the first-stage index with 16
loadings, i.e. component coefficients, for each proxy and its lag. The correlation
between the first-stage index and 8 pairs of a proxy’s lead or lag are then
Page 17
calculated. Consequently, 8 current or lagged proxies which have higher

correlation within each pair are selected. The PCA is repeated to extract the first
principal component of these 8 chosen proxies. After the coefficients are rescaled,
the parsimonious index has unit variance and is estimated as the following
equation:
(1)
As a result, 65.94% of the sample variance can be explained by the first

principal component, suggesting that most of the common variation are captured
by this factor. The pair-wise correlation between the SENTIMENT index (1) and
the first-stage index with 16 loadings is estimated at 0.92, indicating that the
estimating power of the 8 loadings that were left unchosen is not substantial, i.e.
little information is lost (Baker and Wurgler 2006).
Following a similar approach, the first-stage index with 12 loadings for
Vietnam is estimated as the first principle component of 6 proxies including
CEFD, TURN, NIPO, ES, VIX, CCI and their lags CEFDt-1, TURNt-1, NIPO t-1,
ES t-1, VIX t-1, CCI t-1. After the selection between the lead or lag of the 6 proxies
based on its highest correlation with the first-stage index, the PCA procedure is
applied again and extract the parsimonious index for Vietnam, which explains
72.29% of the sample variance and has the pair-wise correlation with the first-
stage index estimated at 0.93.
SENTIMENTt = 0.052CEFDt-1 0.244TURNt-1+ 0.293NIPOt-1

0.382ESt-1 + 0.390VIXt 0.081CCIt (2)
In the SENTIMENT index (1) for Norway, most of the estimated signs
meet expectations, i.e. CEFD, TURN, NIPO, VIX, and CCI, whereas only RIPO,
PD-ND and ES are not consistent. Only CEFD and CCI follow the order of lead-lag
relationship as expected above. However, the SENTIMENT index (1) supports
the result founded by Baker and Wurgler (2007), i.e. TURN, RIPO, and PD-ND lag
behind CEFD, NIPO, and ES. However, the SENTIMENT index (2) for Vietnam
displays quite few expected properties due to the fact that few estimated signs are
as predicted, i.e. CEFD, NIPO. Nevertheless, NIPOt-1, ESt-1, VIXt and CCIt follow
predictions for their lead-lag relationship.
Page 18
These 2 parsimonious indices seem not to be totally consistent with the

predictions due to the fact that the PCA treats and extracts a common component
based on common variation in the same way regardless of sentiment component
or just business cycle component. Moreover, because investors are overly
optimistic or pessimistic based on a series of news, returns, or macro
developments (Qiu and Welch 2005), a sentiment index should be removed the
effects of business cycle. Based on the earlier empirical research, an identical
composition of 4 additional macroeconomic variables motivated by asset pricing
theory is employed. Firstly, Consumer Price Index, i.e. CPI, is to measure the
development of the cost of living (Brown and Cliff 2005, Lemmon and
Portniaguina 2006, Schmeling 2009). Secondly, Industrial production index, i.e.
IPI, is the indicator measuring the output amount of manufacturing, mining,
electric and gas industries. Thirdly, Gross domestic product, i.e. GDP, is an
indicator for total value added in a country as well as gross income from domestic
production (Chen, Roll and Ross 1986, Lemmon and Portniaguina 2006). Finally,
the policy interest rate is employed to reflect the foundations of monetary policy
in each country. All these 4 indicators are obtained from DataStream for Norway
and Vietnam on a yearly basis. Therefore, each of 8 proxies which exist in the
SENTIMENT index (1) and 6 proxies in the SENTIMENT index (2) are regressed
as a function of 4 macroeconomic variables respectively. The residuals of these
regressions, labeled with a superscript ⊥, may serve as cleaner proxies, i.e.
orthogonalized proxies, for investor sentiment.
Using the same PCA procedure, a second sentiment index for Norway is
extracted as the first principal component of 8 orthogonalized proxies, which
explains 48.43% of the sample variance
(3)
The first eigenvalue is estimated at 3.87, compared to 2.02 of the second

eigenvalue. There are three changes of estimated signs in 3 components, i.e.
CEFD, RIPO, ES. Although the pair-wise correlation with the first-stage index at
0.82, lower than before controlling for other macroeconomic effects, the
orthogonalized index still maintains appealing properties from the SENTIMENT
index (1) considering the signs and lead-lag relationship among the proxies.
Page 19
Consequently, the signs of TURN, NIPO, RIPO, ES, VIX and CCI are now as
predicted.
A orthogonalized index for Vietnam is also constructed as
(4)
As a result, the sign of TURN changes and the second index conform to
expectations. The resulting index is the second principal component extracted
through the PCA of 6 orthogonalized proxies. This component, which has
eigenvalue of 2.378 and explains 39.63% of the total variance, has a pair-wise
correlation of 0.74 with the first-stage index. The first principal component is
excluded due to the fact that it fails to capture the common variance of these 6
orthogonalized proxies. The above-mentioned pair-wise correlation is estimated at
-0.69, although the eigenvalue is estimated at 3.071 and 51.19% the total variance
is explained (Appendix 1 and 2).
Table II provides the descriptive statistics for all raw sentiment proxies as
well as for orthogonalized proxies after controlling for macroeconomic conditions
for both countries. The correlations within sentiment components as well as the
correlations with the parsimonious indices (1) and (2), and the orthogonalized
indices (3) and (4) are also given. The correlations among the orthogonalized
proxies tend to be slightly higher than raw proxies.
Figure 1 compares all the raw proxies and the residuals from the
regressions on the composite of macroeconomic variables. Panel E plots the first
principal component index of the 8 raw and orthogonalized proxies for Norway
and the second principal component index of 6 proxies for Vietnam, respectively.
Several proxies, i.e. CEFD, ES in Norway and TURN, ES in Vietnam, are
influenced by economic conditions as illustrated. Therefore, the orthogonalized
proxies are employed in further analyses.
Page 20
Table IIA
Investor Sentiment Data, Norway, 1991-2013
This table presents the means, standard deviations, and correlations for measures of investor sentiment. Panel A reports the raw sentiment proxies. Closed-end mutual funds (CEFD)
are defined as the year-end, value-weighted average discount on closed-end mutual funds. It is calculated as net asset values (NAV) minus market price divided by NAV times 100.
The data on market prices and NAVs are obtained from Datastream and Oslo Bors. TURN is defined as the natural log of turnover. Turnover is the ratio of total value of shares
traded during the period divided by the average market capitalization for the period. Average market capitalization is calculated as the average of the end-of-period values for the
current period and the previous period. NIPO is the annual number of initial public offerings obtained from OBI. RIPO is measured annually and is defined as the average first-day
returns of initial public offerings. RIPO is calculated as the difference between the first trading price and the offer price divided by the offer price. ES is measured as the proportion
of aggregate equity issuance in aggregate equity and debt issuance published annually by Oslo Bors. PD-ND is calculated at the year-end log ratio of the equal-weighted average
market-to-book ratios after sorting all securities on OBX into payers and nonpayers using raw data from Datastream (Baker and Wurgler 2004). TURN, RIPO, PD-ND and VIX are
lagged 1 year relative to the other four measures. VIX is the natural log of the Chicago Board Options Exchange Market Volatility Index, which is measures the implied volatility of
S&P500. CCI is the natural log of the yearly Consumer confidence index, which is based on survey data on consumer’s confidence. SENTIMENT is the first principal component of
the eight sentiment proxies. In panel B, each of the eight proxies is regressed on CPI, IPI, GDP, and Central Bank’s key policy rate. The orthogonalized proxies, labeled with a “⊥,”
are the residuals from these regressions. SENTIMENT⊥ is the first principal component of the eight orthogonalized proxies.
Correlations with Sentiment Correlations with Sentiment Components

