Investor Sentiment and Stock Return: Do Industries Matter?

Abstract

The basic objective of this article is to evaluate the pricing implications of market-wide investor sentiment risk for cross-sectional return variations of Indian listed companies across industry groups. A multivariate time-series regression approach has been used to examine the impact of sentiment risk on stock return behaviour in the presence of other market-wide systematic risk factors. Our results suggest that the role of sentiment risk in the determination of a cross-section of stock returns is not uniform across the test asset portfolios formed on the basis of size, book-to-market equity, liquidity and momentum characteristics. For all portfolios, the impact of sentiment risk on the cross-section of stock returns behaviour has been disproportionately negative. The effect holds even after controlling for systematic market-wide risk factors. Although the impact of sentiment risk on industry-shorted portfolio returns persists in accordance with the theoretical argument, the cross-sectional variation with respect to different industries has been heavily dependent on the availability of stocks in that particular industry. The commonality of the sentiment effect across industry is not similar, as it is for the aggregate market. The results suggest that generalisation of the hard-to-value and difficult-to-arbitrage argument must be judged with caution, keeping the industry effects in mind.

JEL Classification: GI,G12,G14

Keywords

Sentiment Risk Multi Factor Model Liquidity Risk Momentum Strategy Stock Return

1. INTRODUCTION

Traditional asset pricing theories assume that, under efficient equilibrium market conditions, stock prices will always confirm to their fundamental value and any effect of irrationality or sentiment related mispricing among market participants will be eliminated by profit-seeking rational investors or arbitrageurs. However, over the past decades the disastrous socioeconomic consequences of market crashes (see for example, Daniel et al., 2002; Shiller et al., 1984 for a detailed discussion) and the failure of traditional theories to provide convincing explanations for such market behaviour have made both academic and practitioners to question the complete rationality assumption of investor behaviour. As an alternative to the traditional finance paradigm, the behavioural approach to asset-pricing seeks to answer such inconsistent market phenomena by considering market participants imperfectly rational rather fully rational. Following the arguments of animal spirits (Keynes, 1936) and noise trading (Black, 1986; De Long et al., 1990) the proponents of behavioural asset pricing theories try to analyse the role of investor sentiment in the context of irrational exuberance (Shiller, 2000) of the market with unanticipated boom and bust cycles.

In a more formal way, investor sentiment in the behavioural asset-pricing literature has been defined as the systematic error or biases in investor belief about future cash flows and investment risks that are inconsistent with the fundamental facts (Baker and Wurgler, 2006; Shefrin, 2005). The behavioural approach to asset pricing suggests two theoretical arguments towards the sentiment-driven mispricing in financial markets: first, the role of psychological biases (Kahneman and Tversky, 1979) in investor behaviour; second, the costly arbitrage opportunity and short-sell constraints for determining stock prices (Shleifer and Vishny, 1997). Motivated by experimental psychology literature, the behavioural asset pricing theory assumes that investors are not completely rational rather, normal and systematic cognitive biases such as self-attribution, heuristics and representativeness in their decision-making process (see for example, Baker and Nofsinger, 2002; Barberis and Thaler, 2003; Hirshliefer, 2001; Rabin, 1998; Ritter, 2003; Stracca, 2004 for detail review of literature) induce them to invest based on instinct and not on fundamentals.

Following such fundamental arguments of sentiment-driven mispricing, the behavioural approach to asset-pricing attempts to incorporate investor sentiment as an additional source of risk in the well-established multifactor framework. The primary motivation in this regard is to demonstrate the fact that, the expected risk premium on a security is generally the sum of a fundamental component and a sentiment premium (Shefrin, 2005). Earlier theoretical work such as Black (1986), De Long et al. (1990), Shleifer and Summers (1990), Barberis et al. (1998), Daniel et al. (1998), have more formally validated that noise trader risk is a responsible factor for the deviation of equity prices from their fundamental values and investors in the capital market bear both systematic and sentiment risk induced by noise traders. Building upon such theoretical literature and following the top-down approach of Baker and Wurgler (2006), recent literature has investigated empirically the explanatory power of sentiment risk for the cross-section of stock return behaviour in alternative multifactor specifications. According to Baker and Wurgler (2006, 2007) stocks that are the hardest to arbitrage and whose valuations are more subjective, are also most vulnerable to sentiment risk. In common, using alternative sentiment proxies, the existing literature supports a negative relationship between individual investor sentiment and stock returns across different markets (Brown and Cliff, 2004, 2005; Baker and Wurgler, 2006, 2007; Baker et al., 2011; Changsheng and Yongfeng, 2012; Dash and Mahakud, 2012; Fisher and Statman, 2000; Finter et al., 2011; Kumar and Lee, 2006; Schmeling, 2009; Schmeling, 2009).

Although the available empirical literature supports the pricing implication of sentiment risk across several developed markets, there are at least two issues that cast doubt on the generalisation of sentiment pricing evidence. First, despite its popularity in practice, industry analysis has never been explored in the behavioural asset pricing literature. From the practitioner’s perspective, active investment strategy explores tactical asset allocation with respect to the industry analysis. Although behavioural asset pricing researchers and active investment practitioners share a common belief of inefficient markets, the commonality of sentiment pricing across industry groups has never been a source of research. Since, in a financial market, all assets are perfect substitutes and the supply or demand of an asset is infinitely elastic, the broad wave of sentiment effect and uninformed demand or supply shocks at the aggregate market level may have a subtle effect across different industry groups.

Second, the proponents of behavioural asset pricing suggest that the cross-sectional structure of characteristics-based mispricing with respect to the size, value, momentum and liquidity effects stem from investor bahavioral biases (Baker and Wurgler, 2006; Shefrin, 2005). However, available empirical evidence emphasises the fact that even though characteristics are priced in other markets apart from the US market, they have not been uniformly priced by investors. For instance, uniformity in the pricing of size and book-to-market equity effects has been questioned by a number of scholars in emerging markets (Chui and Wei, 1998; Classens et al., 1995, 1998). In this regard it can be argued that, as these characteristics are not uniformly priced across markets, the argument for their uniform sentiment sensitivity may not carry the same degree of consistency in emerging markets as compared to other developed markets. Moreover, little attention has been given to liquidity and momentum characteristics for the investigation of possible sentiment mispricing.

