International Review of Economics and Finance 100 (2025) 104089

Contents lists available at ScienceDirect

International Review of Economics and Finance

journal homepage: www.elsevier.com/locate/iref

Revisiting the role of investor sentiment in the stock market

Quyen Pham a,b, Huy Pham b,* , Tra Pham b, Aviral Kumar Tiwari c
a
Western Sydney University, Viet Nam
b
The Business School, RMIT University, Viet Nam
c
Indian Institute of Management Bodh Gaya (IIM Bodh Gaya), Bodh Gaya, India

A R T I C L E I N F O A B S T R A C T

JEL classification: The stock market is a critical determinant of global economic growth, and investor irrational
JEL behaviors are remarkable forces that are not only forming stock prices but also determining the
G10 stock market’s performance. To examine the significant impacts of those irrationalities, various
G41
methods have been applied to generate an investor sentiment index. However, isolating the ir­
Keywords: rational judgments of investors is a challenge and the existing sentiment indices are inefficient. To
Investor sentiment
overcome this shortcoming, this paper develops a new investor sentiment index. We take into
Sentiment index
Behavioral bias
account only the irrational part of people’s behavior biases that lead to misvaluation in stock
Return predictability markets, neither behavioral biases nor misvaluation per se. We then conducted several robust
tests toward stock returns’ predictability using time-series data for the U.S. and Chinese markets.
Various empirical methods, including OLS, TVP-VAR model, and out-of-sample test are used in
this study. Our results confirm the advantages of our index to assess the predictability towards
stock returns compared to two common existing measures of investor sentiment: survey-based
Consumer Sentiment Index and market-based Baker & Wurgler index. To the best of our
knowledge, this is the first study that presents this novel approach to capture the irrationalities of
investors.

1. Introduction

Various methods have been applied to better quantify investor sentiment. Investor sentiment refers to the behavioral bias of
humans that leads to either noise trading or misvaluation in stock markets (Black, 1986; De Long et al., 1990; Shiller et al., 1984).
Currently, we have survey-based, Internet-based, and market-based sentiment indexes. However, isolating the irrational judgments of
investors is a challenge and we cannot reach a conclusion about investor irrationality using the existing sentiment indexes. The
isolation of irrationality is extremely important because of the following reasons. First, the Efficient Market Hypothesis (EMH) implies
the well-functioning of stock markets conditionally on the rational behavior of investors and only the irrationality of investors distorts
the performance of stock markets. Using the existing sentiment indexes, a high index does not equal a high irrationality and vice versa,
and the findings based on them can be misleading. Secondly, in the bigger context of the stock market–economic growth nexus, we
must point out whether the policymakers should put efforts into managing investor sentiment. The answer depends on the degree of
irrationality that the failure to seize it cannot help us make meaningful policy implications.
To overcome the insufficiency associated with existing sentiment indexes, this paper develops a new method to capture investor

* The corresponding author. RMIT FinTech-Crypto Hub School of Business & Management RMIT University 702 Nguyen Van Linh Blvd, Tan
Phong Ward District 7, Ho Chi Minh City, Viet Nam.
E-mail address: huy.phamnguyenanh@rmit.edu.vn (H. Pham).

https://doi.org/10.1016/j.iref.2025.104089
Received 28 October 2024; Received in revised form 21 January 2025; Accepted 31 March 2025
Available online 2 April 2025
1059-0560/© 2025 The Author(s). Published by Elsevier Inc. This is an open access article under the CC BY license
(http://creativecommons.org/licenses/by/4.0/).
Q. Pham et al. International Review of Economics and Finance 100 (2025) 104089

sentiment. First, we capture four common behavioral biases in stock markets: Heuristic, Overconfidence, Loss aversion, and Herding
behavior. Heuristic happens when investors overweight historical prices over other information when valuing stocks, therefore is
measured by the predictability of technical indicators toward stock returns. Overconfident investors are those who overreact using
their private information while reluctant in update their beliefs. Whenever we capture the short-term momentum but long-term
reversal in stock prices, we record the existence of overconfidence bias. With Loss Aversion bias, investors are not emotionally
indifferent between gains and losses. While they are comfortable to realize profits, the loss positions will be held irrationally longer. We
document this bias whenever there is a positive relationship between stock turnover and stock returns. For Herding behavior, the
predictability of market-wide net position and past firm-level net position toward current net position can capture the extent of this
bias. After capturing the four common behaviroal biases, we produce a Bias Index using Principal Component Analysis. Next, an
indicator of stock misvaluation will be regressed on this Bias index together with other determinants of misvaluation. The statistically
significant coefficient of the Bias index will be our sentiment index, implying the extent to which behavioral biases induce mis­
valuation. We then conducted the test toward stock returns’ predictability using time-series data for the US and Chinese markets.
Besides the traditional OLS regression for the predictability test, we employ the Time-varying Parameter estimation for the VAR model
(TVP-VAR) to capture the time-varying effects of investor sentiment on stock returns. By using this TVP-VAR estimation, we compare
the predictability between the U.S. and China time-series data to examine the performance of our new sentiment index in both
advanced and emerging markets. We also conducted the out-of-sample test to assess the performance of our new sentiment index. The
test confirms the advantages of our index compared to two common existing measures of investor sentiment: the survey-based
Consumer Sentiment Index (CSI) and the market-based Baker & Wurgler index (BW index).
Our contribution to the literature on investor sentiment is two-fold. First, we contribute a novel investor sentiment indicator to the
literature. Our new investor sentiment index isolates investor behavior’s irrational component which is the cause of stock market
misvaluation. Unlike existing indices, such as the survey-based Consumer Sentiment Index or the market-based BW index, that fail to
differentiate between rational and irrational sentiment, our new index focuses specifically on the irrational biases—such as heuristic,
overconfidence, loss aversion, and herding behavior—that distort market prices. This innovation allows for a more precise assessment
of investor sentiment by targeting the elements that contribute directly to mispricing in stock markets. The second contribution is
methodological and involves the application of the Time-Varying Parameter Vector Autoregressive (TVP-VAR) model with stochastic
volatility. This model enables a more nuanced understanding of the time-varying relationship between investor sentiment and stock
returns, a feature that is often overlooked in static models. By incorporating stochastic volatility, the TVP-VAR model allows us to more
accurately capture how investor sentiment evolves and influences market outcomes over time. This dual approach—introducing a new
sentiment index and employing a time-varying, volatility-adjusted model—offers a unique and more comprehensive tool for analyzing
the complex dynamics between investor sentiment and stock market performance.
The remainder of this paper is as follows. Section 2 discusses the literature on investor sentiment measurements. Section 3 describes
the data and methodology. The empirical findings and discussions are presented in Section 4 and Section 5 concludes the paper.

2. Literature review

When walking along the history of stock market literature regarding the field of investment behavior and asset pricing, it is not
difficult to realize the evolution in the ways researchers interpreted the behavior of investors toward stock valuation. During this
evolution, the exclusive agreement is that the market price of a firm’s shares is the equilibrium of diffusion cognition of investors about
a firm’s value. This statement was the foundation for many models describing the process of stock price formation (Basak, 2005;
Bhamra & Uppal, 2014; Campbell & Kyle, 1993; Diether et al., 2002; Grossman, 1976; Jackson, 1991; Shiller et al., 1984). However,
the most aggressive debate relates to the underlying psychological process through which investors form their cognitive valuation. The
traditional school of thought, represented by Maximizing Expected Utility Theory and Efficient Market Hypothesis, assumed that
investors employed only fundamental information and rationally weighted each piece of information to form an expected value of
firms (i.e. informed investors), then trading in the manner that maximizing their utility (i.e. rational investors) (Fama, 1970; Lucas Jr,
1988; Rubinstein, 2001). The behavioral finance strand, on the other hand, supports the argument that investors suffer from behavioral
biases in either valuing firms and trading manner (i.e. irrational investors, noise traders, uninformed traders) (Kahneman & Tversky,
1979; Black, 1986; Shiller et al., 1984; Mendel & Shleifer, 2012; Ramiah, Xu, & Moosa, 2015).
Investor sentiment refers to human behavioral biases that lead to either noise trading or misvaluation (Black, 1986; De Long et al.,
1991; Shiller et al., 1984). Based on this definition, behavioral biases are the reason leading to investor sentiment or irrationality, while
misvaluation is the consequence caused by this irrationality. Therefore, current approaches to measuring investor sentiment fall into
two subcategories: the reason-based approach, which measures sentiment according to behavioral biases, and the consequence-based
approach, which measures sentiment according to the misvaluation of investors. Both approaches have been captured in previous
studies using three tools: surveys (Beer & Zouaoui, 2013; Kurov, 2010; McLean & Zhao, 2014; Zouaoui et al., 2011), Internet-search
text (Da et al., 2015; Joseph et al., 2011; Kim & Kim, 2014), and market data.
Survey-based sentiment indices are extracted from the responses of individual investors and consumers regarding their expecta­
tions of future economic conditions. For instance, the Gallup survey asks investors about their beliefs about stock returns in the next 12
months. The American Association of Individual Investors survey captures the proportion of investors who are bullish, bearish, or
neutral toward stock markets. The Investors Intelligence Sentiment Index collects similar information but from newsletters instead of
from investors. The Michigan Consumer Sentiment Index (MCSI) investigates consumers’ expectations about economic growth and
their planned consumption. Prosad et al. (2015) proposed a survey to capture three common behavioral biases such as overconfidence,
herding behavior, and disposition effects. Previous authors have argued that these investor sentiment surveys can capture the mood of