Mean Median SD Min Max SENTIMENT SENTIMENT⊥ CEFD TURN NIPO RIPO ES PD−ND VIX CCI
Panel A: Raw Data
CEFDt -0.24 -0.02 0.61 -1.22 0.69 -0.22 -0.31 1
TURNt−1 4.39 4.41 0.45 3.32 5.03 0.97 0.91 -0.18 1
NIPOt 22.64 20.50 15.40 3.00 59.00 1.00 0.93 -0.22 0.96 1
RIPOt−1 -0.50 -0.06 1.16 -3.17 1.06 -0.64 -0.44 0.21 -0.74 -0.59 1
ESt 0.05 0.03 0.04 0.01 0.13 -0.40 -0.26 -0.63 -0.52 -0.38 0.54 1
PD−NDt-1 -0.14 -0.19 0.32 -0.45 1.14 0.92 0.82 -0.20 0.89 0.92 -0.60 -0.45 1
VIXt-1 2.93 2.96 0.34 2.45 3.69 -0.86 -0.63 -0.16 -0.87 -0.84 0.75 0.70 -0.80 1
CCI 2.88 3.07 0.58 1.62 3.52 0.84 0.63 -0.02 0.77 0.83 -0.51 -0.35 0.78 -0.88 1
Panel B: Controlling for Macroeconomic Conditions
CEFD⊥t 3.1E-15 -0.02 0.35 -0.84 0.69 -0.19 -0.06 1
TURN⊥t−1 2.2E-16 0.00 0.27 -0.64 0.44 0.95 0.83 0 1.00
NIPO⊥t 2.8E-14 -1.46 11.50 -17.47 25.01 0.91 1.00 -0.08 0.82 1
RIPO⊥t−1 2.8E-15 0.32 1.01 -2.21 1.07 -0.28 -0.23 0.46 -0.24 -0.26 1
ES⊥t -3E-16 0.00 0.02 -0.05 0.03 -0.06 0.00 -0.66 0.06 0.00 0.27 1
PD−ND⊥t-1 -2E-16 -0.06 0.30 -0.30 1.23 0.89 0.70 -0.19 0.84 0.69 -0.09 -0.02 1
VIX⊥t-1 -1E-16 -0.04 0.25 -0.42 0.58 -0.71 -0.64 -0.20 -0.65 -0.65 0.53 0.67 -0.53 1
CCI⊥ -5E-16 0.15 0.46 -1.22 0.60 0.62 0.77 0.25 0.53 0.77 -0.29 -0.29 0.45 -0.68 1
Page 21
Table IIB
Investor Sentiment Data, Vietnam, 2000-2013
This table presents the means, standard deviations, and correlations for measures of investor sentiment. Panel A reports the raw sentiment proxies. Closed-end mutual funds (CEFD)
are defined as the year-end, value-weighted average discount on closed-end mutual funds. It is calculated as net asset values (NAV) minus market price divided by NAV times 100.
The data on market prices and NAVs are obtained from DataStream and HOSE. TURN is defined as the natural log of turnover. Turnover is the ratio of total value of shares traded
during the period divided by the average market capitalization for the period. Average market capitalization is calculated as the average of the end-of-period values for the current
period and the previous period. NIPO is the annual number of initial public offerings obtained from HOSE. RIPO is measured annually and is defined as the average first-day returns
of initial public offerings. CEFD,TURN, NIPO, and ES are lagged 1 year relative to the other two measures. VIX is the natural log of the Chicago Board Options Exchange Market
Volatility Index, which is measures the implied volatility of S&P500. CCI is the natural log of the yearly Consumer confidence index, which is based on survey data on consumer’s
confidence. SENTIMENT is the first principal component of the six sentiment proxies. In panel B, each of the six proxies is regressed on CPI, IPI, GDP, and Central Bank’s key
policy rate. The orthogonalized proxies, labeled with a “⊥,” are the residuals from these regressions. SENTIMENT⊥ is the second principal component of the six orthogonalized
proxies.
Correlations with Sentiment Correlations with Sentiment Components
Mean Median SD Min Max SENTIMENT SENTIMENT⊥ CEFD TURN NIPO ES VIX CCI
Panel A: Raw Data
CEFDt-1 0.40 0.40 0.10 0.26 0.52 -0.82 -0.76 1
TURNt−1 4.21 4.42 0.61 3.39 4.95 0.35 0.75 -0.60 1
NIPOt−1 13.80 9.00 10.18 6.00 31.00 1.00 0.79 -0.84 0.40 1
Est-1 0.79 0.89 0.22 0.40 0.94 -0.92 -0.50 0.68 -0.24 -0.90 1
VIXt 3.14 3.08 0.33 2.88 3.69 0.98 0.67 -0.69 0.23 0.97 -0.94 1
CCIt 4.57 4.58 0.08 4.44 4.66 0.66 0.75 -0.96 0.66 0.69 -0.46 0.49 1
Panel B: Controlling for Macroeconomic Conditions
CEFD⊥t-1 -1E-16 -0.004 0.17 -0.26 0.32 -0.33 -0.40 1
TURN⊥t−1 -3E-15 0.01 0.52 -0.75 0.63 0.42 0.57 0.51 1
NIPO⊥t-1 -9E-14 2.05 5.90 -7.62 5.40 0.75 1.00 -0.46 0.51 1
ES⊥t-1 -4E-16 -0.003 0.12 -0.17 0.24 -0.34 -0.44 1.00 0.49 -0.50 1
VIX⊥t -5E-16 0.0033 0.29 -0.40 0.44 0.76 0.92 -0.19 0.72 0.89 -0.20 1
CCI⊥t 5.7E-16 -0.005 0.05 -0.06 0.10 0.21 0.13 -0.69 -0.38 0.16 -0.62 0.24 1
Page 22
Panel A: Closed-End Fund Discount (%) Panel F: Dividend Premium

80% 80% 1.25 1.25
40% 40% 1.00 1.00

0.75 0.75
0% 0%
0.50 0.50
-40% -40%
0.25 0.25
-80% -80% 0.00 0.00
-120% -120% -0.25 -0.25
-160% -160% -0.50 -0.50
1996 1998 2000 2002 2004 2006 1995 2000 2005 2010
Panel B: Turnover Panel G: VIX

2.0 1.0
1.5
0.5
1.0 4.0
6 0.5 0.0
0.0 3.6
5 -0.5
-0.5 3.2
4 -1.0 -1.0
2.8
3
2.4
2 1995 2000 2005 2010
1995 2000 2005 2010
Panel H: Consumer confidence index

Panel C: Number of IPOs 4 4
50 50 3 3
40 40 2 2
30 30
1 1
20 20
10 10 0 0
0 0 -1 -1
-10 -10 -2 -2
1995 2000 2005 2010
1995 2000 2005 2010
Panel I: Sentiment Index (SENTIMENT)

Panel D: Average first-day return of IPOs 8
2% 2%
1% 1% 4
14
0% 0% 12 0
10
-1% -1% 8 -4
-2% -2% 6
-8
4
-3% -3%
2
-4% -4% 0
2000 2002 2004 2006 2008 2010 2012 2000 2002 2004 2006
Panel E: Equity share in new issues

.16 .16
.12 .12
.08 .08
.04 .04
.00 .00
-.04 -.04
-.08 -.08
2000 2002 2004 2006 2008 2010 2012
Figure 1A. Investor sentiment, Norway, 1991–2013. Panel A presents the year-end, value-weighted average discount on
closed-end mutual funds. The data on market prices and NAVs are obtained from DataStream and Oslo Bors. Panel B plots
the log turnover. Turnover is the ratio of total value of shares traded during the period divided by the average market
capitalization for the period. Panel C presents the annual number of initial public offerings obtained from OBI. Panel D
presents the average annual first-day returns of initial public offerings. Panel E presents as the proportion of aggregate
equity issuance in aggregate equity and debt issuance published annually by Oslo Bors. Panel F presents the year-end log
ratio of the equal-weighted average market-to-book ratios after sorting all securities on OBX into payers and nonpayers
using raw data from DataStream (Baker and Wurgler 2004). Panel G presents the natural log of the Chicago Board Options
Exchange Market Volatility Index (VIX), which measures the implied volatility of S&P500. Panel H presents the natural
log of the yearly Consumer confidence index, which is based on survey data on consumer’s confidence. The blue line (left
axis) is raw data. Each measure are regressed on the CPI, IPI, GDP, and Central Bank’s key policy rate. The green line
(right axis) is the residuals from this regression. The blue line in the final panel is the first principal component of the eight
sentiment raw proxies. The green line in the final panel is the first principal component of the eight orthogonalized proxies.
Both are standardized to have unit variance. In these two indices, turnover, the average annual first-day return, the dividend
premium, and VIX are lagged 1 year relative to the other four proxies.
Page 23
Panel A: Closed-End Fund Discount (%) Panel E: VIX

40% .6
20% .4
4.0 .2
60%
0%
.0
40% 3.6
-20% -.2
20% 3.2 -.4
-40%
0% -.6
2.8
-20% 2.4
2006 2008 2010 2012 2000 2002 2004 2006 2008 2010 2012
Panel B: Turnover
2
Panel F: Consumer confidence index
1.0
1
6 0.5
5.0
0
5 0.0
4.8
-1
4 4.6 -0.5
-2
3 4.4
-1.0
4.2
2
2006 2008 2010 2012 4.0
2006 2008 2010 2012
Panel C: Number of IPOs
40 40
Panel G: Sentiment Index (SENTIMENT)
30 30
6
20 20 4
2
10 10 10 0
8 -2
0 0
-4
6
-10 -10 -6
2008 2009 2010 2011 2012 2013 4
2
Panel D: Equity share in new issues 0
2008 2009 2010 2011 2012
1.0
0.5
1.0 0.0
0.5 -0.5
0.0 -1.0
-0.5
-1.0
2006 2008 2010 2012
Figure 1B. Investor sentiment, Vietnam, 2000–2013. Panel A presents the year-end, value-
weighted average discount on closed-end mutual funds. The data on market prices and NAVs are
obtained from DataStream and HOSE. Panel B plots the log turnover. Turnover is the ratio of total
value of shares traded during the period divided by the average market capitalization for the
period. Panel C presents the annual number of initial public offerings obtained from HOSE. Panel
D presents as the proportion of aggregate equity issuance in aggregate equity and debt issuance
published annually by HOSE. Panel E presents the natural log of the Chicago Board Options
Exchange Market Volatility Index (VIX), which measures the implied volatility of S&P500. Panel
F presents the natural log of the yearly Consumer confidence index, which is based on survey data
on consumer’s confidence, published quarterly by the Nielsen company. The blue line (left axis) is
raw data. Each measure are regressed on the CPI, IPI, GDP, and Central Bank’s key policy rate.
The green line (right axis) is the residuals from this regression. The blue line in the final panel is
the first principal component of the six sentiment raw proxies. The green line in the final panel is
the first principal component of the six orthogonalized proxies. Both are standardized to have unit
variance. In these two indices, closed-end funds discount, turnover, the annual number of initial
public offerings, and the share of equity issuance are lagged 1 year relative to the other two
proxies.
Page 24
4. Empirical Tests
A. Sorting
Table III presents 10 equally-weighted portfolios according to their characteristics