Keeping these arguments in mind, our objectives are to investigate the pricing implications of investor sentiment for the cross-section of stock returns behaviour at an aggregate level and across several industry groups in the Indian stock market after controlling the effects of other market-wide risk factors. For the empirical analysis, by using a multivariate time series approach, we first try to analyse the sentiment pricing effects for single-shorted portfolios constructed across the four well-debated characteristics of size, book-to-market equity, liquidity and momentum. We also check the robustness of sentiment pricing evidence across different industry groups by using single-shorted portfolios with respect to the above-mentioned four characteristics.

In this regard, the contribution of this article to existing asset pricing literature has been two-fold. First, it provides the first-ever evidence on the pricing implications of sentiment risk across several industry groups in an augmented five-factor model specification. In their seminal paper Baker and Wurgler (2006) conjecture that, the influence of investor sentiment on stock mispricing can be materialised through two distinctive channels. ‘In the first channel, sentimental demand shocks vary in the cross section, while arbitrage limits are constant. In the second, the difficulty of arbitrage varies across stocks, but sentiment is generic (Baker and Wurgler, 2006, p. 1648)’. Our two broad set of objectives help us revisit these two distinct channels through which investor sentiment may affect the cross-section of stock returns behaviour. With the first objective, we explore the possibility that the sentiment effect varies across the cross-sectional return variation of several alternative characteristic-based portfolios by assuming that arbitrage limits are constant. Following the results of our first objective that sentiment risk is generic or systematic, in our subsequent analysis we try to explore the possibility of sentiment pricing across different industry groups by assuming that arbitrage constraints vary across industry groups, given the perfectly elastic nature of financial assets. Second, considering a comprehensive set of sentiment proxies, it documents out-of-sample evidence from an emerging market for the pricing implications of sentiment risk across several alternative test asset portfolios. In contrast to prior literature, we provide additional insight into this issue by considering the liquidity factor as a control variable in the multifactor specification. We also extend our empirical analysis in the form of a robustness test to examine whether investor sentiment ‘Granger cause’ test asset portfolio returns and vice versa. Further, all the available literature on sentiment-based risk-pricing evidence has been associated with markets that are characterised by a high level of individual investor participation. The out-of-sample evidence from an emerging stock market with a high level of national and international institutional investor participation can be a more fundamental validation of sentiment risk evidence.

The remainder of this article is structured as follows. The second section elaborates the empirical approach, the third section presents data and variables, the fourth section discusses the empirical results and finally the last section concludes.

2. EMPIRICAL APPROACH

This section has been divided into two parts. The first part describes the empirical approach which tests the pricing implication of sentiment risk after controlling the impact of other systematic market-wide risk factors. In order to test the pricing implication of sentiment risk, we employ the first step Fama and Macbeth (1973) time series regression approach. The second part elaborates the approach for testing the causal link between the sentiment risk and portfolio returns using the Granger’s causality test (1969, 1988). The second approach complements the first step. After testing the pricing implication of sentiment risk in the first step, in the second step we test the causal link between the test asset portfolio returns and the sentiment risk.

2.1 Linear Factor Model Approach

Following the multifactor asset pricing framework, a linear factor model characterises the security payoff with the following factor structure:

R_{i} = E (R_{i}) + b'_{i} \tilde{f} + e_{i}

(1)

Where, f˜ is a K-by-1 vector of demanded factors, E(ε_i) = 0, Ef ε_i = 0, and importantly var(ε_i) → 0.

However, under the argument of the possible impact of sentiment-driven changes in the cross-sectional predictability pattern, the possible inclusion of a sentiment risk component leads to the following specification:

E_{t - 1} [R_{i t}] = α + \sum_{j = 1}^{k} β_{i j} (F_{j t}) + μ_{i} ϕ_{t - 1}

(2)

Where, F_j is the excess return on the j_th factor at time t; $β_{i j}$ is security i’s (i = 1….N) loading on factor j; μ_i indicates the security i’s sensitivity on the sentiment risk component $ϕ_{t - 1}$ .

Under the null hypothesis that μ_i equals zero, the expected return of security i can only be determined by the k number of systematic risk factors. Under the alternative, if μ_i is non-zero and negative, then the sentiment component $ϕ_{t - 1}$ reveals cross-sectional patterns in sentiment-driven mispricing. The negative non-zero sign of μ_i follows the theoretical argument that, if previous time periods’ sentiment-driven mispricing is high for a certain security, then the expected future return will be low. Under the condition of exact pricing, equation (2) suggests that a security’s risk premium, that is, its excess return over the risk-free rate (R_i – R_f), is the sum of a fundamental-based premium and a sentiment-based premium. In other words, when the sentiment premium is largely relative to the fundamental premium, risk premiums reflect both mispricing and compensation for bearing additional sentiment-based risk. To test this hypothesis we consider the following time series specification of an augmented five-factor model with the sentiment risk component:

R_{p t} - R_{f t} = α_{t} + β_{S e n t_{p}} S e n t_{t - 1} + \sum_{k = 1}^{5} β_{p k} F F M_{t} + ε_{p t}

(3)

Where, R_pt represents the equal weighted portfolio return of the test asset portfolios constructed on firm characteristics. R_ft is the risk-free rate of return proxied by the 91-day Treasury bill rate. R_pt – R_ft indicates excess return on the characteristic-based portfolios and $β_{p k}$ represents the factor sensitivities of test asset p for the selected five market-wide risk factors.

FFM shows the vector containing the excess returns of five market-wide risk factors, namely, MRKT (market excess return), SMB (small minus big), HML (high minus low),UMD (up minus down) and LMHL (low liquid minus high liquid). The FFM derives its motivation from the Fama and French (1993) three-factor model (MRKT, SMB, HML), Carhart’s (1997) four-factor model that augments the three-factor model with the momentum factor (UMD) and the five-factor model that extends the four-factor model with a liquidity factor (LMHL). We describe the detailed measurement of these risk factors in the subsequent section. Empirical consistency for the explanation of cross-section of stock returns behaviour of these risk factors in several markets is the basic motivation for the inclusion of such factors in our analysis (Dash and Mahakud, 2012; Her et al., 2004; Keene and Peterson, 2007; Lam and Tam, 2011; Pastor and Stambaugh, 2003; Sehgal and Jain, 2011). In a more elaborative form, equation (3) can be specified as:

\begin{array}{l} R_{p t} - R_{f t} = a_{t} + b_{S e n t_{p}} S e n t_{t - 1} + b_{M R K T_{p}} M R K T_{t} + \\ b_{S M B_{P}} S M B_{t} + b_{H M L_{P}} H M L_{t} + b_{U M D_{P}} U M D_{t} + b_{L M H L_{P}} L M H L_{t} + e_{p t} \end{array}

(4)

Given our objective of examining the importance of sentiment risk on the cross-section of stock return behaviour after controlling for the effects of the five fundamental risk factors (MRKT, SMB, HML, UMD, LMHL), we restrict our model to the first-step time series regression. We measure sentiment risk (Sent_t_–1) by constructing an aggregate Indian investor sentiment index. As we have discussed earlier under the null hypothesis, stock returns are only determined by fundamental risk factors and that sentiment component should not enter the regression significantly. We follow the augmented five-factor model, augmented with the sentiment component in our subsequent analysis, for both the aggregate market level and across the industry groups.