2
Q. Pham et al. International Review of Economics and Finance 100 (2025) 104089

investors, which is the most similar to investor sentiment (Beer & Zouaoui, 2013; Greenwood & Shleifer, 2014; Kurov, 2010; McLean &
Zhao, 2014; Zouaoui et al., 2011). While survey-based measures have been widely applied, they have inherent drawbacks; for example,
respondents may not be entirely truthful because of personal reasons or the survey’s methodology, and even if they are honest, their
investing behaviors may differ from their responses (Greenwood & Shleifer, 2014). In addition, the most important concern is that
survey-based sentiment indices cannot capture the irrational aspects of investor expectations (Zhou, 2018). In other words, investors’
expectations about economic conditions and stock markets may be rational according to the current economic situation. The surveys
measuring behavioral biases also cannot tell whether these biases will lead to irrational valuation in stock markets. Without isolating
the irrationality, survey-based sentiment indices are inconsistent with the theoretical definition of sentiment.
In contrast to capturing the market sentiment from investors’ responses to surveys, the Internet-based approach relies on textual
search analysis and media messages (Da et al., 2015; Garcia, 2013; Sun et al., 2016). To analyze market sentiment, this approach
classifies keywords and messages into “optimism,” “pessimism,” and “neutral” categories. Although the approach has been shown to

Combined approach
- new index

Biases lead to
misvaluation

Fig. 1. Investor sentiment theoretical framework.

3
Q. Pham et al. International Review of Economics and Finance 100 (2025) 104089

generate the highest predictability for stock movement (Zhou, 2018), its drawbacks are similar to those of the survey-based sentiment
index, making it a noisy measure of investor sentiment.
The market-based approach is considered the most objective measurement of investor sentiment (Aggarwal, 2019; Beer & Zouaoui,
2013). Rather than directly capturing investors’ beliefs, composite sentiment indices extract sentiment from market data. The most
well-known study in this area is that by Baker and Wurgler (2007), who used six proxies of investor sentiment: closed-end fund
discounts, the number of IPOs, IPO initial returns, dividend premiums, the ratio of equity issues to total equity and debt issues, and
market turnover. Chen et al. (2014) developed another outstanding composite sentiment index based on PCA and comprising interest
rate, exchange rate, industrial production change, money supply change, market turnover, and the number of new accounts in stock
exchanges. Hudson and Green (2015) developed a composite sentiment index comprising eight indicators: the advance-decline ratio,
closed-end fund discounts, money flow index, put–call volume ratio, put-call–open-interest ratio, the relative strength index, realized
volatility, and trading volume. The Chicago Board of Options Exchange generates an index called the Volatility Index (VIX) to capture
market sentiment. The approach of VIX uses an option valuation model to calculate the expected volatility of the S&P index. A higher
VIX implies higher expected volatility, meaning higher risk and greater fear (Whaley, 2000). Another example of a composite
sentiment index is the CNN Fear and Greed index. This index is a composition of seven market indicators: momentum, stock price
strength, stock price breadth, put-call options, junk bond demand, market volatility, and safe-haven demand. The index takes the value
of 0 for maximum greediness and 100 for maximum fear (CNN Business).
Despite the multidimensional methodology in generating market-based sentiment indexes, the underlying approach is also reason-
based and consequence-based. Baker and Wurgler (2007), Chen et al. (2014), Hudson and Green (2015), or VIX and CNN Fear and Gear
index are existing market-based measures that equate misvaluation with sentiment, thus implicitly ignoring all other factors leading to
misvaluation. In contrast, approaches that equate behavioral biases with investor sentiment (e.g., Adams et al., 2008; Baker et al.,
2019; Hwang et al., 2021) are also not failsafe. Evidence from the psychological sciences shows that behavioral biases do not always
lead to misvaluation or bad decisions (Gigerenzer & Gaissmaier, 2011; Wübben & Wangenheim, 2008). If this is the case, using
behavioral biases per se as a proxy for sentiment may result in measurement errors (Zhou, 2018). The methodology and limitations of
each existing sentiment measure are provided in Appendix 1.
To overcome these limitations, in this study, we question on how to isolate the irrational component out of investor behavior,
capturing only the extent to which behavioral biases relate to misvaluation. We measure investor sentiment by the irrational
component of investors’ behavioral biases which leads to stock market misvaluation. The core theoretical premise is drawn from the
Efficient Market Hypothesis (EMH) and Behavioral Finance. From an EMH standpoint, investor irrationality would not have a long-
lasting impact on stock prices, as any misvaluation would be arbitraged away quickly. By developing a new sentiment index that
targets the irrational aspects of sentiment and examines its predictability towards stock returns, our study shows that misvaluation can
be sustained in the market, contradicting the EMH assumption. On the other hand, the research question directly aligns with the
behavioral finance theory, which challenges the notion of rational investors and acknowledges the presence of systematic biases that
distort decision-making and market prices. Moreover, the behavioral finance school of thought acknowledges that the influence of
investor sentiment on stock returns varies depending on external factors, market conditions, and shifts in investor psychology (Chu
et al., 2016; Dash & Maitra, 2018; Marczak & Beissinger, 2016). Therefore, we use of the Time-Varying Parameter Vector Autore­
gressive (TVP-VAR) model with stochastic volatility in the study to capture this stochastic dynamic. The theoretical framework of the
study is demonstrated in Fig. 1.

3. Data and methodology

3.1. Data and sample

We make the test in the US and Chinese for the comparison of sentiment index performance in both emerging and advanced stock
markets. The New York Stock Exchange and Shanghai Stock Exchange are investigated regarding their highest value of market
capitalization. The period of the research was from 2001 to 2019. Since 2020, the global economy in general and stock markets all over
the world, in particular, have been seriously affected by COVID-19, we exclude these years from the sample and leave the examination
of COVID-19 impacts out of research scope.
To produce a sentiment index, data have to be collected from Refinitiv Eikon to estimate behavioral biases and stock misvaluation.
In terms of behavioral biases, input data related to firm-level stock prices, returns, and trading volume were collected every month for
the annual index and daily basis for the quarterly index. In terms of misvaluation, input data related to firm-level financial figures from
financial statements. The sentiment index is estimated at annual and quarterly frequency for each country.

3.2. Methodology

3.2.1. Behavioral biases

3.2.1.1. Heuristic. If heuristic is defined as the overweight of a type of information over the others, past price movements are agreed as
being commonly over-weighted by investors, as known as “representativeness” (Di Guilmi et al., 2014). Therefore, the higher the level
of heuristic suffered by investors, the higher the predictability of technical indicators toward stock returns. In this research, the
predictability of technical indicators, R-squared, toward stock returns will be an indicator for the heuristic. Following Neely et al.

4
Q. Pham et al. International Review of Economics and Finance 100 (2025) 104089

(2014), OLS will be employed to estimate the regression model:

MON.RETURNSi = α1 + β11 MA3M + β12 MA12M + ε1 (1)

where MON.RETURNSi is the monthly returns of stock i for 12 months of year t, MA3M , MA12M is the monthly average price of 3 months
prior and 12 months before the current month, respectively. Adjusted R-square will be recorded as the level of heuristic for firm i in
year t, named HEURi,t .

3.2.1.2. Overconfidence and conservatism. Synthesizing from the literature on overconfidence and conservatism, a common conclusion
is that investors suffering from these two biases were likely to be overreacted/underreacted in the short term but are reluctant to
update their beliefs. (Daniel et al., 2005; Scheinkman & Xiong, 2003). This bias leads to the anomaly of short-term momentum but the
long-term reversal in stock returns (McLean, 2010; Zaremba et al., 2020). If, at the beginning of year t, investors make the judgments
on the stock’s value with overconfidence bias, then 12-month cumulative returns at the end of year t represent investors’ short-term
valuation and 36-month cumulative returns at the end of year t + 2 – represent the updated valuation. As long as there is a reverse in
36-month returns compared to 12-month returns, the difference indicates the level of overconfidence-conservatism bias in year t of
firm i, named OVERCONi,t . Following the procedure of Blackburn and Cakici (2017):
OVERCONi,t = CUM.RETURNS36M − CUM.RETURNS12M

where CUM.RETURNS36M is cumulative returns for 36 months from the beginning of year t, and CUM.RETURNS12M is cumulative
returns for 12 months from the beginning of year t.