of firm size (ME), total risk (σ), earnings-to-book ratio (E/BE), dividend-to-book
ratio (D/BE), asset tangibility (PPE/A), R&D over assets (RD/A), book-to-market
ratio (BE/ME), and external finance over assets (EF/A) and sales growth (GS). At
the beginning of the month, each monthly return variable is grouped into the
decile rank that the characteristic takes, and then according to the level of
SENTIMENT⊥ at the end of the previous year. Decile 10 represents the largest
values of each characteristic whereas decile 1 represents the smallest. For each
section, the equally-weighted average monthly return is calculated. Furthermore,
each decile portfolio is divided into a positive sentiment group and a negative
sentiment group, in which the difference between these two groups are computed
in order to look for patterns.
In Table IIIA, The ME panel represents the size effect conditional on
sentiment. The results for Norway are in line with Baker and Wurgler’s (2006)
findings that the size effect is apparent when sentiment is low. Particularly, when
sentiment⊥ is negative, the average return is 6.85% per month for decile 1 and
2,30% for decile 10. When sentiment is positive, the average return is -1.55% per
month for decile 1 and -1.31% for decile 10. The difference between the positive
and negative sentiment supports Baker and Wurgler’s (2006) claim that small
stocks are more affected by sentiment, in which the sentiment-return relation is
negative. Vietnam’s results in Table IIIB are also consistent with the authors’
findings. When sentiment is pessimistic, the average return is 2.90% per month
for the bottom decile and 0.45% for the top decile. However, here the size effect is
also apparent when sentiment is high because the average return is 0.65% per
month for the bottom decile and -1.1% for the top decile. Overall, when sentiment
is low, future returns are relatively high for small stocks in both Norway and
Vietnam. On the contrary, when sentiment is high, the size effect is only evident
in Vietnam.
The σ panel shows that the cross-sectional effect of return volatility in
Norway is conditional on sentiment. Again, Norway’s results are consistent with
Baker and Wurlger’s findings. When sentiment is positive, high sigma stocks earn
Page 25
lower returns (-2.99%). When sentiment is negative, they earn higher returns
(7.15%). Intuitively, “riskier” stocks are hard to value and difficult to arbitrage,
which makes them particularly prone to sentiment fluctuations. Figure 2a and 2b
are graphical representations of the results from Table IIIA and Table IIIB. Panel
B: σ documents the unconditional average monthly returns across σ deciles (green
line), which is basically flat; the average monthly return when sentiment is
positive (blue bar), which is declining with risk decile; the average monthly return
when sentiment is negative (purple bar), which rises with risk deciles; and the
difference in conditional returns (red line). The red line clearly shows that riskier
stocks’ future returns are more sensitive to sentiment. However, this is not fully
the case in Vietnam. When sentiment is high, high sigma stocks earn lower
returns (-3.8%). But when sentiment is low, they do not earn higher returns. In
fact, they earn lower returns (-0.55%), which means that the sentiment effect on
high volatility stocks are only apparent when sentiment is high in Vietnam.
E/BE represents the profitability panel, while D/BE represents the dividend
panel. Consistent with Baker and Wurgler’s findings, when sentiment is high, next
year’s monthly returns are lower on unprofitable firms (-4.50% and -4.20%) than
profitable firms in both Norway and Vietnam, respectively. However, when
sentiment is low, unprofitable firms do not earn higher returns. This suggests that
the results on profitability characteristics are only consistent with Baker and
Wurgler’s findings when the sentiment is low. In terms of the dividend
characteristic, nonpayers tend to earn relatively lower (higher) returns when the
sentiment is high (low) in Norway. In Vietnam, this is only consistent when
sentiment is high.
PPE/A represents the tangibility panel under the notion that firms with less
tangible assets, i.e. less PPE/A, are harder to value because they have more
intangible assets, and are therefore more sensitive to sentiment fluctuations. The
results from Norway are consistent with prior theory. Specifically, when the
sentiment is high, the average returns on the bottom decile are lower than the top
decile, i.e. -1.98% versus 0.22%. Whereas, when the sentiment is low, the former
is higher than the latter, i.e. 4.54% versus 3.82%. However, the Vietnamese
results conform to Baker and Wurgler’s (2006) findings only when the sentiment
is high. The results for the RD/A panel are ambiguous mainly because of the lack
of data for both countries.
Page 26
The book-to-market variable shows intriguing patterns, in which it has

some explanatory power. Consistent with Baker and Wurgler (2006), future
returns are higher for high BE/ME stocks in both Norway and Vietnam. However,
the results are inconsistent in terms of the EF/A stocks in Norway. In Vietnam,
however, when sentiment is low, low EF/A stocks have generally higher returns
than high EF/A stocks. Due to lack of data, the GS variable could not be
investigated. Appendix 3 compares and summarizes the results.
In general, consistent with Baker and Wurgler’s (2006) statement, when
beginning-of-period proxies for sentiment are low, future returns are high for
small stocks, high volatility stocks, non-dividend-paying stocks, and value stocks.
Page 27
Table IIIA
Future Returns by Sentiment Index and Firm Characteristics, Norway, 1991-2013
Table III presents the average monthly returns of portfolios sorted by the firm characteristic and the sentiment index. 10 equally-weighted portfolios for each month are established
according to their characteristics of firm size (ME), total risk (σ), earnings-to-book ratio (E/BE), dividend-to-book ratio (D/BE), asset tangibility (PPE/A), book-to-market ratio
(BE/ME), and external finance over assets (EF/A). Decile 1 represents the smallest whereas decile 10 represents the largest values of each characteristic. Furthermore, according to
the level of SENTIMENT⊥ at the end of the previous year, each decile portfolio is divided into a positive sentiment group and a negative sentiment group. The average returns for
each portfolio as well as the difference of average returns between two groups in the same portfolio, i.e. positive and negative sentiment groups, are then calculated. Portfolio returns
for unprofitable firms and nonpayers are also computed. SENTIMENT⊥ is positive for 2000-2001 and 2005-2007.
Decile Comparisons
SENTIMENT⊥t-1 ≤0 1 2 3 4 5 6 7 8 9 10 10 1 10 5 51
ME Positive -1.55 -0.80 -2.67 -1.23 -1.75 -1.30 -1.03 -2.37 -1.18 -1.31 0.24 0.44 -0.20
Negative 6.85 4.87 3.80 3.32 3.84 3.26 3.64 4.54 3.42 2.30 -4.55 -1.54 -3.00
Difference -8.39 -5.66 -6.47 -4.54 -5.59 -4.56 -4.67 -6.91 -4.60 -3.61 4.79 1.99 2.80
σ Positive -0.52 -1.87 -0.91 -1.29 -1.21 -1.48 -1.74 -1.10 -3.30 -3.00 -2.48 -1.79 -0.69
Negative 2.49 2.69 3.20 4.51 3.46 3.37 4.26 3.49 4.56 7.15 4.66 3.70 0.97
Difference -3.01 -4.56 -4.11 -5.81 -4.67 -4.86 -6.01 -4.59 -7.87 -10.15 -7.14 -5.49 -1.66
E/BE Positive -4.50 -1.05 -1.46 -0.84 -0.72 -1.00 0.21 0.53 0.73 1.16 2.97 4.02 3.97 0.05
Negative 3.41 3.87 2.56 2.64 2.58 3.90 3.42 5.80 4.00 5.17 5.13 1.26 1.23 0.03
Difference -7.92 -4.92 -4.02 -3.48 -3.30 -4.90 -3.22 -5.28 -3.27 -4.01 -2.16 2.76 2.74 0.02
D/BE Positive -2.23 -0.22 -0.72 -0.56 -0.54 -0.14 -0.52 -0.47 -0.60 -1.24 -1.34 -1.12 -1.20 0.08
Negative 5.01 3.45 3.20 3.46 3.62 2.66 2.33 2.95 2.19 3.80 3.19 -0.27 0.52 -0.79
Difference -7.24 -3.68 -3.93 -4.02 -4.16 -2.81 -2.85 -3.42 -2.78 -5.04 -4.53 -0.85 -1.72 0.87
PPE/A Positive -1.98 -1.86 -2.75 -1.72 -1.67 -2.11 -1.60 -1.16 -1.76 0.21 2.20 1.88 0.32
Negative 4.53 3.76 4.81 2.98 2.96 4.05 5.00 5.03 4.21 3.82 -0.72 0.86 -1.58
Difference -6.52 -5.63 -7.56 -4.70 -4.62 -6.16 -6.61 -6.19 -5.97 -3.60 2.92 1.02 1.90
BE/ME Positive -8.52 -2.03 -2.12 -2.13 -1.29 -1.35 -0.50 0.17 -0.44 0.04 8.55 1.33 7.22
Negative -3.27 3.83 2.57 4.15 3.61 3.89 4.53 4.39 5.88 6.06 9.33 2.45 6.88
Difference -5.25 -5.86 -4.69 -6.28 -4.91 -5.24 -5.03 -4.22 -6.32 -6.02 -0.77 -1.12 0.35
EF/A Positive -2.01 -1.15 -1.86 -1.88 -2.22 -0.97 -1.62 -1.79 -2.03 -0.46 1.55 1.76 -0.21
Negative 3.94 4.16 3.36 1.89 4.42 3.81 3.47 3.06 4.69 6.01 2.07 1.59 0.48
Difference -5.95 -5.31 -5.21 -3.77 -6.64 -4.77 -5.08 -4.84 -6.72 -6.47 -0.52 0.17 -0.69
Page 28
Table IIIB
Future Returns by Sentiment Index and Firm Characteristics, Vietnam, 2000-2013
Table III presents the average monthly returns of portfolios sorted by the firm characteristic and the sentiment index. 10 equally-weighted portfolios for each month are established
according to their characteristics of firm size (ME), total risk (σ), earnings-to-book ratio (E/BE), dividend-to-book ratio (D/BE), asset tangibility (PPE/A), book-to-market ratio
(BE/ME), and external finance over assets (EF/A). Decile 1 represents the smallest whereas decile 10 represents the largest values of each characteristic. Furthermore, according to
the level of SENTIMENT⊥ at the end of the previous year, each decile portfolio is divided into a positive sentiment group and a negative sentiment group. The average returns for
each portfolio as well as the difference of average returns between two groups in the same portfolio, i.e. positive and negative sentiment groups, are then calculated. Portfolio returns
for unprofitable firms and nonpayers are also computed. SENTIMENT ⊥ is positive for 2008 and 2010-2011.
Decile Comparisons
SENTIMENT⊥t-1 ≤0 1 2 3 4 5 6 7 8 9 10 10 1 10 5 51
ME Positive 0.65 -1.16 -0.61 -0.41 -1.59 -1.08 -2.51 -1.53 -1.54 -1.10 -1.75 0.49 -2.25
Negative 2.91 1.12 2.04 1.13 0.87 0.70 0.49 -0.01 -0.80 0.45 -2.47 -0.42 -2.05
Difference -2.26 -2.28 -2.65 -1.54 -2.46 -1.78 -3.00 -1.52 -0.73 -1.55 0.71 0.91 -0.20
σ Positive -0.30 -0.12 -1.25 -1.12 -0.49 -2.49 -1.57 -0.11 -3.66 -3.80 -3.49 -3.31 -0.18
Negative 2.54 3.16 4.43 3.07 2.59 0.91 0.88 0.60 -0.28 -0.55 -3.08 -3.13 0.05
Difference -2.84 -3.27 -5.68 -4.19 -3.07 -3.39 -2.45 -0.70 -3.39 -3.25 -0.41 -0.18 -0.24
E/BE Positive -4.19 -3.48 -1.77 -1.72 -0.45 0.26 0.24 2.33 1.33 2.81 0.79 4.27 0.53 3.74
Negative -4.04 -3.23 -2.02 -2.45 -1.61 -2.80 -1.92 -1.66 -1.82 0.40 -1.02 2.21 1.78 0.43
Difference -0.16 -0.25 0.26 0.73 1.16 3.06 2.16 4.00 3.15 2.42 1.81 2.06 -1.25 3.31
D/BE Positive -3.75 -2.23 -1.66 -0.68 -0.83 0.35 0.26 -1.03 -0.86 0.08 1.62 3.85 1.27 2.57
Negative -2.76 -1.54 -1.64 -1.24 -1.77 -1.82 -1.47 -1.44 -2.05 -0.84 -2.35 -0.82 -0.53 -0.28
Difference -0.99 -0.69 -0.02 0.55 0.93 2.17 1.72 0.41 1.18 0.92 3.97 4.66 1.81 2.86
PPE/A Positive -3.02 -1.94 -1.50 -1.31 -0.02 -0.93 -0.88 1.08 -1.67 -0.32 2.71 -0.29 3.00
Negative -3.25 -2.36 -1.64 -0.91 -1.22 -1.38 -1.80 -1.32 -2.38 -2.03 1.22 -0.81 2.03
Difference 0.23 0.42 0.14 -0.41 1.20 0.45 0.92 2.41 0.71 1.72 1.49 0.52 0.97
BE/ME Positive -3.55 -3.24 -2.93 -2.61 -2.32 -0.92 -1.48 -0.96 0.47 1.54 5.09 3.87 1.23
Negative -3.25 -2.03 -2.13 -0.76 -2.31 -0.87 -0.86 -0.43 1.42 3.87 7.12 6.18 0.94
Difference -0.30 -1.21 -0.81 -1.84 -0.01 -0.05 -0.62 -0.53 -0.95 -2.32 -2.03 -2.31 0.29
EF/A Positive -0.14 -0.64 -0.55 -0.90 -1.23 -1.11 -3.99 -1.12 -1.72 0.48 0.62 1.71 -1.09
Negative -1.11 -2.02 -1.39 -1.99 -1.72 -2.27 -1.71 -1.15 -3.08 -1.93 -0.82 -0.21 -0.61
Difference 0.97 1.38 0.84 1.09 0.49 1.16 -2.28 0.03 1.36 2.40 1.43 1.91 -0.48
Page 29
Panel A: ME Panel E: PPE/A