2.2 Granger Causality Test

This section elaborates the empirical approach, concentrating on the causality test between portfolio returns (R_p) and sentiment risk (Sent). The concept of Granger’s causality test (1969, 1988) examines the dynamic linkage between the two series. A time series x_t Granger-causes another time series y_t, if the series y_t can be predicted with better accuracy by using past values of x_t rather than by not doing so. In our empirical analysis the causality between sentiment index (Sent) and portfolio return (R_p) is tested using a bivariate vector autoregression (VAR) model of the following kind:

R_{p t} = a + \sum_{j = 1}^{k} m_{j} R_{p t - j} + \sum_{j = 1}^{k} g_{j} S e n t_{t - j} + u_{x t}

(5)

S e n t_{t} = a + \sum_{j = 1}^{k} m_{j} R_{p t - j} + \sum_{j = 1}^{k} g_{j} S e n t i_{t - j} + u_{x t}

(6)

Where, k is a suitably chosen positive integer with j = 0, 1… k; μ_i and γ_j are parameters, α’s are constants and u_t’s are disturbance terms with zero means and finite variances. The null hypothesis that Sent_t does not Granger-cause (→) R_pt is not accepted if the γ_j (j > 0) in equation (5) and (6) is jointly significantly different from zero using a standard joint test (for example, an F test). Similarly, R_pt Granger-causes Sent_t if the μ_j, j > 0 coefficients in equation (6) are jointly different from zero. Based on the standard econometric approach of Granger’s causality test, the causal relationship may be unidirectional causality (μ_j≠ 0, γ = 0 or γ_j≠ 0, μ = 0), bidirectional, that is, feedback causality (μ_j≠ 0, γ ≠ 0), or no Granger causality (μ_j≠ 0, γ ≠ 0) in any direction. In principle, the causality test refers that Sent is a cause of R_p if it is useful in forecasting R_p. Put differently, Sent is able to increase the accuracy of the prediction of R_p with respect to a forecast, considering only past values of R_p.

3. DATA AND VARIABLES

The basic data consists of monthly returns and other firm-specific risk characteristics of NSE-listed (National Stock Exchange of India) companies for the period January 2003 to March 2011. The choice of the sample period is conditioned upon data availability for construction of the sentiment index. The S&P CNX Nifty has been taken as the market proxy. The 91-days Treasury bill rate is taken as the proxy for a risk-free rate. While considering the S&P CNX Nifty as the market proxy, we assume implicitly that the Indian stock market is segmented and the market risk premium is priced because of local macroeconomic and firm-specific factors (Bekaert and Harvey, 2003; Misra and Mahakud, 2009). For the construction of test asset portfolios and market-wide risk factors we begin portfolio formation at the beginning of September every year to compare accounting data for the year ending March of year y, with stock returns from September of year y to August of year y+1. The purpose of giving a period a five-month lag for the disclosure of available accounting information to market participants is to avoid the look-ahead bias. Controlling for various stock selection criteria as discussed in Fama and French (1992, 1993) and Amihud (2002) there are between 676 (September 2002) and 1,156 (September 2010) companies available for our analysis. The required data on stock returns and other firm-specific information have been collected from the Centre for Monitoring the Indian Economy (CMIE) PROWESS database. Data for the risk-free rate and macroeconomic factors have been collected from the Reserve Bank of India (RBI) website. Data for the sentiment proxies have been collected from the NSE, Securities and Exchange Board of India (SEBI) and the Association of Mutual Funds in India (AMFI) websites.

3.1 Construction of the Sentiment Index

Prior, related literatures support three different methods to estimate the unobservable sentiment variable. First, is based on the survey method of individual investor responses to anticipated movements of the stock market and aggregate economy (Fisher and Statman, 2000; Schmeling, 2009). The second approach is the implicit sentiment proxy (ISP) derived from selected market statistics (Baker and Wurgler, 2006, 2007; Brown and Cliff, 2004, 2005; Dash and Mahakud, 2012). The third approach is a combination of both the explicit and implicit sentiment proxies in the form of a composite sentiment measure. Although there is no uncontroversial and universal proxy for measuring investor sentiment, our approach for constructing the Indian investor sentiment index (Sent_t) closely follows the top-down approach of Baker and Wurgler (2006, 2007). As there is no theoretical argument as to the exact number of proxies that can be considered for constructing the sentiment index, we closely follow the aggregate sentiment index constructed by Dash and Mahakud (2012) by using the common variations in 11 ISPs. Following the approach of Dash and Mahakud (2012), the selected 11 ISPs are: turnover volatility ratio (tvr), buy-sell imbalance ratio (bsir), put-call ratio (pcr), advance decline ratio (adr), share turnover velocity (stv), number of IPOs (nipo), equity issue in total issue (eiti), dividend premium (div.P), change in margin borrowing (cmb), fund flow (ff) and cash-to-total assets (cta) in the mutual fund market.