3.2.1.3. Loss Aversion. Under the paradigm of Prospect Theory, people are loss aversion rather than risk aversion by which they
evaluate investment opportunities subject to the potential gains/losses compared to a reference point of wealth rather than the risk
level of the investment (Kahneman et al., 1991). Associating with loss aversion in the form of the disposition effect, investors tend to
close the profitable positions too early, observed by the subsequent excessive trading following a gain, while being resistant to closing
the loss-making positions, observed by the low trading following a loss (Li & Yang, 2013). Following the procedure of Statman et al.
(2006), the coefficient between lagged returns and stock turnover demonstrates the level of loss aversion. The authors run a VAR model
regressed on 9 lagged turnover as an endogenous variable and 2 lagged stock returns as an exogenous variable, following SIC criteria
for choosing optimal lag lengths. However, this research examines Loss aversion across individual stocks through 12 months, therefore
a 9-lagged turnover will make the observations insufficient. Therefore, this research employed the criteria of maximum lag length from
Schwert (1989) that:
( )1/4
T
pmax = 12 x 100 = 7 with T as the number of observations.

By using this formula for maximum lag lengths, this research is aware of the fact that it is not the optimal lag length for time-series
analysis. However, using either AIC or SIC criteria will return different lags for different individual stocks that lead to inconsistency in
comparing and running regressions. Finally, k = 7 and l = 2 will be used in the following VAR model for all individual stocks:
7
∑ 2
∑
MON.TURNi,m = α2 + β2k MON.TURNi,m− k + θ2l MON.RETURNSi,m− l + ε2 (2)
k=1 l=0

where MON.TURNi,m equals the monthly value of traded/value of shares in month m, MONTH.RETURNSi,m equals monthly returns in
month m. ADF test will be conducted to test the stationary. Any firm’s turnover time series that is non-stationary will be excluded and
treated as a missing value. Because l = 2, the average of significant θ21 and θ22 will be recorded as a proxy of loss aversion of firm i in
year t, named LAi,t , and expected to be positive. Any insignificant θ2l will not be considered for taking the average. If both θ21 and θ22
are insignificant, LAi,t equals zero.

3.2.1.4. Herding behavior. Although herding behavior has been well-recorded in literature, a common approach to detect the herding
behavior is using an aggregate market bias toward cross-section stocks rather than addressing its effects on individual securities. The
argument is that investors usually herd toward market-wide returns that eventually lead to the co-movements of cross-section stock
returns (Chiang & Zheng, 2010; Hwang & Salmon, 2004). However, the majority’s trading positions toward a particular stock can
affect other traders of the same stock, especially when there is not any obvious market-wide trend. Therefore, this research proposes a
combined approach to measuring herding behavior in an individual stock, using the regression model of the individual stock’s trading
position regressed on market-wide trading volume and individual stock’s trading volume.
The term "net position" refers to the difference between total buying volume minus selling volume. Net position can be either “net
long” (the difference is positive) or “net short” (the difference is negative). Stock with net long (net short) demonstrates that the
majority of holders are buying (selling) and expecting a price increase (decrease). The predictability of market-wide net position and
past firm-level net position toward current net position can capture the extent of herding behavior in an individual stock. 3-month is
applied for the lagged net position of both market-wide and individual stocks, following the herding formation period from Jegadeesh
and Titman (2001), which is:
LAG.NETi = BUYINGi - SELLINGi of stock i averaging over 3 months before the current month.

5
Q. Pham et al. International Review of Economics and Finance 100 (2025) 104089

LAG.NETm = BUYINGm - SELLINGm of aggregate market averaging over 3 months before the current month.
The model for the stock i’s net position is as follows:
NET.POSi = α3 + β31 LAG.NETi + β32 LAG.NETm + ε3 (3)
The R-square will be recorded as the level of herding for firm i in year t, named HERDi,t .
Unfortunately, the data for buying and selling volume are not available for the cross-country stock markets in the sample.
Therefore, the study of Easley et al. (2016) is employed to estimate the buying and selling volume from the aggregate trading volume.
In this paper, the authors argued that based on the price changes, one can estimate the percentage of buying volume and selling volume
in the total volume traded, given the price follows the Student’s distribution. The accuracy also has been justified using multiple robust
checks. For details, consider the following equation:
( )
̂ B = Vτ .t Pτ − Pτ− 1 , df
V (4)
τ
σΔP
[ ( )]
̂ S = Vτ . 1 − t Pτ − Pτ− 1 , df
V (5)
τ
σ ΔP

̂ B, V
in which V ̂ S are estimated buying and selling volume from the total traded volume V
̂ S for month τ, df = 1. Pτ is the stock price at the
τ τ τ
end of month τ. σ P is the standard deviation of price changes over 12 months.

3.2.1.5. Composite behavioral bias. Principal Component Analysis (PCA) was initially developed by Pearson (1901) and Hotelling
(1933) and is applied commonly in economics and finance. Eigenvalues of variables will be estimated to identify variables with similar
components to be representatives of the same factor. In this study, PCA is applied to generate a composite behavioral bias (BIASi,t ) for
firm i in time t, which is in the form of:
BIASi,t = π1 HEURi,t + π2 OVERCONi,t + π3 LAi,t + π4 HERDi,t (6)

3.2.2. Misvaluation
If misvaluation is defined as the divergence of the market’s valuation from the issuing firm’s fundamentals (Black, 1986; Shiller
et al., 1984; De Long et al., 1991), then it should be measured by the residuals between these two. The most common indicator for the
market valuation of stocks is stock returns.
The fundamentals of firms can be synthesized in the literature of four different dimensions: earnings, human capital, physical
capital, productivity, and innovation. In various economic growth models, the aggerate output of the whole economy is the sum of all
firms’ output that was driven by these fundamentals (see neoclassical growth model – Solow (1999), endogeneous growth model -
Romer (1994)). Stock returns were found to be affected by fundamental variables related to a firm’s earnings such as accounting
earnings, cash flow, or EVA (Allen et al., 2013; Kothari et al., 2006; Sharma & Kumar, 2010; Vuolteenaho, 2002). In terms of a firm’s
human capital, stock returns were proved to be predicted by the firm’s labor’s quantity (Belo et al., 2014; Kuehn et al., 2017), labor’s
quality (Pantzalis & Park, 2009), value added (Hansson, 2004; Riahi-Belkaoui & Picur, 1994). Regarding the innovative aspect of
firms, stock returns were examined in the relationship with R&D investments (Li, 2011; Lin, 2012), and innovative efficiency
(Hirshleifer et al., 2013; Kumar & Li, 2016). In terms of productivity, it is recorded in literature the connection between TFP and stock
returns (Chun et al., 2016; İmrohoroğlu & Tüzel, 2014). Concerning physical capital, it could be found in the literature the connection
between stock returns with capital expenditures (Lamont, 2000; Titman et al., 2009), with asset growth (Cooper et al., 2008; Gray &
Johnson, 2011). In addition to the firm’s fundamentals, stock returns fluctuated with the level of investor sentiment (Baker & Wurgler,
2007; Stambaugh, Yu, & Yuan, 2012; Chung et al., 2012; Huang et al., 2015; Shen, Yu, & Zhao, 2017).
In this research, Value-added growth (TVAi,t ) will be an indicator of earnings and labor quantity, asset growth (ASSETi,t ), TFP
(TFPi,t ) for productivity, and R&D expenditures growth (RDi,t ) for innovation.
For calculating TFP, this research follows the procedure of Bournakis and Mallick (2018) that:
Revenuei,t
TFPi,t = (7)
Assets1− a .Labor costsa

in which the weight a = Labor costs/Value-added.