10 10
5
0 0
-5
-10 -10
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Panel B: σ Panel F: BE/ME

10
0 0
-10 -10
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Panel C: E/BE Panel G: EF/A

10 10
5
0 0
-5
-10 -10
≤0 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Panel D: D/BE
10
5
0
-5
-10
≤0 1 2 3 4 5 6 7 8 9 10
Figure 2A. Sorting approach: Future returns by sentiment index and firm
characteristics, Norway, 1963–2001. 10 portfolios are formed based on firm
characteristics of firm size (ME), total risk, earnings-book ratio (E/BE), dividend-book
ratio (D/BE), fixed assets (PPE/A), book-to-market ratio (BE/ME), and external finance
over assets (EF/A). Portfolio returns are computed for unprofitable, and nonpaying firms.
Returns following positive SENTIMENT⊥ periods are represented by the blue bars, and
returns following negative sentiment periods are presented by the purple bars. The red
line is the difference between both periods and the green line is the average.
SENTIMENT⊥ is positive for 2000-2001 and 2005-2007.
Page 30
Panel A: ME Panel E: PPE/A

5 5
0 0
-5 -5
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Panel B: σ Panel F BE/ME

5 5
0
0
-5
-10 -5
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Panel C: E/BE Panel G: EF/A

5 5
0 0
-5 -5
≤0 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Panel D: D/BE
5
-5
≤0 1 2 3 4 5 6 7 8 9 10
Figure 2B. Sorting Approach: Future returns by sentiment index and firm
characteristics, Vietnam, 2000-2013. 10 portfolios are formed based on firm
characteristics of firm size (ME), total risk, earnings-book ratio (E/BE), dividend-book
ratio (D/BE), fixed assets (PPE/A), book-to-market ratio (BE/ME), and external finance
over assets (EF/A). Portfolio returns are computed for unprofitable, and nonpaying firms.
Returns following positive SENTIMENT⊥ periods are represented by the blue bars, and
returns following negative sentiment periods are presented by the purple bars. The red
line is the difference between both periods and the green line is the average.
SENTIMENT⊥ is positive for 2008 and 2010-2011.
Page 31
B. Long–Short Portfolios Regressions
An alternative method that also looks for conditional characteristics effects is the
method that forecasts the equally-weighted portfolios using the sentiment. These
portfolios are long on stocks with high characteristic values, i.e. the three top
deciles, and short on stocks with low values, i.e. the bottom three deciles. Medium
is defined as the four middle deciles. This regression approach is of special
interest because it permits us to conclude which characteristics have conditional
effects that are different from recognized unconditional effects, incorporate the
sentiment indexes’ continuous nature, and perform formal significance tests.
Table IVA and IVB reports the correlations among characteristics-based
long- short portfolios, in which the samples period includes the average monthly
returns.
Following Baker and Wurgler (2006), the growth and distress variables are
also divided into portfolios of high-minus-medium and medium-minus-low. This
is conducted because a simple high-minus-low analysis of the variables would
omit important cross-sectional aspects. For instance, in Norway, the portfolios of
the BE/ME variable are negatively correlated with each other at −0.20, suggesting
that high and low BE/ME firms move together relative to middle BE/ME firms.
Similarly, the medium-minus-low correlation of the variable EF/A is −0.31. In
Vietnam, however, only the correlation of the portfolios formed according to the
variable BE/ME is negative at -0.42. The portfolios of the variable EF/A are
positively correlated at 0.40.
In order to test whether sentiment can predict the several long–short
portfolios created in Table IVA and IVB, the following regressions have been
conducted:
RXit=High,t − RXit=Low,t = α1 + βSENTIMENTt−1 + εit.
The dependent variable, RXit=High,t − RXit=Low,t, is the long–short portfolio’s

monthly return, such as the size effect portfolio (SMB). The monthly returns are
regressed on the sentiment index, which is lagged one year (SENTIMENTt-1).
Also, a multivariate regression is conducted in order to differentiate new
predictability effects from well-known co-movement:
RXit=High,t − RXit=Low,t = α2+ β1SENTIMENTt−1 + β2RMRFt + β3SMBt + β4HMLt + β5UMDt + εit.
Employing the definitions introduced by Fama and French (1993), RMRF is