It has been argued in related literature that when an investor is bullish or bearish, this could be a rational expectation of the future period’s expected cash flow, an irrational optimism, or a combination of both (Baker and Wurgler, 2006, 2007; Brown and Cliff, 2004, 2005). Therefore, it is likely that each of the ISPs may include a non-fundamental and fundamental component. Since our objective is to deal with the irrational component of sentiment, following related literature (Baker and Wurgler, 2006, 2007; Brown and Cliff, 2004, 2005; Verma and Soydemir, 2009) we have tried to circumvent this problem by regressing each of the 11 proxies on certain fundamental factors. Consistent with related literature, the selected fundamental factors are the industrial production growth rate, term spread, exchange rate, rate of inflation, per cent change in net FII inflows and five systematic market-wide risk factors—MRKT, SMB, HML, LMHL and UMD. The fitted values of such regressions are considered to capture the rational component of sentiment and the residuals are assumed to be cleaner proxies for the irrational component of investor sentiment (Verma and Soydemir, 2009). We call these modified proxies the orthogonal sentiment proxies. After making the selected raw sentiment proxies orthogonal to the fundamental factors, we employ principal component analysis for constructing our sentiment index. The principal component analysis filters out idiosyncratic noise in the orthogonal sentiment measures and captures their common component. The first principal component, with 42 per cent of the sample variance, gives the following measure of our sentiment index:

\begin{array}{l} S e n t_{t} = 0.328 (t v r_{t - 1}) + 0.394 (s t v_{t - 1}) + 0.060 (b s i r_{t}) - 0.164 (p c r_{t - 1}) \\ + 0.194 (a d r_{t - 1}) + 0.209 (n i p o_{t}) + 0.213 (e i t i_{t}) - 0.053 (d i v . P_{t}) \\ + 0.273 (c m b_{t - 1}) + 0.025 (f f_{t - 1}) - 0.308 (c t a_{t - 1}) \end{array}

(7)

It has been also argued that some ISPs may reflect a shift in aggregate market-wide sentiment earlier than others (Baker and Wurgler, 2006). In this regard, the time subscript in equation (7) shows the relative timing of each orthogonal ISPs. The respective lead-lag relationship of each ISP with aggregate market-wide investor sentiment is indentified by the following steps. The first step is, before validating the final sentiment index, a raw sentiment index has been constructed by estimating the first principal component of 11 orthogonal ISPs and their lags. This results in a first-stage raw sentiment index composed of 22 loadings for each of the current and lagged values of orthogonal ISPs. In the second step, we compute the correlation between the first-stage raw sentiment index and the current and lagged value of each of the orthogonal ISPs. In the third phase, we construct the final sentiment index with the first principal component of orthogonal ISPs (with their respective lead or lag), whichever has a higher correlation with the first-stage raw sentiment index. In our subsequent analysis we find that the correlation between the raw sentiment index and the final sentiment index is 82 per cent, which suggests we are not running the risk of losing substantial information by dropping the 11 orthogonal ISPs with other time subscripts. Figure 1 shows the co-movements of our newly constructed sentiment index and the value weighted CNX Nifty market index returns.

Figure 1

Source: Authors’ own analysis.

The sentiment index is scaled to have a zero mean and unit standard deviation. The positive or negative impact of such proxies on our aggregate sentiment measure is consistent with the theoretical arguments given in the related literature (Brown and Cliff, 2004, 2005; Baker and Wurgler, 2006, 2007; Kumar and Lee, 2006; Finter et al., 2011; Baker et al., 2011; Dash and Mahakud, 2012).

3.2 Construction of Test Asset Portfolios and Risk Factors

Following Fama and French (1992), Jegadeesh and Titman (1993), Keene and Peterson (2007) the firm size (SZ) is measured by market capitalisation (the product of the stock price and the number of shares outstanding at the end of August in year y); the book-to-market equity (BM) is computed as the ratio between a firm’s book equity at the fiscal year-end (March) in calendar year y and its market value of equity at the end of August in year y; liquidity (LQ) in the month of September y is measured by the annual average of monthly turnover ratio (number of shares traded to the number of shares outstanding) from September y – 1 to August y; and momentum (MM) in the month of September y is measured with respect to the prior performance of stocks from September y – 1 to June y (that is, the 12–2 strategy). The most recent two-month returns are excluded to avoid the continuation effect caused by the bid-ask spread (Lam and Tam, 2011).

For our analysis, for each year we have constructed ten single-shorted, equal-weighted portfolios for the SZ, BM, LQ and MM characteristics. For a more microscopic approach we have constructed single-shorted portfolios for 14 different industry groups (represented as I-n, where n = 1…..14) with respect to the above-mentioned four risk characteristics. Industry classifications and subsequent allocation of companies into different groups have been made with respect to the PROWES database classification. The selected 14 industry groups are: Cement (I-1), Chemical (I-2), Pharmaceuticals (I-3), Construction and real estate (I-4), Food and Beverages (I-5), Machinery Manufacturing (I-6), Textile (I-7), Automobile (I-8), Mining (I-9), Power and Energy (I-10), Retail (I-11), IT Services (I-12), Financial Services (I-13) and Banking (I-14).

For brevity we have reported the results across all industries for two portfolios (P-1 and P-2) in the form of I-n/P-1 and I-n/P-2. Here the portfolios P-1 and P-2 for SZ, BM, LQ and MM characteristics indicate small and large SZ, low and high BM, low and high LQ, loser and winner MM portfolios, respectively. We have opted for 30 per cent (top i.e, P-1), 70 per cent (middle) and 30 per cent (bottom i.e., P-2) as the break points for portfolio construction across industry groups for all the risk characteristics. For single-shorted portfolios across the whole sample and for respective industry groups we have decided to consider equal-weighted approach because the value-weighted approach tends to obscure the relevant effect of respective firm characteristics.

Following previous studies by Fama and French (1993), Carhart (1997), Her et al. (2004), Keene and Peterson (2007) and Lam and Tam (2011), we have constructed five market-wide risk factors, namely: MRKT, that is, market return in excess of a risk-free rate; SMB, the simple average of the returns on three small-stock portfolios (small–low, small–medium, small–high) minus the returns on three big-stock portfolios (big–low, big–medium, big–high); HML, the simple average of returns on the two high-BM portfolios minus the returns on the two low-BM portfolios; UMD, the simple average of returns on the winner-stock portfolios (small winner, big winner) minus returns on the loser-stock (small-loser, big-loser) portfolios; and LMHL, the simple average of the return on the two low liquid portfolios (big–low liquid, small–low liquid) minus the return on two high liquid portfolios (small–high liquid, big–high liquid). The details of five market wide risk factor construction have been provided in the Appendix 1.

Table 1 provides the summary statistics and correlation matrix of the market-wide risk factors. The reported figures in Panel (A) of Table 1 suggest that apart from the market risk premium all the systematic risk factors earn positive premiums. Consistent with the findings of Rouwenhorst (1999), our results suggest that market risk factors in emerging markets are qualitatively similar to those documented for many developed markets. Panel (B) of Table 1 suggests that our sentiment index (Sent) measure shares a negligible correlation structure with the other market-wide risk factors. Consistent with the related literature, the correlation among other market-wide risk factors is also found to be negligible.