The regression model for generating misvaluation will be:
ANN.RETURNSi,t = α4 + β41 TVAi,t + β42 ASSETi,t + β43 TFPi,t + β44 RDi,t + ε4 (8)

From the above equation, the residuals ε4 are the proxy of the misvaluation of stock i in year t. The residuals take negative values in
case of undervaluation and positive values in case of overvaluation, named MISVALi,t .
For the robust test, we modify the way we extract misvaluation indicator. Rather than using fundamental factors as described in
Equation (8), we employ Fama-French 5 factors to extract abnormal returns as indicator of stock misvaluation. By this combination of
two approaches, we examine misvaluation in term of both fundamental and market factors.
For FF 5-factor, we extract expected returns from the following regression:

6
Q. Pham et al. International Review of Economics and Finance 100 (2025) 104089

( )
− rf = α + β1 * rM − rf + β2 *SMB + β3 *HML + β4 *RMW + β5 *CMA
ri,t̂ (9)

In which SMB represents size premium and HML represents value premium, RMW – profitability premium and CMA – investment
factor. Misvaluation in this approach will be the abnormal returns between predicted and actual returns at firm level.

3.2.3. Generating investor sentiment index

To capture the extent to which behavioral biases induce misvaluation as an indicator of investor sentiment, each firm’s mis­
valuation will be regressed on the PCA bias index:
MISVALi,t = α5 + β5 BIASi,t + h5 Ki,t + ε5 (10)

β5 is the sentiment index of year t, named SENTt , and can take either negative (bias induces undervaluation) or positive (bias induces
overvaluation) values. β5 is treated as a missing value if it is statistically insignificant.
Ki,t is the set of other misvaluation antecedents: idiosyncratic risk, liquidity risk, and firm value uncertainty. Besides investor
sentiment, the literature on stock market inefficiency showed various determinants of stock misvaluation in which the main factors
were arbitrage risk and the uncertainty about the firm’s value. Arbitrage risk was defined as the costs paid by arbitragers if they want to
take advantage of mispricing and thereby correct the price, which includes holding costs and transaction costs. The higher this risk, the
more persistent and intensive mispricing (Cao & Han, 2016; Doukas et al., 2010; Pontiff, 2006; Stambaugh, Yu, & Yuan, 2015). The
holding costs of an individual stock were caused by the idiosyncratic risk associated with that stock while the transaction costs were
caused by the liquidity risk. Following previous studies in estimating idiosyncratic risk (Guo & Savickas, 2006; Herskovic et al., 2016),
we run the ARCH regression for each stock every year:
ri,t = α + βRM,t + ε (11)

In which ri,t is the weekly premium returns of stock i, RM,t is the weekly premium returns of the market index over the riskless US
monthly Treasury bills. Idiosyncratic risk of stock i in year t, IVOLi,t , will be estimated as the standard deviation of ε. Liquidity risk of
stock i in year t, LIQUIDi,t will be calculated as the ratio of the value of traded volume to the value of shares at the end of the year.
Another determinant of mispricing, the uncertainty about a firm’s value, referred to the hard-to-value stocks. These are stocks with
specific characteristics that make it difficult to estimate the fair value, leading to misvaluation. To capture the extent of value un­
certainty, previous studies employed firm age as the proxy (Baker & Wurgler, 2003; Kumar, 2009; Zhang, 2006). However, because
data about firm age was not widely available, the number of listed years will be used for AGEi,t . A number of listed years represented
the period that the firm’s information has been available for public access which makes this indicator a valid replacement for firm age.
In this approach, the sentiment index is constructed at an annual level for each country. Although investor sentiment and stock
market volatility are changing on a high-frequency basis and previous studies use daily or even intraday data for sentiment analysi, our
misvaluation is constructed from the firm’s fundamental data in which the highest frequency is quarterly. Therefore, we also construct
the sentiment index for the quarterly level. The increasing frequency from annually to quarterly level allows us to better capture the
fluctuation of investor sentiment, stock markets, and economic conditions. To construct the quarterly sentiment index, daily stock
data, and quarterly economic data have been collected from Datastream. The methods for constructing and extracting variables are
similar to the annual sentiment index, except for data frequency.

3.2.4. Testing the predictability towards stock returns

3.2.4.1. ARCH estimation. To test the power of the new sentiment index in stock markets, this research conducts a comparative study
between the new sentiment index and the two common ones which are the market-based BW sentiment index and the survey-based
Consumer Confidence Index (CSI). The comparative study includes the test for the predictability of the sentiment index toward stock
returns using time-series data in the US and China.
BW sentiment index is available only in the US stock market. For China, we construct the BW index following the procedure in
Baker et al. (2012). To do this, we need economic data, stock market data, and sentiment data for both annual and quarterly frequency.
For the predictability test toward stock returns, the procedure follows the study of Zhou (2018) in which aggregate market returns
will be regressed on proxies of investor sentiment. ARCH estimator for time-series data following regression model:
RETURNSt = αH + βH X + εH (12)
X is a sentiment proxy. The comparison of log-likelihood in this equation for each sentiment proxy: SENTt , BWt , and CSIt will be
made to justify the explanatory power of the new sentiment index towards aggregate stock returns.

3.2.4.2. TVP-VAR model estimation. The traditional OLS regression model assumes that stock returns volatility is constant over time,
as well as the effects of its determinants. These assumptions insufficiently reflect the nature of the real world that both volatility and
the causal effects of stock returns’ determinants are time-varying. The critical role of time-varying parameter vector autoregressive
(TVP-VAR) models has been emphasized in a large body of financial econometrics literature (Chan, 2022) because they allow the
model’s coefficients to change over time. However, the causal relationships between economic variables toward stock returns also
exhibit stochastic volatility shocks. In such cases, the model with time-varying coefficients but constant volatility leads to the potential

7
Q. Pham et al. International Review of Economics and Finance 100 (2025) 104089

bias due to the fluctuations in disturbance volatility. Among the various models employed for volatility modeling and forecasting, the
stochastic volatility model proves its advantages because the model accounts for the fact that asset returns volatility changes over time
(Le et al., 2023; Yu, 2012). In this research, the robustness test for stock returns’ predictability is expanded to the employment of the
TVP-VAR model with stochastic volatility. Compared to conventional models with fixed parameters such as OLS estimation, the
TVP-VAR model with stochastic volatility offers more flexibility in modeling, enabling more precise predictions. Moreover, it proves to
be a potent tool for effectively capturing and modeling the ever-changing interactions between variables, ultimately leading to more
accurate forecasts and a deeper comprehension of the underlying economic or social processes (Koop & Korobilis, 2013).
We estimate the TVP-VAR model with stochastic volatility using the Markov chain Monte Carlo (MCMC) method (Primiceri, 2005).
Given the bivariate process Zt between investor sentiment (X1,t ) and stock returns (X2,t ), the TVP-VAR model under stochastic volatility
is as follows:
p
∑
Zt = αt + Ak (t)Zt− k + εt (13)
k=1

where εt ∼ N(0, σ2t ), σ 2t = α exp(ht ), ht+1 = βht + γ t and γ t ∼ N(0, σ 2γ ). In addition, αt is the intercept, p is lag order and Ak (t) are (2x2)
matrix of time-varying coefficients. Each component of this Ak (t) matrix satisfies Ai,j (t + 1) = Ai,j (t) + μt with μt ∼ N(0, φ). By this
( )
specification, A2,1 (t) k will be the lag k time-varying coefficients for the causal effect of investor sentiment on stock returns. The
MCMC method generates 5000 time series of A2,1 (t) for each lag and a t-test will be employed to determine the statistical significance.
We then compare the proportion of insignificance coefficients between our new sentiment index with the BW index and CSI for the U.S.
and China time-series market data.

3.2.4.3. Out-of-sample predictability. Another robust test we employ is out-of-sample predictability. By the set of coefficients extracted
from the U.S. data, we measure their forecast accuracy for China data using Mean Squared Error (MSE):
1∑
MSE = (yi − ŷi )2 (14)
n

where n: the total number of observations.

yi : the actual value of observation i in out-of-sample data
ŷi : the predicted value generated from the estimated in-sample model
The MSE will be compared between our new sentiment index with the BW index and CSI to determine which sentiment index will
perform better in out-of-sample predictability. We also make a similar procedure for the predictability of the China model toward US
data. Through these tests, we can examine which investor sentiment index will generate a model with better out-of-sample
predictability.