defined as the excess return of the value-weighted market less the risk-free rate.
This factor controls for the correlation between the returns in each portfolio of
individual stocks and the market portfolio returns. In particular, the market
portfolio returns for Norway and Vietnam are computed from the Total return
index (RI) of OBX SE and VNINDEX. The official risk-free rate for Norway is
the 3-month interbank rate. The risk-free rate for Vietnam is base interest rate. All
data is obtained from DataStream. The SMB variable is calculated as the
difference between the returns on small ME portfolios and big ME portfolios.
HML is computed as high minus low BE/ME stocks. When SML or HML are the
Page 32
portfolios being forecasted, they are excluded from the right hand side. The
additional factor, UMD, is the return on high-minus-low momentum stocks. This
factor is calculated as the cumulative return for the 11-month period between 12
and 2 months prior to the given month.
Tables VA and VB present the results from the regressions of the portfolio
returns. The size column in Norway demonstrates that when sentiment is high,
next year’s returns on small stocks are relatively low, and vice versa. For
example, the SMB coefficient suggests that a one-unit increase in sentiment results
in a lower monthly return of −0.43% on the SMB portfolio. However, inconsistent
with Baker and Wurgler’s (2006) findings, the sentiment coefficient increases
after controlling for RMRF, SMB, HML, and UMD. This means that the
forecasting power of sentiment on stock returns increases after controlling for
RMRF, SMB, HML, and UMD. The results for Vietnam show that when the
sentiment is high, the returns on high volatility stocks next year tend to be lower,
and vice versa. However, this effect is only apparent for the sentiment index that
is not orthogonalized to macroeconomic conditions, ie. The SENTIMENT index
(2). However, the size effect is evident in SENTIMENT⊥ after controlling for
RMRF, SMB, HML, and UMD (-0.60%).
Inconsistent with Baker and Wurgler’s (2006) results, the coefficients on
SENTIMENT and SENTIMENT⊥ for both Norway and Vietnam are not similar.
This implies that macroeconomic conditions play a major role in both countries.
In terms of the profitability and dividend payment, regressions are run in
order to forecast the difference between the portfolios that are profitable and
dividend paying and the portfolios that are unprofitable and nonpaying,
respectively. This is because the sorting approach implied that these are expected
to capture the key contrasts. In Norway, higher sentiment can forecast lower
returns on unprofitable stocks. For example, the coefficient of profitability
characteristic suggests that a one-unit increase in sentiment results in a lower
monthly return of −0.31% on the profitability portfolio. However, the sentiment
index in Vietnam does not have a significant forecasting power on these
portfolios.
The remaining portfolios show no significant relationship between
sentiment and subsequent stock returns.
Since the size effect in Vietnam is only significant at 12%, the sample
period is shortened in order to test whether the findings are driven by an overall
trend. The subsample period is from 2008 to 2012. Table VI reports that the size
effect is more significant at the 11.5% level, and has substantially stronger impact
on stock returns than the entire sample.
To summarize, the regressions from Table V partially confirm the
significance of the patterns proposed in the sorting approach. In Norway, when
sentiment is high, subsequent returns are relatively low for small firms and
unprofitable firms. In Vietnam, when sentiment is high, subsequent returns are
relatively low for small firms and firms with volatile stock returns. And vice-
versa. Generally, the results are consistent with the statement that sentiment has
stronger effects on stocks that are hard to value and difficult to arbitrage.
Page 33
Table IVA
Correlations of Portfolio Returns in Norway, 1991-2013
This table presents the correlations among portfolios that are characteristics-based. The sample
period comprises of monthly returns from 1991 to 2013. The long–short portfolios are designed
according to firm characteristics: firm size (ME), age, profitability (E), dividends (D), fixed assets
(PPE), research and development (RD), book-to-market ratio (BE/ME), and external finance over
assets (EF/A). High describes a stock in the top three deciles, low describes a stock in the bottom
three deciles, and medium describes a stock in the middle four deciles.
Growth
Profitability,
Opportunities Growth
Size, Risk Dividends Tangibility & Distress Opportunities Distress
ME σ E D PPE/A RD/A BE/ME EF/A BE/ME EF/A BE/ME EF/A
ME SMB 1.00
σ High-Low 0.07 1.00
E >0 - <0 -0.34 -0.48 1.00
D >0 - =0 -0.38 -0.62 0.76 1.00
PPE/A High-Low 0.03 -0.17 0.35 0.19 1.00
RD/A High-Low -0.32 0.15 0.05 0.03 -0.07 1.00
BE/ME HML -0.22 -0.17 0.43 0.32 0.14 0.40 1.00
EF/A High-Low -0.05 0.37 -0.32 -0.32 -0.19 0.22 0.11 1.00
BE/ME Medium-Low -0.39 -0.08 0.39 0.30 0.16 0.48 0.86 0.21 1.00
EF/A High-Medium -0.01 0.33 -0.31 -0.35 -0.19 0.04 -0.12 0.82 -0.02 1.00
BE/ME High-Medium 0.30 -0.18 0.11 0.05 -0.03 -0.12 0.34 -0.18 -0.20 -0.19 1.00
EF/A Medium-Low -0.08 0.06 -0.02 0.05 0.01 0.29 0.36 0.29 0.37 -0.31 0.03 1.00
Table IVB
Correlations of Portfolio Returns in Vietnam, 2000-2013
This table presents the correlations among portfolios that are characteristics-based. The sample
period comprises of monthly returns from 2000 to 2013. The long–short portfolios are designed
according to firm characteristics: firm size (ME), age, profitability (E), dividends (D), fixed assets
(PPE), research and development (RD), book-to-market ratio (BE/ME), and external finance over
assets (EF/A). High describes a stock in the top three deciles, low describes a stock in the bottom
three deciles, and medium describes a stock in the middle four deciles.
Growth
Profitability,
ME σ E D PPE/A BE/ME EF/A GS BE/ME EF/A BE/ME EF/A
ME SMB 1.00
σ High-Low 0.78 1.00
E >0 - <0 -0.02 -0.38 1.00
D >0 - =0 -0.39 -0.49 0.88 1.00
PPE/A High-Low 0.42 0.50 0.52 0.50 1.00
BE/ME HML 0.35 0.46 0.16 0.05 0.43 1.00
EF/A High-Low -0.16 0.06 -0.90 -0.83 -0.80 -0.11 1.00
GS High-Low 0.18 0.35 -0.45 -0.53 -0.27 0.72 0.60 1.00
BE/ME Medium-Low -0.29 -0.10 0.57 0.73 0.56 0.63 -0.52 0.11 1.00
EF/A High-Medium -0.57 -0.05 -0.62 -0.30 -0.45 0.18 0.70 0.58 0.18 1.00
BE/ME High-Medium 0.75 0.66 -0.48 -0.80 -0.15 0.43 0.48 0.70 -0.42 0.00 1.00
EF/A Medium-Low 0.08 0.11 -0.84 -0.92 -0.80 -0.24 0.93 0.48 -0.77 0.40 0.62 1.00
Page 34
Table VA1 and VA2

Time Series Regressions of Portfolio Returns, Norway, 1991 to 2013
This table represents the regressions of long–short portfolio monthly returns on the lagged
SENTIMENT index, the market risk premium (RMRF), the Fama–French factors (HML and
SMB), and a momentum factor (UMD) in Norway.
RXit=High,t − RXit=Low,t = α1 + βSENTIMENTi,t−1 + εit.
RXit=High,t − RXit=Low,t = α2+ β1SENTIMENTi, t−1 + β2RMRFt + β3SMBt + β4HMLt + β5UMDt + εit.
The long–short portfolios are sorted based on firm characteristics: firm size (ME), total risk (σ),
profitability (E), dividends (D), asset tangibility (PPE/A), book-to-market ratio (BE/ME), and
external finance over assets (EF/A). These portfolios are long on stocks with high characteristic
values, i.e. the three top deciles, and short on stocks with low values, i.e. the bottom three deciles.
Medium is defined as the four middle deciles. Average monthly portfolio returns are matched to
the lagged SENTIMENT index for the previous calendar year before controlling for
macroeconomic conditions. The first panel presents the results of univariate regressions, whereas
the second panel present the results of multivariate regressions after controlling for RMRF, HML,
SMD, and UMD. SMB and HML are excluded from the control variables when they are regressed.
Coefficients for each variable are presented first and p-values are presented in brackets.
Growth
Profitability,
High- High- Med- High-
SMB High-Low >0 - <0 >0 - =0 Low Low HML High-Low Low Med High-Med Med-Low
Rxit=High,t - Rxit=Low,t = α1 + βSentimentit-1 + εit
1.72 -0.33 4.11 0.17 1.76 -1.42 2.60 1.96 0.16 1.54 2.44 0.42
Intercept α1
(0.27) (0.86) (0.00) (0.89) (0.15) (0.82) (0.17) (0.08) (0.93) (0.13) (0.01) (0.55)
-0.19 0.14 -0.22 -0.04 -0.13 -0.20 -0.02 -0.16 0.12 -0.10 -0.13 -0.07
Sentiment β
(0.38) (0.58) (0.21) (0.82) (0.43) (0.77) (0.95) (0.28) (0.67) (0.49) (0.29) (0.48)
Rxit=High,t - Rxit=Low,t = α2 + β1Sentimentit-1 + β2RMKT + β3SMB + β4HML + β5UMD + εit
2.15 1.51 3.59 -0.78 0.97 -0.48 3.74 1.46 -0.43 1.63 0.43 -0.18
Intercept α2
(0.08) (0.15) (0.00) (0.30) (0.41) (0.44) (0.02) (0.12) (0.62) (0.04) (0.62) (0.80)
-0.33 0.08 -0.32 -0.06 -0.15 -0.42 -0.23 -0.05 -0.07 0.01 0.07 -0.06
Sentiment β1
(0.05) (0.58) (0.05) (0.56) (0.37) (0.08) (0.31) (0.70) (0.58) (0.94) (0.58) (0.55)
-0.22 0.42 -0.33 -0.30 -0.13 -0.36 -0.23 0.08 0.07 0.08 -0.07 0.00
RMKT β2
(0.00) (0.00) (0.00) (0.00) (0.09) (0.33) (0.02) (0.19) (0.20) (0.12) (0.20) (0.97)
0.45 -0.35 -0.50 -0.22 -1.52 -0.54 0.21 -0.27 0.26 0.27 -0.05
SMB β3
(0.00) (0.00) (0.00) (0.03) (0.00) (0.00) (0.01) (0.00) (0.00) (0.00) (0.39)
-0.32 -0.26 -0.10 0.13 0.12 -0.33 -0.04 0.76 -0.10 0.24 0.06
HML β4
(0.00) (0.00) (0.21) (0.01) (0.13) (0.27) (0.48) (0.00) (0.06) (0.00) (0.23)
-0.36 -0.25 -0.19 -0.02 -0.08 4.32 0.29 0.06 0.06 0.14 -0.06 -0.08
UMD β5
(0.00) (0.00) (0.05) (0.79) (0.38) (0.42) (0.02) (0.40) (0.42) (0.03) (0.42) (0.15)
Table VA2
Growth
Profitability,
High- High- High- Med- High- High- Med-
SMB Low >0 - <0 >0 - =0 Low Low HML High-Low Low Med Med Low
Rxit=High,t - Rxit=Low,t = α1 + βSentiment⊥it-1 + εit
0.65 0.59 2.84 -0.19 1.01 -3.19 2.44 1.17 0.68 1.14 1.75 0.02
Intercept α1
(0.44) (0.55) (0.00) (0.78) (0.12) (0.38) (0.02) (0.05) (0.52) (0.04) (0.00) (0.95)
-0.26 -0.02 -0.11 0.08 -0.06 0.03 0.04 -0.19 0.16 -0.16 -0.12 -0.03
Sentiment⊥ β
(0.26) (0.93) (0.57) (0.64) (0.72) (0.97) (0.89) (0.24) (0.57) (0.27) (0.36) (0.80)
Rxit=High,t - Rxit=Low,t = α2 + β1Sentiment⊥it-1 + β2RMKT + β3SMB + β4HML + β5UMD + εit
0.49 1.93 1.91 -1.09 0.17 0.65 2.60 1.15 -0.79 1.66 0.79 -0.51
Intercept α2
(0.49) (0.00) (0.01) (0.01) (0.80) (0.76) (0.00) (0.04) (0.12) (0.00) (0.12) (0.21)
-0.43 0.07 -0.31 -0.06 -0.11 -0.11 -0.31 0.01 -0.06 0.04 0.06 -0.02
Sentiment β1
(0.02) (0.68) (0.10) (0.61) (0.54) (0.85) (0.21) (0.92) (0.65) (0.78) (0.65) (0.85)
-0.25 0.42 -0.35 -0.30 -0.12 -0.36 -0.25 0.09 0.07 0.08 -0.07 0.00
RMKT β2
(0.00) (0.00) (0.00) (0.00) (0.11) (0.15) (0.02) (0.16) (0.24) (0.12) (0.24) (0.94)
0.45 -0.35 -0.50 -0.21 -0.32 -0.55 0.21 -0.27 0.26 0.27 -0.05
SMB β3
(0.00) (0.00) (0.00) (0.04) (0.41) (0.00) (0.01) (0.00) (0.00) (0.00) (0.44)
-0.32 -0.26 -0.10 0.13 0.12 -1.47 -0.04 0.76 -0.10 0.24 0.06
HML β4
(0.00) (0.00) (0.22) (0.01) (0.13) (0.00) (0.52) (0.00) (0.07) (0.00) (0.22)
-0.35 -0.25 -0.19 -0.02 -0.08 -0.31 0.29 0.07 0.06 0.15 -0.06 -0.08
UMD β5
(0.00) (0.00) (0.05) (0.79) (0.39) (0.29) (0.02) (0.38) (0.42) (0.03) (0.42) (0.16)
Page 35
Table VB1 and VB2