Table 1

Descriptive Statistics of Risk Factors and Sentiment Index

Panel (A) Descriptive statistics of risk factors
	Mean	Standard Deviation	Skewness	Kurtosis
MRKT	-8.12	7.48	-0.23	0.33
SMB	0.30	9.97	-1.37	8.00
HML	1.55	11.12	1.80	4.40
UMD	0.48	3.50	0.22	1.54
LMHL	1.09	10.35	-0.24	3.56

Panel (B) Correlation matrix of sentiment index with risk factors
	MRKT	SMB	HML	UMD	LMHL	Sent
MRKT	1.00
SMB	0.15	1.00
HML	-0.08	-0.06	1.00
UMD	-0.26#	0.04	0.01	1.00
LMHL	0.24#	0.01	0.21#	0.05	1.00
Sent	0.02	0.05	-0.13	-0.06	0.02	1.00

Source: Authors.

Note: # represents statistical significance at the 5% level. Sample period ranges from January 2003 to March 2011.

4. DISCUSSION OF THE RESULTS

This section is divided into three main parts. The first part discusses the sentiment pricing evidence across the 40 single-shorted equal-weighted portfolios constructed on the basis of four firm characteristics (SZ, BM, LQ and MM). In the second part we extend our discussion to examine the commonality in sentiment pricing across the selected industry groups. The third part discusses the causality test among the sentiment risk and portfolio returns.

4.1 Estimation Results of Equal-Weighted, Single-Shorted Portfolios

Table 2 presents the multivariate time-series regression results of a sentiment-augmented five-factor model on 20 single-shorted equal-weighted portfolios of SZ and BM characteristics. Panel (A) of Table 2 suggest that the sentiment coefficient has been priced negatively across all the SZ deciles and is significant for the five SZ portfolios. In particular, for SZ, portfolios results in Panel (A) of Table 2 suggest that sentiment fluctuations can lead to significant variation in stock returns for small and large stocks. More specifically, one standard deviation of sentiment index is associated with –168 per year lower equal-weighted returns for small-size portfolios. For medium-size stock portfolios (SZ-5), the lower equal-weighted returns were found to be –0.19 per cent on a monthly basis and –2.28 per cent on a yearly basis. For the large-size stocks (SZ-9), sentiment-driven mispricing resulted in lower equal-weighted returns of –0.20 per cent on a monthly basis and –2.4 per cent on a yearly basis. The reported results suggest that sentiment-driven mispricing is relatively higher for large stocks.

Reported results in Panel (B) of Table 2 indicate that the BM characteristic is also susceptible to sentiment-driven mispricing. The sentiment coefficient has been priced significantly for three BM-shorted portfolios. The sentiment fluctuations can have significant variation in stock returns for both value and growth stocks. Specifically, because of sentiment-driven mispricing, one standard deviation of sentiment index is associated with –2.64, –2.64 and –2.88 percentages per year lower equal-weighted returns for the BM-3, BM-5 and BM-9 portfolios, respectively. The higher returns for the BM-9 portfolio suggest that although the high sentiment period influences lower expected returns across all BM deciles, the impact is higher for growth stocks.

Table 2

Estimation Results of Equation (4) for SZ and BM Portfolio Excess Returns

Panel (A) Time Series Regression Results for Size (SZ) Portfolios
Portfolio Groups	α_t	$β_{S e n t_{p}}$	$β_{M A R T_{p}}$	$β_{S M B_{P}}$	$β_{H M L_{P}}$	$β_{U M D_{P}}$	$β_{L M H L_{P}}$	${\bar{R}}^{2}$
SZ1	-0.26(-0.20)	-0.14* (-1.72)	0.91** (5.08)	1.77*** (12.30)	0.14 (0.55)	-0.32 (-1.63)	-0.50 (-1.57)	0.77
SZ2	5.98** (14.24)	0.01 (-0.12)	0.01 (0.21)	0.36** (4.55)	0.13 (0.96)	-0.02 (-0.23)	-0.10 (-0.75)	0.27
SZ3	-0.10 (-0.09)	-0.16* (-1.74)	0.93** (5.64)	1.50** (9.69)	0.25 (1.05)	-0.26 (-1.34)	-0.36 (-1.10)	0.74
SZ4	5.91** (10.25)	0.03 (0.54)	0.03 (0.51)	0.47** (5.42)	0.14 (1.03)	0.01 (0.15)	-0.11 (-0.64)	0.29
SZ5	-0.06 (-0.05)	-0.19# (-2.00)	0.90** (5.64)	1.37** (8.82)	0.32 (1.43)	-0.25 (-1.23)	-0.48 (-1.42)	0.71
SZ6	6.25** (9.36)	0.03 (0.46)	0.08 (1.08)	0.62** (6.13)	0.23* (1.86)	0.02 (0.26)	-0.04 (-0.21)	0.37
SZ7	0.41(0.36)	-0.17* (-1.75)	0.89** (5.48)	1.23** (8.50)	0.28 (1.27)	-0.17 (-0.92)	-0.49 (-1.55)	0.68
SZ8	5.74** (9.03)	0.02 (0.30)	0.05 (0.77)	0.72** (7.25)	0.32# (2.40)	-0.02 (-0.25)	0.05 (0.22)	0.45
SZ9	0.50(0.38)	-0.20* (-1.86)	0.91** (5.01)	1.06** (6.54)	0.33 (1.50)	-0.13 (-0.65)	-0.44 (-1.33)	0.61
SZ10	5.54** (8.58)	0.03 (0.35)	-0.09 (-1.32)	0.90** (9.48)	0.34# (2.73)	0.01 (0.03)	-0.07 (-0.33)	0.59
Panel (B) Time Series Regression Results for Book-to-market (BM) Portfolios
Portfolio Groups	α_t	$β_{S e n t_{p}}$	$β_{M A R T_{p}}$	$β_{S M B_{P}}$	$β_{H M L_{P}}$	$β_{U M D_{P}}$	$β_{L M H L_{P}}$	${\bar{R}}^{2}$
BM-1	-0.17 (-0.36)	-0.15 (-1.65)	0.85** (6.11)	0.16** (1.75)	0.06 (0.42)	-0.21 (-1.52)	-0.24 (-1.07)	0.52
BM-2	5.74** (12.30)	0.02 (0.38)	0.01 (0.17)	0.08 (1.19)	0.02 (0.17)	0.09 (1.05)	-0.17 (-0.87)	0.01
BM-3	-0.43 (-0.65)	-0.20# (-2.15)	0.76** (4.12)	0.24* (1.88)	0.18 (0.86)	-0.04 (-0.24)	-0.10 (-0.29)	0.35
BM-4	6.65** (11.20)	-0.01 (-0.07)	0.11 (1.53)	0.11 (1.40)	0.23 (1.51)	-0.01 (-0.02)	-0.35 (-0.59)	0.06
BM-5	0.24 (0.17)	-0.22# (-2.37)	0.78*** (4.00)	0.26* (1.77)	0.40* (1.70)	-0.27 (-1.38)	-0.25 (-0.76)	0.33
BM-6	6.53** (7.12)	0.05 (0.94)	0.02 (0.16)	0.21 (1.59)	0.62# (2.96)	0.30* (1.82)	-0.47 (-1.49)	0.13
BM-7	-1.16 (-0.67)	-0.15 (-1.22)	0.76** (3.34)	0.27 (1.59)	0.48 (1.47)	-0.37* (-1.75)	0.25 (0.50)	0.19
BM-8	7.72** (6.10)	0.03 (0.37)	0.19 (1.15)	0.36# (2.12)	0.66# (2..35)	0.02 (0.08)	-0.73 (-1.53)	0.12
BM-9	-1.35 (-0.81)	-0.22* (-1.83)	0.62# (2.79)	0.24 (1.23)	0.68* (1.82)	-0.32 (-1.43)	0.10 (0.17)	0.13
BM-10	7.55** (5.17)	0.02 (0.20)	0.22 (1.20)	0.80** (3.60)	0.83# (2.28)	0.16 (0.59)	-0.84 (-1.64)	0.24