4. Findings and discussions

4.1. Misvaluation in stock markets

To generate the sentiment index, we have to measure the misvaluation and the behavioral biases at the firm-level. Following
Equation (4), misvaluation at the firm-level is the residual from the regression model of annual stock returns regressing on fundamental
determinants. Table 1 reports the mean of variables in Equation (4). Because misvaluation are residuals that have zero mean, we
separate firms in each country into an overvaluation group (positive residuals) and undervaluation group (negative residuals) and
report the mean of each group for a more meaningful interpretation (see Table 2). On average, the stock market in either the Chinese
market or the US market commonly witnessed undervaluation. The advanced economy such as the U.S. is usually accompanied by a

Table 1
Firm’s fundamentals descriptive statistics.
Variables China US

Annual
ANN.RETURNSi,t 0.060 0.124
TFPi,t 2.185 2.285
TVAi,t 0.051 0.080
ASSETi,t 0.144 0.088
RDi,t 0.282 0.140

Quarterly
QUAR.RETURNSi,t 0.017 0.030
TFPi,t 1.120 0.060
TVAi,t 0.052 0.026
ASSETi,t 0.020 0.022
RDi,t 0.037 0.091

8
Q. Pham et al. International Review of Economics and Finance 100 (2025) 104089

stable growth rate that stock undervaluation is widely observed (Huang & Wirjanto, 2012). Similarly, emerging markets with higher
economic growth rates should demonstrate a larger P/E ratio as an indicator of overvaluation. However, undervaluation can also be
documented in these countries due to other reasons. Huang and Wirjanto (2012) claim the undervaluation of earnings volatility in
China. Chien, Lee, Hu, and Hu (2015) shows the drop-down of Asian stock markets since 2002 and only partially converged so far.
Caporale et al. (2019) provide some macroeconomic factors such as interest rate, financial crisis, or global integration that may affect
the speed of convergence and lead to stock market undervaluation in Asia.

4.2. Behavioral biases and the new sentiment index

This section reports the findings of four behavioral biases and our new sentiment index. The four common behavioral biases
Heuristic, Overconfidence, Loss Aversion, and Herding behavior were estimated annually and quarterly for each firm. The results in
Table 3 are averaged across firms and over time for each country. Both annual and quarterly data show a lower level of behavioral bias
in the US in comparison with China. The reason for differences in bias level, according to previous studies, may relate to the inequality
of financial literacy, investor sophistication, and national culture between an advanced market and an emerging market.
To capture the irrationality element in behavioral bias that leads to misevaluation, equation (5) was regressed using cross-sectional
data of firms for each country every year. The annual sentiment index is a significant coefficient at the 10 % level of these regressions
which are estimated by OLS. Because the homoscedasticity assumption cannot be held for all countries and all periods, the robust
standard error will be estimated and employed in hypothesis testing. This section reports descriptive statistics of the sentiment level of
two countries in the period from 2001 to 2019 in Table 4.
For China, behavioral bias enhances undervaluation at the annual level but overvaluation at the quarterly level. Behavioral biases
cause a short-term overvaluation but undervaluation for a longer time horizon in the Chinese market. This observation is also wit­
nessed in previous studies examining the misvaluation in Chinese stock markets. Tanl et al. (2010) concluded the short-term over­
valuation of Chinese stocks was caused by investors’ irrationality and market inefficiency. This overvaluation should be corrected in a
long time. The situation is the opposite in the U.S. as indicated by the mean value of SENT. This finding confirms the short-term
momentum and long-term convergence in both advanced and emerging markets.
These findings suggest that sentiment does not remain constant over time and that its impact on stock returns fluctuates based on
broader economic conditions and market events. The TVP-VAR model used in the study allows for the modelling of these time-varying
effects, providing a dynamic view of how sentiment influences stock returns over different periods. Therefore, by applying the TVP-
VAR model, we contribute to the current literature a more efficient estimation of this relationship.

4.3. Predictability towards stock returns

4.3.1. The US market

In the US, the annual and quarterly sentiment index SENTt will be compared with Baker and Wurgler sentiment index and Michigan
Consumer Sentiment Index (MCSI) subject to their predictability toward NYSE annual and quarterly returns. For comparison purposes,
we transform the BW index and MCSI monthly data into quarterly and annual data. The results are documented in Table 5 using the
ARCH estimator for time-series data. We compare the value of log-likelihood to assess the predictability of three sentiment indexes.
Regarding the predictive power of the sentiment index toward aggregate market returns, our annual investor sentiment index (SENT)
has the highest predictability compared with the other two indexes. However, the quarterly sentiment index has lower predictability
than MCSI but higher than the BW index.
Besides ARCH estimation, we employ quarterly data for stock returns and investor sentiment in the U.S. to estimate Equation (7)
under the framework of the TVP-VAR model with stochastic volatility, using 1 lag. The results are visualized in Fig. 2. Significant
coefficients are presented in grey color while insignificant ones are in red. Fig. 2a, b and 2c illustrate the coefficient distribution of

Table 2
Misvaluation in stock markets.
Misvaluation China US

Annual
Overvaluation
Proportion 44.03 % 46.43 %
Mean 0.235 0.232
Undervaluation
Proportion 55.97 % 53.57 %
Mean − 0.185 − 0.201

Quarterly
Overvaluation
Proportion 43.07 % 48.12 %
Mean 0.117 0.099
Undervaluation
Proportion 56.93 % 51.88 %
Mean − 0.089 − 0.092

9
Q. Pham et al. International Review of Economics and Finance 100 (2025) 104089

Table 3
Descriptive statistics for HEURi,t , OVERCONi,t , LAi,t and HERDi,t
Country Mean

HEURi,t OVERCONi,t LAi,t HERDi,t BIASi,t

Annually
China 0.444 0.306 3.790 0.185 0.150
US 0.460 0.124 3.140 0.174 0.067

Quarterly
China 0.527 0.597 108.097 0.073 0.060
US 0.526 0.494 88.480 0.037 0.023

Table 4
Descriptive statistics for SENT i,t 2001–2019.
Annual SENTi,t China US

Mean − 0.004 0.012

S.D 0.056 0.070
Skewness 0.328 − 0.930
Kurtosis 5.772 4.564

< ! − − Col Count : 3 − − > Quarterly SENTi,t China US

Mean 0.005 − 0.001

S.D 0.038 0.025
Skewness − 0.462 0.592
Kurtosis 10.778 4.113

Table 5
Predictability toward NYSE market returns.
Annual β z-statistic Log-likelihood

SENT 0.069 1.53 54.097

BW − 0.008 − 1.10 51.433
MCSI 0.000 0.95 53.396
FF5_SENT − 58.041 − 4.19 37.165

Quarterly β z-statistic Log-likelihood

SENT − 0.453 − 1.64 84.200

BW − 0.025 − 1.65 79.596
MCSI 0.001 1.64 84.333
FF5_SENT 1.199 0.09 40.849

SENT, BW, and MCSI, respectively, toward the U.S. stock returns. The number of insignificant coefficients (Insig. Count) amongst 5000
iterations implies the outperformance of MCSI toward predicting stock returns compared to SENT and BW. Our SENT’s predictability is
better than the BW index for the U.S. data.

4.3.2. The China market

We estimate Equation (6) in China using the ARCH estimator and the results are presented in Table 6. According to the value of log-
likelihood after estimation, our quarterly sentiment index performs better than the other two indices.
The results from TVP-VAR model estimation are visualized in Fig. 3. Our sentiment index outperforms the other two when applied
to China data, represented by the lowest number of insignificant coefficients. The robust test for predictability using the TVP-VAR
model under stochastic volatility confirms the advantage of our new sentiment index in predicting stock returns, especially in the
context of emerging markets.

4.3.3. Out-of-sample predictability

To examine the out-of-sample predictability, rather than conducting the traditional out-of-sample test, we adjust the method to use
one country’s model to forecast for another and vice versa. We estimate Equation (6) for the US and China, then record the parameters.
With three sentiment indices, we have 3 sets of model parameters. Then, we use one country’s parameters to predict the market returns
of the other. We measure the Mean Squared Error (MSE) of each sentiment index and the results are presented in Table 7. When using
the Chinese model to predict the U.S. stock market returns, our sentiment index performs significantly better than BW. When we use
the U.S. model to forecast China’s stock returns, our sentiment index has a higher MSE than the BW index.
We presented a ranking of predictability amongst three indexes in Table 8, using different methods and data frequency. Our
sentiment index ranks as number 1 and 2 across various robust tests while the other two’s ranking is unstable. The findings reflect the

10
Q. Pham et al. International Review of Economics and Finance 100 (2025) 104089

Fig. 2. Time-varying coefficients in the US.

Table 6
Predictability towards Shanghai market returns.
Annual β z-statistic Log-likelihood

SENT 0.021 0.13 35.207

BW − 0.000 − 0.08 35.202
MCSI 0.030 1.54 36.451
FF5_SENT 88.101 0.24 8.859

Quarterly β z-statistic Log-likelihood

SENT 0.655** 2.30 27.746

BW 0.000 0.11 27.623
MCSI 0.014 0.81 26.994
FF5_SENT − 39.200 − 1.52 7.081

outperformance of our sentiment index in comparison with two common ones.