Time Series Regressions of Portfolio Returns, Vietnam, 2000 to 2013
SENTIMENT index, the market risk premium (RMRF), the Fama–French factors (HML and
SMB), and a momentum factor (UMD) in Vietnam.
RXit=High,t − RXit=Low,t = α1 + βSENTIMENTi,t−1 + εit.
RXit=High,t − RXit=Low,t = α2+ β1SENTIMENTi,t−1 + β2RMRFt + β3SMBt + β4HMLt + β5UMDt + εit.
the lagged SENTIMENT index for the previous calendar year before controlling for
macroeconomic conditions. The first panel presents the results of univariate regressions, whereas
the second panel present the results of multivariate regressions after controlling for RMRF, HML,
SMD, and UMD. SMB and HML are excluded from the control variables when they are regressed.
Coefficients for each variable are presented first and p-values are presented in brackets.
Profitability, Growth Opportunities Growth
ME σ E D PPE/A BE/ME EF/A BE/ME EF/A BE/ME EF/A
High- >0 - >0 - High- Med- High- High-
SMB Low <0 =0 High-Low HML Low Low Med Med Med-Low
Rxit=High,t - Rxit=Low,t = α1 + βSentimentit-1 + εit
-0.21 -5.27 3.84 2.12 0.97 4.87 -0.52 1.52 0.46 1.28 -0.98
Intercept α1
(0.85) (0.16) (0.00) (0.01) (0.12) (0.05) (0.43) (0.16) (0.28) (0.38) (0.10)
0.31 0.80 -0.21 -0.01 -0.07 -1.06 0.07 0.08 0.05 -0.14 0.02
Sentiment β
(0.20) (0.31) (0.39) (0.93) (0.64) (0.31) (0.63) (0.73) (0.56) (0.81) (0.91)
-4.24 5.34 -2.00 -1.97 0.55 7.55 1.52 -0.29 1.12 0.29 0.41
Intercept α2
(0.06) (0.69) (0.54) (0.34) (0.74) (0.00) (0.43) (0.85) (0.38) (0.85) (0.83)
0.07 -3.71 0.56 0.36 -0.08 -0.22 0.00 -0.06 -0.01 0.06 0.01
Sentiment β1
(0.81) (0.07) (0.20) (0.19) (0.72) (0.48) (0.99) (0.76) (0.94) (0.76) (0.97)
-0.25 -0.06 -0.26 -0.26 -0.11 0.38 0.14 -0.08 0.03 0.08 0.11
RMKT β2
(0.01) (0.93) (0.08) (0.01) (0.16) (0.00) (0.12) (0.22) (0.61) (0.22) (0.19)
-1.92 -0.40 -0.25 0.01 -0.10 -0.04 0.06 0.04 -0.16
SMB β3
(0.04) (0.12) (0.13) (0.96) (0.50) (0.74) (0.57) (0.74) (0.28)
0.70 1.63 0.26 0.07 -0.08 0.75 -0.07 0.58 -0.08 0.42 0.02
HML β4
(0.00) (0.13) (0.29) (0.64) (0.54) (0.00) (0.65) (0.00) (0.38) (0.00) (0.90)
-0.01 0.07 0.06 0.01 0.02 0.01 0.14 -0.13 0.01 0.13 0.13
UMD β5
(0.90) (0.92) (0.55) (0.84) (0.66) (0.91) (0.03) (0.01) (0.80) (0.01) (0.03)
Table VB2
High- >0 - >0 - High- Med- High- High- Med-
SMB Low <0 =0 High-Low HML Low Low Med Med Low
Rxit=High,t - Rxit=Low,t = α1 + βSentiment⊥it-1 + εit
0.93 -2.41 3.06 2.08 0.73 2.12 -0.26 1.78 0.66 0.93 -0.92
Intercept α1
(0.17) (0.29) (0.00) (0.00) (0.06) (0.05) (0.53) (0.01) (0.01) (0.13) (0.02)
0.25 0.74 -0.27 0.04 -0.05 -0.53 0.08 -0.21 0.05 -0.04 0.03
Sentiment⊥ β
(0.43) (0.48) (0.40) (0.85) (0.78) (0.27) (0.66) (0.49) (0.68) (0.88) (0.85)
Rxit=High,t - Rxit=Low,t = α2 + β1Sentiment⊥it-1 + β2RMKT + β3SMB + β4HML + β5UMD + εit
-2.94 -8.67 0.52 -0.34 -0.35 6.00 1.74 -0.85 0.89 0.85 0.85
Intercept α2
(0.04) (0.44) (0.81) (0.80) (0.74) (0.00) (0.17) (0.38) (0.28) (0.38) (0.47)
-0.60 -2.34 0.74 0.49 0.44 0.27 -0.24 0.20 0.15 -0.20 -0.39
Sentiment β1
(0.12) (0.26) (0.21) (0.18) (0.14) (0.53) (0.48) (0.45) (0.51) (0.45) (0.23)
-0.21 -0.21 -0.20 -0.22 -0.13 0.33 0.14 -0.10 0.02 0.10 0.12
RMKT β2
(0.01) (0.78) (0.13) (0.01) (0.05) (0.00) (0.07) (0.11) (0.66) (0.11) (0.10)
-2.50 -0.29 -0.18 0.06 -0.13 -0.02 0.07 0.02 -0.21
SMB β3
(0.02) (0.27) (0.29) (0.66) (0.40) (0.89) (0.47) (0.89) (0.17)
0.67 2.33 0.19 0.02 -0.09 0.78 -0.05 0.58 -0.09 0.42 0.04
HML β4
(0.00) (0.04) (0.44) (0.89) (0.45) (0.00) (0.71) (0.00) (0.34) (0.00) (0.79)
-0.04 -0.06 0.03 -0.01 0.05 0.05 0.13 -0.11 0.02 0.11 0.11
UMD β5
(0.50) (0.93) (0.74) (0.91) (0.27) (0.51) (0.03) (0.01) (0.62) (0.01) (0.04)
Page 36
Table VI
Time Series Regressions of Portfolio Returns, Vietnam, Subsample 2009 to 2012
orthogonalized SENTIMENT⊥ index, the market risk premium (RMRF), the Fama–French factors
(HML and SMB), and a momentum factor (UMD) in Vietnam.
RXit=High,t − RXit=Low,t = α1 + βSENTIMENT⊥t−1 + εit.
RXit=High,t − RXit=Low,t = α2+ β1SENTIMENT⊥t−1 + β2RMRFt + β3SMBt + β4HMLt + β5UMDt + εit.

the lagged orthogonalized SENTIMENT ⊥ index for the previous calendar year after controlling for
macroeconomic conditions, i.e. CPI, IPI, GDP, and Central Bank’s key policy rate. The first panel
presents the results of univariate regressions, whereas the second panel present the results of
multivariate regressions after controlling for RMRF, HML, SMD, and UMD. SMB and HML are
excluded from the control variables when they are regressed. Coefficients for each variable are
presented first and p-values are presented in brackets.