Source: Authors.

Notes: The portfolio symbolised as P-1 (i.e., SZ-1 or BM-1) represents the smallest decile and portfolio P-10 (i.e., SZ-10 or BM-10) represents the largest one. All t-statistics in parentheses are hetroskedasticity-consistent standard errors of White (1980). *, # and ** represent significance at the levels of 0.1, 0.05 and 0.01 per cent, respectively.

Our findings are found to be consistent with Baker and Wurgler (2006, 2007) which suggest that stocks which are hard-to-value and difficult-to-arbitrage are subject to sentiment-driven overvaluation. Significant sentiment coefficients for large stocks may be an indication of large firms which can also be accounted for as high growth in nature. The valuation of small-size and value stocks is considered difficult and prone to subjective valuation partly for their information asymmetry and partly for uncertain future earnings potential. Because of such subjective valuation the arbitrage cost for these stocks is too high to counterbalance the uninformed demand shock for such stocks.

Table 3 presents the multivariate time-series regression results of the sentiment-augmented, five-factor model on 20 single-shorted, equal-weighted portfolios with LQ and MM characteristics. Panel (A) of Table 3 suggests that across the LQ-shorted portfolios, sentiment-influenced mispricing is significant only for the extreme low LQ (LQ-1) and high liquid stocks (LQ-10). Consistent with the theoretical argument, the sentiment coefficient is found to be negative. The effect holds after controlling for market, size, book-to-market, momentum and liquidity risk factors. Reported results in Panel (A) of Table 3 suggest that the illiquid (LQ-1) and extreme liquid (LQ-10) portfolios generate –0.84 and –2.52 percentages, respectively, per year lower equal-weighted returns, following one standard deviation of the sentiment index. The commonality of significant sentiment pricing for low and high liquid portfolios can be justified with two arguments derived from prior literature. First, it has been suggested by Amihud (2002) and Keene and Peterson (2007) that the size or market value of a stock is related to liquidity since a larger stock issue has a smaller price impact for a given order flow and a smaller bid-ask spread. In their respective analyses, the authors suggest that higher returns on small-firm stocks are because of the fact that small firms are more likely to be low liquid and characterised with information asymmetry. The literature on sentiment pricing across small-firm stocks suggests that small firms are hard to value and therefore difficult to arbitrage. Therefore, the pricing evidence of sentiment risk in low liquid stocks can be more generalised with the small firm effect which makes them difficult to arbitrage and thus prone to sentiment fluctuations. Second, Baker and Stein (2004) suggest that the high liquidity characterised with lower bid-ask spreads and lower price impact trade is associated with lower subsequent returns because of persistent overconfident irrational investors’ subjective valuations.