4.4. Robustness test using Fama-French 5-factor model

We use the 5-factor Fama-French version of SENT as a robustness test for the predictability of our sentiment index. The results show
that our index does not outperform either the original SENT or BW index and CSI (Table 5). We believe that the reason lies on the
underlying methodology to extract misvaluation. Our original SENT extracts misvaluation as the residual of actual stock returns
relative to the predicted one from fundamental factors of firms, including earnings and labor quantity, asset growth, productivity, and
innovation. Then, sentiment level is the extent to which behavioral biases cause these residual returns after eliminating the impacts of
idiosyncratic and systematic risk. By doing this, we consider all potential determinants of stock misvaluation and we successfully
isolate the irrationality of behavioral biases. The major difference of this approach with Fama French 5-factor is that Fama French 5-
factor employs factors that already incorporate the effects of behavioral biases, therefore there is a multicollinearity problem in
Equation (5). In the presence of multicollinearity, the estimated coefficient, which is our sentiment index, will be unreliable leading to
its poor performance relative to the original method.

11
Q. Pham et al. International Review of Economics and Finance 100 (2025) 104089

Fig. 3. Time-varying coefficients in China.

Table 7
Out-of-sample predictability using MSE.
Investor sentiment index US model’s MSE Chinese model’s MSE

SENT 0.849145 0.0108102

BW 0.8440671 0.0891763
CSI 0.8494204 0.0091461

Table 8
Predictability ranking of sentiment indexes.
Index ARCH regression TVP-VAR model Out-of-sample test

Annually US China US China US China

SENT 1 2 – – – –
BW 3 3 – – – –
CSI 2 1 – – – –

Quarterly US China US China US China

SENT 2 1 1 2 2 2
BW 3 3 2 1 1 3
CSI 1 2 3 3 3 1

5. Conclusions and limitations

A review of the literature on investor sentiment measurements implies the inefficiency of existing measures. Isolating the irra­
tionality of investors is a challenge that recent sentiment measures are insufficient. This research fills this gap by proposing a new
measurement of investor sentiment, then we conduct the tests to justify the efficiency of our new sentiment index relative to two
common ones: the market-based BW index and survey-based Consumer Sentiment Index. Several conclusions can be drawn from our
findings. Firstly, regarding the predictability towards stock returns, our sentiment index performs better than the BW index in both the
US and China. Amongst the three sentiment indexes, the log-likelihood using our sentiment index is highest with annual data in the US

12
Q. Pham et al. International Review of Economics and Finance 100 (2025) 104089

and quarterly data in China. Second, when we conduct the TVP-VAR model estimation, our sentiment index performs better in China.
This conclusion is supported by the out-of-sample predictability test.
Our study, however, is not without limitations. Despite the critical role of Asian emerging markets in the global financial system,
market-based investor sentiment data for Asian emerging markets remains limited and our study only covers China and leaves out
other potential emerging markets. Future research could expand the sample to include a broader range of markets, which would allow
for a more comprehensive understanding of how the sentiment index performs globally, particularly in other Asian or European
markets. Additionally, the study uses quarterly and annual data to construct the sentiment index, which might not fully capture high-
frequency changes in sentiment. Lastly, while the new sentiment index offers a more accurate reflection of investor irrationality, there
may be limitations in the utility of market-based data. Future studies could explore the robustness of the sentiment index by applying
different methodological approaches using alternative sentiment measures, such as survey-based or text-based index. This could
potentially enhance the robustness of the findings and provide a more comprehensive measure of sentiment dynamics.
We report the mean of each variable in Equation (4) for China and the US through the period from 2001 to 2019. Equation (4) is
estimated using OLS for panel data of listed firms through the years. The residual for each firm every year is documented as MISVAL.
Listed firms in each country are separated into the Overvaluation group (positive MISVAL) and Undervaluation group (negative
MISVAL). This table reports the mean of each group and the proportion of each group in the total amount of firms.
The four common behavioral biases Heuristic, Overconfidence, Loss Aversion, and Herding behavior were estimated annually and
quarterly for each firm. The results in this table are averaged across firms and over time for each country. The bias index BIAS is the
Principal Component Analysis outcome from these four.
SENTi,t is the coefficient of equation (5) regressed at the country level every year from 2001 to 2019, significant at 10 %. SENTi,t
takes the value of 0 for insignificant coefficients.
The table reports the predictive regression results of the NYSE aggregate returns on our new sentiment index, BW sentiment index,
and MCSI as Equation (6). The regression employs the ARCH estimator.
The table reports the predictive regression results of Shanghai market returns on our new sentiment index, BW index, and Con­
sumer Sentiment Index (CSI) as Equation (6). The regression employs the ARCH estimator.
The table reports the results from the out-of-sample test for predictability. Parameters estimated from the US’s model with each
sentiment index will be used to predict Chinese market returns and vice versa. The out-of-sample predictability is measured by MSE.
The table ranks the performance of three investor sentiment indexes across various tests and methods. We use both annual and
quarterly indexes for ARCH regression. For the TVP-VAR model and out-of-sample test, we use only the quarterly index for a higher
number of observations during the observed period. Rank 1 implies best performance and rank 3 implies the worst.

Declaration of competing interest

On behalf of the authors, I declare that there is no conflict of interest in this paper.

Appendix 1Methodology and Limitation of Investor Sentiment Measures

Source Type Name of index Methodology Limitation

University of Survey, Michigan The index asks consumers about their We cannot determine whether these
Michigan consequence- Consumer expectations of three aspects: their financial expectations are rational or irrational.
based approach Confidence situation, and short-term and long-term
Index economic conditions.
Gallup Survey, U.S. Economic Two questions will be asked Americans: We cannot determine whether these
consequence- Confidence economic conditions today and whether they expectations are rational or irrational.
based approach Index are getting better. A positive index means a
positive view of the economy and vice versa.
American Survey, AAII Sentiment The survey seeks individual investors’ This index implies the current valuation of
Association of consequence- Survey opinions about stock markets in the next six investors about stock prices. In case they
Individual based approach months. The answers can be Bullish, Neutral, perceive the undervaluation, the answer will
Investors or Bearish. be bullish, otherwise is bearish. We cannot
determine whether an investor’s valuation is
rational or irrational.
​
Prosad, J. M., Survey, reason- _ Using a questionnaire to extract common The way respondents think when they do the
Kapoor, S., & based approach behavioral biases of investors: survey may be different from what they think
Sengupta, J. Overconfidence, Herding, Disposition effects. in a real investment decision. Besides, we do
(2015) not have evidence that these behavioral
biases will lead to misvaluation in stock
markets.
Oliveira, N., Cortez, Internet, _ Using stock-related texts in Tweets Twitters to A bullish and bearish attitude implies the
P., & Areal, N. consequence- generate a sentiment score representing for valuation of investors toward stock prices.
(2017) based approach bullish or bearish attitude Again, we cannot determine whether this
valuation is rational or irrational. In addition,
(continued on next page)

13
Q. Pham et al. International Review of Economics and Finance 100 (2025) 104089

(continued )
Source Type Name of index Methodology Limitation

we do not have evidence that this attitude

causing misvaluation.
Da, Z., Engelberg, Internet, FEARS Using daily Internet searches to classify This is an insufficient measure of investor
J., & Gao, P. consequence- economic keywords into positive or negative sentiment in stock markets. The authors
(2015) based approach sentiment cannot separate stock investors from the
overall search words. Moreover, a positive/
negative classification in this case does not
present for stock valuation.
Baker and Wurgler Market, BW index Determining six proxies of investor sentiment: These proxies are known as market anomalies
(2006, 2007) consequence- closed-end fund discounts, the number of that are assumed to be caused by investor
based approach IPOs, IPO initial returns, dividend premiums, sentiment. This assumption is incorrect
the ratio of equity issues to total equity and because there are other determinants of
debt issues, and market turnover. The authors market anomalies. Therefore, this approach
then use these proxies to generate a overestimates investor sentiment in stock
composite sentiment index using PCA. markets.
Hudson and Green Market, _ The methodology is similar to the BW index Again, the authors assumed these proxies
(2015) consequence- but uses a different set of sentiment proxies. represent investor sentiment but they are not.
based approach Partial least squares (PLS) is employed For example, trading volume and market
instead of PCA. volatility have several determinants and
investor sentiment is only one of them.
Neely, C. J., Market, reason- – Using the predictability of historical data The dependence on information from
Rapach, D. E., based approach toward stock prices to capture investor historical data to make current valuation is
Tu, J., & Zhou, sentiment in the type of Heuristic rational as long as investors do not
G. (2014) overweight this information over
fundamental factors. If we do not have
evidence that this heuristic leads to
misvaluation, we cannot equate heuristic
with irrationality.
Blackburn and Market, reason- – Using short-term momentum and long-term There is evidence that this bias does not
Cakici (2017) based approach reversal as indicators of overconfidence and always lead to inefficient markets.
conservatism Overconfidence may induce investors to
trade more aggressively with their private
signals that ultimately bring more
information into stock prices. Conservatism
makes investors wait longer for price
convergence, which reverses the effects of
limit to arbitrage. In this case, this bias is not
irrational.
Statman et al. Market, reason- ​ Using the coefficient between lagged returns Loss aversion will cause the phenomenon that
(2006) based approach and stock turnover to demonstrate the level of winning shares are much more sold than
loss aversion. The positive coefficient losing shares. In case these winning shares
demonstrates that inactive status follows a and losing shares are misevaluated, loss
loss, while excessive trading follows a gain. aversion reverses the misvaluation. Then, loss
aversion reduces misvaluation rather than
leading to misvaluation.
Chiang and Zheng Market, reason- ​ Using cross-sectional stock returns dispersion The low dispersion of stock returns does not
(2010) based approach to capture herding behavior. The lower the always imply herding behavior. When there
dispersion, the higher the degree of herding at is a systematic shock that affects market-wide
the market level. stocks, this low dispersion of returns will
happen but this herding is not a sign of
irrationality.