High- >0 - >0 - High- Med- High- High- Med-
SMB Low <0 =0 High-Low HML Low Low Med Med Low
-3.00 -9.12 0.63 -0.27 -0.28 6.02 1.71 -0.82 0.91 0.82 0.79
Intercept α2
(0.03) (0.41) (0.76) (0.84) (0.79) (0.00) (0.18) (0.40) (0.26) (0.40) (0.50)
-1.15 -4.44 1.38 0.92 0.80 0.53 -0.46 0.37 0.28 -0.37 -0.74
Sentiment β1
(0.11) (0.25) (0.21) (0.18) (0.14) (0.51) (0.48) (0.46) (0.51) (0.46) (0.23)
-0.21 -0.23 -0.20 -0.22 -0.13 0.33 0.14 -0.10 0.02 0.10 0.12
RMKT β2
(0.01) (0.76) (0.13) (0.01) (0.05) (0.00) (0.07) (0.11) (0.65) (0.11) (0.10)
-2.51 -0.29 -0.18 0.06 0.78 -0.13 -0.02 0.07 0.02 -0.21
SMB β3
(0.02) (0.27) (0.29) (0.66) (0.00) (0.40) (0.90) (0.47) (0.90) (0.17)
0.67 2.34 0.19 0.02 -0.09 -0.05 0.57 -0.09 0.43 0.04
HML β4
(0.00) (0.03) (0.45) (0.89) (0.45) (0.71) (0.00) (0.34) (0.00) (0.79)
-0.04 -0.06 0.03 -0.01 0.05 0.05 0.13 -0.11 0.02 0.11 0.11
UMD β5
(0.50) (0.93) (0.75) (0.91) (0.27) (0.51) (0.03) (0.01) (0.62) (0.01) (0.04)
C. Investor sentiment indices
In order to examine the robustness of the resulting orthogonalized

sentiment index for Norway and Vietnam, several sentiment indices are
constructed on the basis of diverse sentiment proxies after controlling for
macroeconomic conditions. Specifically, the robustness of the predictability of the
Baker and Wurgler’s (2006) sentiment index, the predictive power of VIX and
CCI will be also examined across stock markets.
Table VIIA and Table VIIB show the correlations among 6 orthogonalized
sentiment indices. Firstly, the first-stage index is the first principal component of
all lead and lagged proxies, i.e. 16 loadings for Norway and 12 loadings for
Page 37
Vietnam, respectively. This is considered as the base sentiment index, which

reflects most information about investor sentiment. The fit of the remaining
indices in capturing investor sentiment is evaluated based on their correlations
with the first-stage index. For each market, the remaining indices includes: (i) the
parsimonious index (the SENTIMENT index (1) and (2)); (ii) the orthogonalized
index (the SENTIMENT index (3) and (4)); (iii) the orthogonalized index
excluding VIX; (iv) the orthogonalized index excluding CCI; (v) and finally the
orthogonalized index excluding VIX and CCI, i.e. the Baker and Wurgler’s (2006)
index.
Table VIIA
Correlations of Sentiment Indices in Norway, 1991-2013
This table presents the correlations among sentiment indices that are constructed in order
to test the robustness of the resulting orthogonalized sentiment index for Norway.
SENTIMENT1 reprsents the first-stage index with 16 loadings. SENTIMENT2

represents the parsimonious index, which includes the 8 lead or lagged proxies that highly
correlated with the first-stage index. SENTIMENT3 represents the orthogonalized index
after controlling for the macroeconomic conditions. SENTIMENT4 reprsents the
orthogonalized index excluding VIX. SENTIMENT5 reprsents the orthogonalized index
excluding CCI. SENTIMENT6 reprsents the orthogonalized index excluding VIX and
CCI, i.e. the Baker and Wurgler’s (2006) index.
SENTIMENT1 SENTIMENT2 SENTIMENT3 SENTIMENT4 SENTIMENT5 SENTIMENT6

SENTIMENT1 1.00
SENTIMENT2 0.92 1.00
SENTIMENT3 0.82 0.92 1.00
SENTIMENT4 -0.71 -0.83 -0.98 1.00
SENTIMENT5 0.74 0.83 0.86 -0.84 1.00
SENTIMENT6 -0.80 -0.90 -0.99 0.99 -0.86 1.00
.
Page 38
Table VIIB
Correlations of Sentiment Indices in Vietnam, 2000-2013
This table presents the correlations among sentiment indices that are constructed in order to test
the robustness of the resulting orthogonalized sentiment index for Vietnam.
SENTIMENT1 reprsents the first-stage index with 12 loadings. SENTIMENT2 represents the
parsimonious index, which includes the 6 lead or lagged proxies that highly correlated with the
first-stage index. SENTIMENT3 represents the orthogonalized index after controlling for the
macroeconomic conditions. SENTIMENT4 reprsents the orthogonalized index excluding VIX.
SENTIMENT5 reprsents the orthogonalized index excluding CCI. SENTIMENT6 reprsents the
orthogonalized index excluding VIX and CCI, i.e. the Baker and Wurgler’s (2006) index.
SENTIMENT1 SENTIMENT2 SENTIMENT3 SENTIMENT4 SENTIMENT5 SENTIMENT6

SENTIMENT1 1.00
SENTIMENT2 0.93 1.00
SENTIMENT3 0.74 0.75 1.00
SENTIMENT4 0.74 0.75 0.99 1.00
SENTIMENT5 0.74 0.75 0.99 0.99 1.00
SENTIMENT6 0.74 o.75 0.99 0.99 0.99 1.00
As a result, SENTIMENT3, i.e. the orthogonalized index excluding VIX, and

SENTIMENT5, i.e. the Baker and Wurgler’s (2006) index for Norway, have
significant negative correlations with the first-stage index (-0.71 and -0.80),
suggesting that these indices fail to reflect investor sentiment in Norway.
However, SENTIMENT4, i.e. the orthogonalized index excluding CCI for
Norway, still succeeds to capture investor sentiment with the correlation of 0.74.
This may suggest that the predictive power of VIX is only significant in Norway,
as a developed market. CCI as a sentiment proxy can also forecast stock returns in
Norway, however, its predictive power is not as strong as VIX. In the Vietnamese
stock market, the remaining indices lost little information after removing VIX,
CCI and both proxies (0.74).
In general, the robustness check shows that the sentiment indices for
Norway are sensitive to the Chicago Board Options Exchange Market Volatility
Index (VIX) whereas the sentiment indices in Vietnam show no pattern. This
suggests that the U.S. sentiment proxy, i.e. VIX, plays an important role in
constructing the sentiment index in a developed stock market, i.e. Norway, than in
an emerging stock market, i.e. Vietnam. Further research on this may provide
insights into whether the U.S. investor sentiment has stronger effect on stock
returns in developed markets than in emerging markets.
Page 39
5. Conclusions
Standard finance theories neglect the role of investor sentiment in the cross-
section of stock returns. Consistent with Baker and Wurgler’s (2006) findings,
this study contests this claim and argue that the investor sentiment has significant
predictive power in the cross-section of stock returns. Our key findings after
conducting a sorting approach are that when beginning-of-period proxies for
sentiment are low, future returns are higher for small stocks, high volatility stocks,
non-dividend-paying stocks, and value stocks. Furthermore, after conducting a
regression approach, our results partially confirm the significance of the patterns
proposed in the sorting approach. Particularly in Norway, when sentiment is high,
subsequent returns are relatively low for small firms and unprofitable firms. In
Vietnam, when sentiment is high, subsequent returns are relatively low for small
firms and firms with highly volatile stock returns. And vice-versa. Generally, our
results are consistent with the predictions that sentiment has stronger effects on
stocks that are hard to value and difficult to arbitrage.
Moreover, a robustness test of the resulting orthogonalized sentiment
index for Norway and Vietnam is implemented. Several sentiment indices are
constructed on the basis of diverse orthogonalized sentiment proxies. As a result,
the robustness test shows that the sentiment indices for Norway are sensitive to
the Chicago Board Options Exchange Market Volatility Index (VIX) whereas the
sentiment indices in Vietnam reveals no pattern. This implies that VIX, also
known as the U.S. sentiment proxy, plays an important role when constructing the
sentiment index in a developed stock market, i.e. Norway, than in an emerging
stock market, i.e. Vietnam. CCI as a sentiment proxy can also forecast stock
returns in Norway, however, its predictive power is not as strong as VIX. Further
research on this may provide insights into whether the U.S. investor sentiment has
a more pronounced effect on stock returns in developed markets than in emerging
markets.
Page 40
6. Bibliography
Baker, Malcolm, and Jeremy Stein. 2004. “Market Liquidity as a Sentiment

Indicator”. Journal of Financial Markets 7 (3): 271-299
Baker, Malcolm, and Jeffrey Wurgler. 2004. “A catering Theory of Dividends”.

Journal of Finance 59: 1125-1165
Baker, Malcolm, and Jeffrey Wurgler. 2006. “Investor sentiment and the cross-
section of stock returns”. Journal of Finance 61: 1645–1680.
Baker, Malcolm, and Jeffrey Wurgler. 2007. “Investor Sentiment in the Stock
Market”. SSRN working paper.
Baker, Malcolm, Jeffrey Wurgler, and Yu Yuan. 2010. “Global, Local, and
Contagious Investor Sentiment”. Journal of Financial Economics 104 (2): 272-
287
Brown, Gregory W., and Michael T. Cliff. 2004. “Investor Sentiment and the
Near-Term Stock Market.” Journal of Empirical Finance 11(1): 1–27.
Brown, Gregory. W., and Michael. T. Cliff. 2005. “Investor Sentiment and Asset
Valuation”. Journal of Business 78 (2): 405−440.
Chen, Nai-Fu, Richard Roll, and Stephen A. Ross. 1986. “Economic Forces and
the Stock Market”. The Journal of Business 59 (3): 383-403
Chung, San-Lin, Chi-Hsiou Hung, and Chung-Ying Yeh. 2012. “When does
investor sentiment predict stock returns?” Journal of Empirical Finance 19: 217-
240.
De Long, J. Braford, Andrei Shleifer, Lawrence H. Summers and Robert J.