Table 3

Estimation Results of Equation (4) for LQ and MM Portfolio Excess Returns

Panel (A) Time Series Regression Results for Liquidity (LQ) Portfolios
Portfolio Groups	α_t	$β_{S e n t_{p}}$	$β_{M A R T_{p}}$	$β_{S M B_{P}}$	$β_{H M L_{P}}$	$β_{U M D_{P}}$	$β_{L M H L_{P}}$	${\bar{R}}^{2}$
LQ-1	0.39 (0.14)	-0.07* (-1.87)	1.08** (3.22)	0.64** (3.13)	0.20 (0.52)	-0.07 (-0.21)	0.26 (0.49)	0.29
LQ-2	0.16(0.90)	-0.09(-1.04)	0.87** (4.55)	0.58** (3.14)	0.12 (0.49)	0.13 (0.54)	0.30 (0.72)	0.37
LQ-3	-0.33 (-0.12)	-0.06 (0.24)	0.71** (4.12)	0.49** (3.20)	0.40 (1.69)	-0.14 (-0.45)	0.11 (0.35)	0.38
LQ-4	-0.42 (-0.30)	-0.06 (-0.76)	0.83** (4.05)	0.41# (2.54)	0.36 (1.32)	0.18 (0.67)	0.23 (0.63)	0.34
LQ-5	-1.71* (-1.93)	-0.02 (-0.18)	0.73# (2.99)	0.51# (2.95)	0.31 (1.04)	0.08 (0.56)	-0.20 (-0.34)	0.40
LQ-6	-0.24 (-0.26)	-0.06 (-0.61)	0.70** (4.92)	0.35# (2.89)	0.51# (2.77)	-0.12 (-0.43)	-0.14 (-0.27)	0.42
LQ-7	0.47 (0.32)	-0.05 (-0.31)	0.95** (4.66)	0.40 (1.90)	0.02 (0.05)	-0.10 (-0.31)	-0.29 (-0.42)	0.38
LQ-8	-0.47 (-0.40)	-0.07 (-0.61)	0.79 (4.77)	0.25# (2.00)	0.21 (0.62)	-0.21 (-0.72)	-0.34 (-0.57)	0.43
LQ-9	0.65 (0.45)	-0.02 (0.08)	-1.21** (3.29)	0.35 (1.55)	0.07 (0.16)	-0.39 (-1.05)	-0.23 (-0.33)	0.47
LQ-10	1.12 (0.65)	-0.21# (-2.41)	0.16** (5.17)	0.23 (1.28)	0.16 (0.45)	-0.19 (-0.69)	-0.75# (-2.01)	0.46
Panel (B) Time Series Regression Results for Momentum* (MM) Portfolios*
Portfolio Groups	$α_{t}$	$β_{S e n t_{p}}$	$β_{M A R T_{p}}$	$β_{S M B_{P}}$	$β_{H M L_{P}}$	$β_{U M D_{P}}$	$β_{L M H L_{P}}$	${\bar{R}}^{2}$
MM-1	1.59 (1.08)	-0.22* (-1.70)	1.03** (5.15)	0.41# (2.92)	0.41 (1.61)	-0.50# (-2.59)	-0.90 (-0.61)	0.55
MM-2	5.16** (6.63)	-0.04 (-0.59)	0.20# (2.21)	0.18 (1.47)	0.60** (3.30)	0.19* (-1.96)	0.11 (1.49)	0.24
MM-3	1.28 (0.70)	-0.11 (-0.86)	1.17** (4.70)	0.34* (1.88)	0.73# (2.41)	-0.59# (2.31)	-0.72 (-0.90)	0.46
MM-4	6.10** (6.57)	-0.08 (-1.20)	0.28# (2.49)	0.21* (1.87)	0.70** (3.70)	0.32# (-2.59)	0.27 (0.76)	0.29
MM-5	-1.63 (-1.28)	-0.14 (-1.22)	0.94** (5.44)	0.54** (3.08)	0.80# (2.71)	-0.25 (1.59)	-0.57** (-3.01)	0.45
MM-6	6.53** (7.32)	-0.01 (-0.15)	0.22# (2.16)	0.25* (1.91)	0.69** (3.25)	0.38# (-2.12)	0.61 (0.53)	0.27
MM-7	-2.27* (-1.85)	-0.21 (-1.38)	0.91** (5.88)	0.41# (2.09)	0.46 (1.52)	-0.17# (2.94)	-0.86 (-1.49)	0.38
MM-8	7.70** (5.80)	0.08 (0.52)	0.11 (0.69)	0.18 (1.02)	0.68# (2.91)	0.32# (2.94)	0.53 (1.20)	0.13
MM-9	-2.14 (-1.46)	-0.22 (-1.05)	1.03** (5.88)	0.45# (2.01)	0.63* (1.96)	-0.29 (-1.14)	-0.88 (-1.57)	0.38
MM-10	10.74** (5.36)	-0.40# (-2.27)	0.19 (0.69)	0.29 (1.21)	0.77# (2.31)	0.22# (2.70)	0.75 (1.36)	0.49

Source: Authors.

Notes: The portfolio symbolised as P-1 (i.e., LQ-1 or MM-1) represents the smallest decile and portfolio P-10 (i.e., LQ-10 or MM-10) represents the largest decile. All t-statistics in parentheses are hetroskedasticity-consistent standard errors of White (1980). *, # and ** represent significance at the levels of 0.1, 0.05, and 0.01 per cent, respectively.

The results in Panel (B) of Table 3 present sentiment-pricing evidence for the momentum (MM) deciles portfolios. Similar to the liquidity deciles in Panel (A), sentiment-driven mispricing is statistically significant only for the extreme loser and winner portfolios. The commonality of a significant negative sentiment coefficient holds even after controlling for market, size, book-to-market, momentum and liquidity risk factors. The extreme loser (MM-1) and winner (MM-10) portfolios generate –2.64 and –4.80 percentages lower subsequent returns, respectively, with one standard deviation change in the sentiment index. Our results for the sentiment effect in the case of momentum portfolios can be better explained with the bias associated with investor behaviour. Daniel et al. (1998) suggest that the self-attribution bias among investors leads to overconfidence about the performance of ex post winners and, thus, investors may overestimate the prices of winners above fundamental values. In a similar vein, Hong and Stein (1999) attribute momentum profits to the bounded rationality of investors and suggest that momentum profits arise because of investors’ cognitive errors when incorporating information into prices.

In both Tables 2 and 3, out of the ten portfolios only five portfolios across the SZ, BM and MM have significant non-zero intercepts at the 1 per cent level. The results suggest that the five market-wide risk factors provide a dependable explanation for the market-wide systematic risks for describing the cross-sectional variations in stock return. Among the five fundamental risk factors used as control variables in both Table 2 and 3, risk factors like MRKT and SMB have been priced significantly for all the four characteristic-based decile shorted portfolios. The pricing evidence of HML is prominent in the case of large size, high BM and across all MM portfolios. The pricing evidence of the HML factor is negligible in the case of LQ-shorted deciles. The UMD factor is more prominent for the momentum deciles with winner stocks compared to the other characteristic deciles. The pricing evidence of the LMHL factor with a negative coefficient has never been satisfactory except for the large size stocks. In aggregate, the results suggest that four fundamental risk factors (MRKT, SMB, HML and UMD) are prominent for explaining the cross-section of stock return behaviour in the Indian stock market. Our result is consistent with the four factor pricing evidence of Sehgal and Jain (2011) in the Indian stock market, Her et al. (2004) findings for Canadian stock market and Keene and Peterson’s (2007) findings for the US stock market. However, in contrast to the findings of Lam and Tam (2011) in the case of the Korean stock market, we find evidence of significant momentum pricing in the context of the Indian stock market.

Overall, the reported results in Tables 2 and 3 suggest that the commonality of sentiment pricing is also evident in the Indian stock market across the four characteristic-based portfolios. Consistent with prior literature in developed markets, the negative coefficients of sentiment risk in Tables 2 and 3 suggest that, irrespective of the nature of the selected four characteristics, sentiment-driven mispricing is accompanied with lower subsequent returns. The proposition here is that, since positive sentiment results in over-valuation of stocks, the expected return for such stocks will be lower. This finding supports the prediction of the noise trader model of De Long et al. (1990) and the finding of Brown and Cliff (2004, 2005) that stock prices could be influenced by investor sentiment apart from the fundamental risk factors.