Data availability

Data will be made available on request.

References

Adams, M., Thornton, B., & Hall, G. (2008). IPO pricing phenomena: Empirical evidence of behavioral biases. Journal of Business & Economics Research, 6(4).
Aggarwal, D. (2019). Defining and measuring market sentiments: A review of the literature. Qualitative Research in Financial Markets, 14(2), 270–288.
Allen, E., Larson, C., & Sloan, R. (2013). Accrual reversals, earnings, and stock returns. Journal of Accounting and Economics, 56(1), 113–129.
Baker, H. K., Kumar, S., Goyal, N., & Gaur, V. (2019). How financial literacy and demographic variables relate to behavioral biases. Managerial Finance, 45(1),
124–146.
Baker, M., & Wurgler, J. (2003). A conditional characteristics model of stock returns. Working paper, HBS, and NBER.
Baker, M., & Wurgler, J. (2007). Investor sentiment in the stock market. The Journal of Economic Perspectives, 21(2), 129–152.
Baker, M., Wurgler, J., & Yuan, Y. (2012). Global, local, and contagious investor sentiment. Journal of Financial Economics, 104(2).
Basak, S. (2005). Asset pricing with heterogeneous beliefs. Journal of Banking & Finance, 29(11), 2849–2881.

14
Q. Pham et al. International Review of Economics and Finance 100 (2025) 104089

Beer, F., & Zouaoui, M. (2013). Measuring stock market investor sentiment. Journal of Applied Business Research, 29(1), 51–68.
Belo, F., Lin, X., & Bazdresch, S. (2014). Labor hiring, investment, and stock return predictability in the cross-section. Journal of Political Economy, 122(1), 129–177.
Bhamra, H., & Uppal, R. (2014). Asset prices with heterogeneity in preferences and beliefs. Review of Financial Studies, 27(2), 519–580.
Black, F. (1986). Noise. The Journal of Finance, 41(3), 528–543.
Blackburn, D., & Cakici, N. (2017). Overreaction and the cross-section of returns: International evidence. Journal of Empirical Finance, 42, 1–14.
Bournakis, I., & Mallick, S. (2018). TFP estimation at the firm level: The fiscal aspect of productivity convergence in the UK. Economic Modelling, 70, 579–590.
Campbell, J., & Kyle, A. (1993). Smart money, noise trading, and stock price behavior. The Review of Economic Studies, 60(1), 1–34.
Cao, J., & Han, B. (2016). Idiosyncratic risk, costly arbitrage, and the cross-section of stock returns. Journal of Banking & Finance, 73, 1–15.
Caporale, G., You, K., & Chen, L. (2019). Global and regional stock market integration in Asia: A panel convergence approach. International Review of Financial
Analysis, 65, 101–381.
Chan, J. C. (2022). Large hybrid time-varying parameter VARs. Journal of Business & Economic Statistics, 1–16.
Chen, H., Chong, T., & She, Y. (2014). A principal component approach to measuring investor sentiment in China. Quantitative Finance, 14(4), 573–579.
Chiang, T., & Zheng, D. (2010). An empirical analysis of herd behavior in global stock markets. Journal of Banking & Finance, 34(8), 1911–1921.
Chien, M., Lee, C., Hu, T., & Hu, H. (2015). Dynamic Asian stock market convergence: Evidence from dynamic cointegration analysis among China and ASEAN-5.
Economic Modelling, 51, 84–98.
Chu, X., Wu, C., & Qiu, J. (2016). A nonlinear granger causality test between stock returns and investor sentiment for Chinese stock market: A wavelet-based
approach. Applied Economics, 48(21), 1915–1924.
Chun, H., Kim, J., & Morck, R. (2016). Productivity growth and stock returns: Firm-and aggregate-level analyses. Applied Economics, 48(38), 3644–3664.
Chung, S., Hung, C., & Yeh, C. (2012). When does investor sentiment predict stock returns? Journal of Empirical Finance, 19(2), 217–240.
Cooper, M., Gulen, H., & Schill, M. (2008). Asset growth and the cross-section of stock returns. The Journal of Finance, 63(4), 1609–1651.
Da, Z., Engelberg, J., & Gao, P. (2015). The sum of all FEARS investor sentiment and asset prices. Review of Financial Studies, 28(1), 1–32.
Daniel, K., Hirshleifer, D., & Subrahmanyam, A. (2005). Investor psychology and tests of factor pricing models. Available at: SSRN 854024.
Dash, S. R., & Maitra, D. (2018). Does sentiment matter for stock returns? Evidence from Indian stock market using wavelet approach. Finance Research Letters, 26,
32–39.
De Long, J. B., Shleifer, A., Summers, L. H., & Waldmann, R. J. (1990). Noise trader risk in financial markets. Journal of Political Economy, 98(4), 703–738.
De Long, J. B., Shleifer, A., Summers, L. H., & Waldmann, R. J. (1991). The survival of noise traders in financial markets. Journal of Business, 64(1), 1–19.
Di Guilmi, C., He, X. Z., & Li, K. (2014). Herding, trend-chasing, and market volatility. Journal of Economic Dynamics and Control, 48, 349–373.
Diether, K., Malloy, C., & Scherbina, A. (2002). Differences of opinion and the cross-section of stock returns. The Journal of Finance, 57(5), 2113–2141.
Doukas, J., Kim, C., & Pantzalis, C. (2010). Arbitrage risk and stock mispricing. Journal of Financial and Quantitative Analysis, 45(4), 907–934.
Easley, D., de Prado, M., & O’Hara, M. (2016). Discerning information from trade data. Journal of Financial Economics, 120(2), 269–285.
Fama, E. (1970). Efficient capital markets: A review of theory and empirical work. The Journal of Finance, 25(2), 383–417.
Garcia, D. (2013). Sentiment during recessions. The Journal of Finance, 68(3), 1267–1300.
Gigerenzer, G., & Gaissmaier, W. (2011). Heuristic decision-making. Annual Review of Psychology, 62(1), 451–482.
Gray, P., & Johnson, J. (2011). The relationship between asset growth and the cross-section of stock returns. Journal of Banking & Finance, 35(3), 670–680.
Greenwood, R., & Shleifer, A. (2014). Expectations of returns and expected returns. Review of Financial Studies, 27(3), 714–746.
Grossman, S. (1976). On the efficiency of competitive stock markets where trades have diverse information. The Journal of Finance, 31(2), 573–585.
Guo, H., & Savickas, R. (2006). Idiosyncratic volatility, stock market volatility, and expected stock returns. Journal of Business & Economic Statistics, 24(1), 43–56.
Hansson, B. (2004). Human capital and stock returns: Is the value premium an approximation for return on human capital? Journal of Business Finance & Accounting,
31(3), 333–358.
Herskovic, B., Kelly, B., Lustig, H., & Van Nieuwerburgh, S. (2016). The common factor in idiosyncratic volatility: Quantitative asset pricing implications. Journal of
Financial Economics, 119(2), 249–283.
Hirshleifer, D., Hsu, P., & Li, D. (2013). Innovative efficiency and stock returns. Journal of Financial Economics, 107(3), 632–654.
Hotelling, H. (1933). Analysis of a complex of statistical variables into principal components. Journal of Educational Psychology, 24(6), 417.
Huang, D., Jiang, F., Tu, J., & Zhou, G. (2015). Investor sentiment aligned: A powerful predictor of stock returns. Review of Financial Studies, 28(3), 791–837.
Huang, A., & Wirjanto, T. (2012). Is China’s P/E ratio too low? Examining the role of earnings volatility. Pacific-Basin Finance Journal, 20(1), 41–61.
Hudson, Y., & Green, C. (2015). Is investor sentiment contagious? International sentiment and UK equity returns. Journal of Behavioral and Experimental Finance, 5,
46–59.
Hwang, S., Rubesam, A., & Salmon, M. (2021). Beta herding through overconfidence: A behavioral explanation of the low-beta anomaly. Journal of International Money
and Finance, 111, 102–318, 2021.
Hwang, S., & Salmon, M. (2004). Market stress and herding. Journal of Empirical Finance, 11(4), 585–616.
İmrohoroğlu, A., & Tüzel, Ş. (2014). Firm-level productivity, risk, and return. Management Science, 60(8), 2073–2090.
Jackson, M. (1991). Equilibrium, price formation, and the value of private information. Review of Financial Studies, 4(1), 1–16.
Jegadeesh, N., & Titman, S. (2001). Profitability of momentum strategies: An evaluation of alternative explanations. The Journal of Finance, 56(2), 699–720.
Joseph, K., Wintoki, M., & Zhang, Z. (2011). Forecasting abnormal stock returns and trading volume using investor sentiment: Evidence from online search.
International Journal of Forecasting, 27(4), 1116–1127.
Kahneman, D., Knetsch, J., & Thaler, R. (1991). Anomalies: The endowment effect, loss aversion, and status quo bias. The Journal of Economic Perspectives, 5(1),
193–206.
Kahneman, D., & Tversky, A. (1979). On the interpretation of intuitive probability: A reply to jonathan cohen. Cognition, 7(4), 409–411.
Kim, S., & Kim, D. (2014). Investor sentiment from internet message postings and the predictability of stock returns. Journal of Economic Behavior & Organization, 107,
708–729.
Koop, G., & Korobilis, D. (2013). Large time-varying parameter VARs. Journal of Econometrics, 177(2), 185–198.
Kothari, S., Lewellen, J., & Warner, J. (2006). Stock returns, aggregate earnings surprises, and behavioral finance. Journal of Financial Economics, 79(3), 537–568.
Kuehn, L., Simutin, M., & Wang, J. (2017). A labor capital asset pricing model. The Journal of Finance, 72(5), 2131–2178.
Kumar, A. (2009). Hard-to-value stocks, behavioral biases, and informed trading. Journal of Financial and Quantitative Analysis, 44(6), 1375–1401.
Kumar, P., & Li, D. (2016). Capital investment, innovative capacity, and stock returns. The Journal of Finance, 71(5), 2059–2094.
Kurov, A. (2010). Investor sentiment and the stock market’s reaction to monetary policy. Journal of Banking & Finance, 34(1), 139–149.
Lamont, O. (2000). Investment plans and stock returns. The Journal of Finance, 55(6), 2719–2745.
Le, T. H., Boubaker, S., Bui, M. T., & Park, D. (2023). On the volatility of wti crude oil prices: A time-varying approach with stochastic volatility. Energy Economics,
117, 106–474.
Li, D. (2011). Financial constraints, R&D investment, and stock returns. Review of Financial Studies, 24(9), 2974–3007.
Li, Y., & Yang, L. (2013). Prospect theory, the disposition effect, and asset prices. Journal of Financial Economics, 107(3), 715–739.
Lin, X. (2012). Endogenous technological progress and the cross-section of stock returns. Journal of Financial Economics, 103(2), 411–427.
Lucas, J. R. (1988). On the mechanics of economic development. Journal of Monetary Economics, 22(1), 3–42.
Marczak, M., & Beissinger, T. (2016). Bidirectional relationship between investor sentiment and excess returns: New evidence from the Wavelet perspective. Applied
Economics Letters, 23(18), 1305–1311.
McLean, R. (2010). Idiosyncratic risk, long-term reversal, and momentum. Journal of Financial and Quantitative Analysis, 45(4), 883–906.
McLean, R., & Zhao, M. (2014). The business cycle, investor sentiment, and costly external finance. The Journal of Finance, 69(3), 1377–1409.
Mendel, B., & Shleifer, A. (2012). Chasing noise. Journal of Financial Economics, 104(2), 303–320.
Neely, C., Rapach, D., Tu, J., & Zhou, G. (2014). Forecasting the equity risk premium: The role of technical indicators. Management Science, 60(7), 1772–1791.