Waldmann. 1990. “Noise trader risk in financial markets”. Journal of Political
Economy 98(4): 703-738.
Page 41
Dergiades, Theologos. 2012. “Do investors’ sentiment dynamics affect stock

returns? Evidence from the US economy”. Economics Letters 116: 404-407
Doms, Mark, and Norman Morin. 2004. “Consumer Sentiment, the Economy, and
the News Media”. Federal Reserve Bank of San Francisco. Working Paper
Fama, Eugene F., and Kenneth R. French. 1993. “Common risk factors in the
Returns on stocks and bonds”. Journal of Financial Economics 33: 3–56.
Fisher, Kenneth. L., and Meir. Statman. 2000. “Investor sentiment and stock
returns”. Financial Analysis Journal 56 (2): 16-23
Fisher, Kenneth. L., and Meir. Statman. 2003. “Consumer confidence and stock
returns”. Journal of Portfolio Management 30: 115–128
Glushkov, Denys. 2006. “Sentiment beta”. SSRN working paper.
Grigaliuniene, Zana, and Diana Cibulskiene. 2010. “Investor Sentiment Effect on

Stock Returns in Scandinavian Stock Market”. Economics and Management 15:
929-940
Ho Chi Minh Stock Exchange. Accessed December 01, 2012.

http://www.hsx.vn/hsx/Modules/Statistic/VnindexStatistic.aspx
Hofstede, Geert. 1980. Culture’s consequences: International Differences in

Work-Related Values. Sage Publications. London: UK.
Ibbotson, Roger, and Jeffrey F. Jaffe. 1975. “Hot issue markets”. Journal of
Finance 30: 1027–1042.
Jones, Charles. 2001. “A century of stock market liquidity and trading costs”.
Working paper,
Columbia University.
Page 42
Kling, Gerhard and Lei Gao. 2008. “Chinese institutional investors’ sentiment”.
Journal of International Financial Markets, Institutions & Money 18: 374-387
Kumar, Alok, and Charles M. C. Lee. 2006. “Retail investor sentiment and return
co-movement”. Journal of Finance 61: 2451–2486.
Kurov, Alexander. 2010. “Investor sentiment and the stock market’s reaction to
monetary policy”. Journal of Banking & Finance 34 (1): 139-149
Lee, Charles M. C., Andrei Shleifer, and Richard H. Thaler. 1991. “Investor
sentiment and the closed-end fund puzzle”. Journal of Finance 46 (!): 75–110.
Lemmon, Michael, and Evgenia Portniaguina. 2006. “Consumer confidence and

asset prices: some empirical evidence”. Review of Financial Studies (19): 1499–
1529.
Lin, Mei Chen. 2010. “The Effects of Investor Sentiment on Returns and
Idiosyncratic Risk in the Japanese Stock Market”. International Research Journal
of Finance and Economics 60: 29-43
Lowry, Michelle, and G. William Schwert. 2002. “IPO market cycles: Bubbles or
sequential learning?”. Journal of Finance 57: 1171–1200.
Qiu, Lily and Ivo Welch. 2004. “Investor sentiment measures”. No. w10794.
National Bureau of Economic Research.
Ritter, Jay R. 1991. “The Long-Run Performance of Initial Public Offerings”.

Journal of Finance 46 (1): 3-27.
Rouwenhorst, Geert K. 1998. “International Momentum Strategies”. Journal of

Finance 53: 267-284.
Schmeling, Maik. 2009. “Investor sentiment and stock returns: Some international
evidence”. Journal of Empirical Finance 16 (3): 394-408.
Page 43
Stambaugh, Robert F., Jianfeng Yu and Yu Yuan. 2010. “The short of it: Investor
sentiment and anomalies”. Journal of Financial Economics 104 (2): 288-302
Swedroe, Larry. 2012. “How investor sentiment affects returns - CBS news”.
http://www.cbsnews.com/8301-505123_162-57458690/how-investor-sentiment-
affects-returns/. Accessed 04.01.2013.
Whaley, Robert E. 2000. “The investor fear gauge”. Journal of Portfolio

Management 26 (3): 12–17.
World Bank. 2011. “Market capitalization of listed companies (% of GDP)”.

Accessed December 01, 2012.
http://data.worldbank.org/indicator/CM.MKT.LCAP.GD.ZS.
Zweig, Martin E. 1973. “An Investor Expectations Stock Price Predictive Model
Using Closed-end Fund Premiums”. Journal of Finance 28: 67
Page 44
7. Appendices
Appendix 1: Comparison of signs of the sentiment proxies
Proxies Predicted Baker and Norway Vietnam

Wurgler (2006) SENTIMENTt SENTIMENTt
CEFD
TURN
NIPO
RIPO
ES
PD-ND
VIX
CCI
Appendix 2: Comparison of the lead-lag relationship among sentiment proxies
Proxies Predicted Baker and Wurgler (2006) Norway Vietnam

CEFD Lead Lead Lead Lag
TURN Lead Lag Lag Lag
NIPO Lag Lead Lead Lag
RIPO Lead Lag Lag
ES Lag Lead Lead Lag
PD-ND Lead Lag Lag
VIX (Lead) Lag Lead
CCI (Lead) Lead Lead
Appendix 3: Comparisons of sentiment effect on firm characteristics
Characteristics Sentiment Baker and Wurlger (2006) Norway Vietnam

ME + X X ✓
- ✓ ✓ ✓
σ + ✓ ✓ ✓
- ✓ ✓ X
E/BE + ✓ ✓ ✓
- ✓ X X
D/BE + ✓ ✓ ✓
- ✓ ✓ X
PPE/A + ✓ ✓ ✓
- ✓ ✓ X
BE/ME + ✓ ✓ ✓
- ✓ ✓ ✓
EF/A + ✓ X X
- ✓ X ✓
Page 45

The Effect of Investor Sentiment On Stock Return in Norway and Vietnam - Master Thesis

Uploaded by

Copyright:

Available Formats

The Effect of Investor Sentiment On Stock Return in Norway and Vietnam - Master Thesis

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

The Effect of Investor Sentiment On Stock Return in Norway and Vietnam - Master Thesis

Uploaded by

Copyright:

Available Formats

BI Norwegian Business School – Master Thesis

- The effect of investor sentiment on

This study investigates the effects of investor sentiment on stock returns in

Whether investor sentiment affects stock returns has been an important

(2006) also provided evidence of investor sentiment predicting the returns on

The common findings of the sentiment-return relation opposes the premise

economies are characterized by the lack of an earnings history and unsophisticated

Hypothesis 3. Consumer confidence as a proxy for investor sentiment can

3. Empirical Method and Data

Employing Baker and Wurgler’s (2006) empirical method, the cross-

where i represents firms, t represents time, x represents a vector of firms’

B. Returns and characteristics

Table I provides descriptive statistics for all returns and characteristic

C. Potential sentiment proxies

In previous literature, investor sentiment can be investigated by two

D. A composite sentiment index

calculated. Consequently, 8 current or lagged proxies which have higher

As a result, 65.94% of the sample variance can be explained by the first

SENTIMENTt = 0.052CEFDt-1 0.244TURNt-1+ 0.293NIPOt-1

These 2 parsimonious indices seem not to be totally consistent with the

The first eigenvalue is estimated at 3.87, compared to 2.02 of the second

Correlations with Sentiment Correlations with Sentiment Components

Panel A: Closed-End Fund Discount (%) Panel F: Dividend Premium

40% 40% 1.00 1.00

Panel B: Turnover Panel G: VIX

Panel H: Consumer confidence index

Panel I: Sentiment Index (SENTIMENT)

Panel E: Equity share in new issues

Panel A: Closed-End Fund Discount (%) Panel E: VIX

Table III presents 10 equally-weighted portfolios according to their characteristics

The book-to-market variable shows intriguing patterns, in which it has

Panel A: ME Panel E: PPE/A

Panel B: σ Panel F: BE/ME

Panel C: E/BE Panel G: EF/A

Panel A: ME Panel E: PPE/A

Panel B: σ Panel F BE/ME

Panel C: E/BE Panel G: EF/A

B. Long–Short Portfolios Regressions

RXit=High,t − RXit=Low,t = α1 + βSENTIMENTt−1 + εit.

The dependent variable, RXit=High,t − RXit=Low,t, is the long–short portfolio’s

RXit=High,t − RXit=Low,t = α2+ β1SENTIMENTt−1 + β2RMRFt + β3SMBt + β4HMLt + β5UMDt + εit.

Employing the definitions introduced by Fama and French (1993), RMRF is

Table VA1 and VA2

RXit=High,t − RXit=Low,t = α1 + βSENTIMENTi,t−1 + εit.

Table VB1 and VB2

RXit=High,t − RXit=Low,t = α1 + βSENTIMENTi,t−1 + εit.

RXit=High,t − RXit=Low,t = α2+ β1SENTIMENTi,t−1 + β2RMRFt + β3SMBt + β4HMLt + β5UMDt + εit.

RXit=High,t − RXit=Low,t = α1 + βSENTIMENT⊥t−1 + εit.

RXit=High,t − RXit=Low,t = α2+ β1SENTIMENT⊥t−1 + β2RMRFt + β3SMBt + β4HMLt + β5UMDt + εit.

Profitability, Growth Opportunities Growth

C. Investor sentiment indices

In order to examine the robustness of the resulting orthogonalized

Vietnam, respectively. This is considered as the base sentiment index, which

SENTIMENT1 reprsents the first-stage index with 16 loadings. SENTIMENT2

SENTIMENT1 SENTIMENT2 SENTIMENT3 SENTIMENT4 SENTIMENT5 SENTIMENT6

SENTIMENT1 SENTIMENT2 SENTIMENT3 SENTIMENT4 SENTIMENT5 SENTIMENT6

As a result, SENTIMENT3, i.e. the orthogonalized index excluding VIX, and

Baker, Malcolm, and Jeremy Stein. 2004. “Market Liquidity as a Sentiment