15
Q. Pham et al. International Review of Economics and Finance 100 (2025) 104089

Pantzalis, C., & Park, J. (2009). Equity market valuation of human capital and stock returns. Journal of Banking & Finance, 33(9), 1610–1623.
Pearson, K. (1901). Principal components analysis. The London, Edinburgh and Dublin Philosophical Magazine and Journal of Science, 6(2), 559.
Pontiff, J. (2006). Costly arbitrage and the myth of idiosyncratic risk. Journal of Accounting and Economics, 42(1), 35–52.
Primiceri, G. E. (2005). Time-varying structural vector autoregressions and monetary policy. The Review of Economic Studies, 72(3), 821–852.
Prosad, J. M., Kapoor, S., & Sengupta, J. (2015). Behavioral biases of Indian investors: A survey of Delhi-NCR region. Qualitative Research in Financial Markets, 7(3),
230–263.
Ramiah, V., Xu, X., & Moosa, I. (2015). Neoclassical finance, behavioral finance, and noise traders: A review and assessment of the literature. International Review of
Financial Analysis, 41, 89–100.
Riahi-Belkaoui, A., & Picur, R. D. (1994). Explaining market returns earnings versus value-added data. Managerial Finance, 20(9), 44–55.
Romer, P. (1994). The origins of endogenous growth. The Journal of Economic Perspectives, 8(1), 3–22.
Rubinstein, M. (2001). Rational markets: Yes or no? The affirmative case. Financial Analysts Journal, 57(3), 15–29.
Scheinkman, J., & Xiong, W. (2003). Overconfidence and speculative bubbles. Journal of Political Economy, 111(6), 1183–1220.
Schwert, G. (1989). Why does stock market volatility change over time? The Journal of Finance, 44(5), 1115–1153.
Sharma, A., & Kumar, S. (2010). Economic value added (EVA)-literature review and relevant issues. International Journal of Economics and Finance, 2(2), 200–220.
Shen, J., Yu, J., & Zhao, S. (2017). Investor sentiment and economic forces. Journal of Monetary Economics, 86, 1–21.
Shiller, R., Fischer, S., & Friedman, B. (1984). Stock prices and social dynamics. Brookings Papers on Economic Activity, 1984(2), 457–510.
Solow, R. (1999). Neoclassical growth theory. Handbook of Macroeconomics, 1, 637–667.
Stambaugh, R., Yu, J., & Yuan, Y. (2012). The short of it: Investor sentiment and anomalies. Journal of Financial Economics, 104(2), 288–302.
Stambaugh, R., Yu, J., & Yuan, Y. (2015). Arbitrage asymmetry and the idiosyncratic volatility puzzle. The Journal of Finance, 70(5), 1903–1948.
Statman, M., Thorley, S., & Vorkink, K. (2006). Investor overconfidence and trading volume. Review of Financial Studies, 19(4), 1531–1565.
Sun, L., Najand, M., & Shen, J. (2016). Stock return predictability and investor sentiment: A high-frequency perspective. Journal of Banking & Finance, 73, 147–164.
Tanl, Z. H., Gan, C., & Li, Z. (2010). An empirical analysis of the Chinese stock market: Overvalued/undervalued. International Journal, 18(1), 44–74.
Titman, S., Wei, K., & Xie, F. (2009). Capital investments and stock returns in Japan. International Review of Finance, 9(1), 111–131.
Vuolteenaho, T. (2002). What drives firm-level stock returns? The Journal of Finance, 57(1), 233–264.
Whaley, R. E. (2000). The investor fear gauge. Journal of Portfolio Management, 26(3), 12.
Wübben, M., & Wangenheim, F. (2008). Instant customer base analysis: Managerial heuristics often “get it right”. Journal of Marketing, 72(3), 82–93.
Yu, J. (2012). A semiparametric stochastic volatility model. Journal of Econometrics, 167(2), 473–482.
Zaremba, A., Kizys, R., & Raza, M. (2020). The long-run reversal in the long run: Insights from two centuries of international equity returns. Journal of Empirical
Finance, 55, 177–199.
Zhang, X. (2006). Information uncertainty and stock returns. The Journal of Finance, 61(1), 105–137.
Zhou, G. (2018). Measuring investor sentiment. Annual Review of Financial Economics, 10, 239–259.
Zouaoui, M., Nouyrigat, G., & Beer, F. (2011). How does investor sentiment affect stock market crises? Evidence from panel data. Financial Review, 46(4), 723–747.

16