4.2 Estimation Results of Equal-Weighted, Single-Shorted Industry Portfolios

Table 4 reports the results of sentiment pricing evidence across the 14 selected industry groups. In our reported results of Table 4 we only concentrate on the coefficients of intercept and sentiment risk for the purpose of brevity. Consistent with the theoretical argument, the sentiment effect results in a negative coefficient across all industry groups and with all the portfolio-shorting criteria. This indicates that irrespective of the different characteristic-based portfolios, the good sentiment period in the previous month results in a lower return in the subsequent month. The size-shorted portfolios results suggest that industries such as I-1, I-7, I-9, I-10, I-12 and I-13 display significant sentiment-pricing evidence. In particular, sentiment risk is priced significantly for large-size stocks in the case of the I-1, I-7, I-10, I-12 and I-13 industry groups and for I-9 and I-12, sentiment risk pricing is evident in the small-size portfolios. The large-size stock sentiment pricing is however inconsistent with the general acceptability that only the small-size stocks are prone to sentiment mispricing. Consistent with our results from the market-wide analysis (Tables 2 and 3) we find evidence that, apart from the small-size stocks, the large-size stocks are also prone to sentiment-risk pricing in selected industry groups. If we allow the argument that, given a generic sentiment mispricing across the market and a variability of arbitrage constraints across industries, the results suggest that the arbitrage constraint across industries with a limited number of stocks also allow investors to have correlated sentiment-driven demand shocks for selected large-size stocks. To be more specific, the results in Table 4 indicate that a one standard deviation increase in the total investor sentiment index is associated with a 1.08, 0.72, 0.72, 0.82 and 0.72 per cent per year lower equal weighted-returns for large-size portfolios in the case of I-1, I-7, I-10, I-12 and I-13 industry groups, respectively. In the case of equal weighted small-size portfolios of the I-9 and I-12 industry groups, a one standard deviation increase in the total investor sentiment index is associated with 0.96 and 0.72 per cent lower returns, respectively.

In the case of the BM-shorted portfolios, sentiment pricing is evident in the I-1, I-7, I-9, I-12, I-13 and I-14 industry groups. Sentiment risk is priced significantly for the growth stocks (low BM) in the case of the I-1, I-7, I-9, I-13 and I-14 industry groups and for the I-7, I-9, I-12 and I-14 the sentiment risk pricing is evident in value stock (high BM) portfolios. Across the growth stock portfolios, our results suggest that a one standard deviation increase in the total investor sentiment index is associated with 0.96, 0.72, 1.08, 0.72 and 0.96 per cent per year lower equal weighted returns for the I-1, I-7, I-9, I-13 and I-14 industry groups, respectively. Similarly for the I-7, I-9, I-12 and I-14 industry groups, a one standard deviation increase in the total investor sentiment index is associated with 0.60, 0.84, 0.72 and 0.96 percentages per year lower equal weighted value stock returns, respectively. This sentiment pricing evidence is consistent with the prior related literature argument that value stocks and extreme growth stocks are prone to sentiment mispricing (Baker and Wurgler, 2006, 2007; Changsheng and Yongfeng, 2012). For the liquidity shorted portfolios across industry groups, our results suggest that industries such as I-1, I-4, I-7, I-8, I-9, I-10, I-12, I-13 and I-14 are more prone to sentiment risk pricing across different liquidity characteristic portfolios.

High liquid stocks are more prone to sentiment pricing for the I-1, I-4, I-7, I-8, I-9, I-10, I-12, I-13 and I-14 industry groups and the low liquid portfolios for the I-9, I-10, I-12 and I-14 groups. In particular, a one standard deviation increase in the total investor sentiment index is associated with 1.08, 1.32, 0.72, 0.72, 1.08, 0.72, 0.96, 0.84 and 0.96 per cent per year lower equal weighted stock returns for the I-1, I-4, I-7, I-8, I-9, I-10, I-12, I-13 and I-14 industry groups, respectively. For the low liquid portfolios in case of I-9, I-10, I- 12 and I-14, one standard deviation increase in total investor sentiment index is associated with 0.60, 0.72, 0.72 and 0.72 percent per year lower equal weighted stock returns, respectively. Our results for sentiment pricing in the low and high-liquid firms are consistent with results in Tables 2 and 3.

Furthermore, in the case of momentum-shorted portfolios, industry groups such as I-1, I-7, I-9, I-10, I-12, I-13 and I-14 have shown significant return variation with respect to sentiment fluctuations. While sentiment risk is priced significantly across winner stock portfolios in the case of the I-7, I-9, I-10, I-12 and I-14 industry groups, loser stock portfolios are priced significantly for the I-1, I-7, I-10, I-12 and I-14 industry groups. To be more specific, the results in Table 4 suggest that on a yearly basis for winner stock portfolios in the I-7, I-9, I-10, I-12 and I-14 groups, a one standard deviation increase in the total investor sentiment index is associated with 0.72, 1.44, 0.72, 0.72 and 0.84 per cent lower equal weighted stock returns, respectively. Similarly on a yearly basis, one standard deviation increase in the total investor sentiment index is associated with 1.08, 0.60, 0.84, 0.72 and 0.84 per cent lower equal weighted stock returns for the I-1, I-7, I-10, I-12 and I-14 industry groups, respectively. Across all the characteristic-based shorting criteria and across all the industry groups we find that all the intercepts are negative and are significantly different from zero at the 1 per cent level.

Table 4

Regression Results for Single-Shorted Industry Portfolios

Industry Groups	Portfolio Shorts	SZ			BM			LQ			MM
Industry Groups	Portfolio Shorts	α_t	$β_{S e n t_{p}}$	${\bar{R}}^{2}$	α_t	$β_{S e n t_{p}}$	${\bar{R}}^{2}$	α_t	$β_{S e n t_{p}}$	${\bar{R}}^{2}$	α_t	$β_{S e n t_{p}}$	${\bar{R}}^{2}$
I-1[18]	P-1	-5.24** (-15.99)	-0.07 (-1.17)	0.12	-4.99** (-16.85)	-0.08** (-1.75)	0.14	-5.13** (-17.17)	-0.07 (-1.23)	0.10	-5.07** (-15.39)	-0.09* (-1.79)	0.14
	P-10	-4.82** (-15.62)	-0.09** (-1.90)	0.17	-4.89** (-14.42)	-0.08 (-1.41)	0.16	-4.90** (-14.60)	-0.09* (-1.94)	0.19	-4.84** (-16.10)	-0.06 (-1.20)	0.11
I-2 [110]	P-1	-5.32** (-21.34)	-0.01 (-0.53)	0.13	-4.85** (-22.40)	-0.05 (-1.35)	0.14	-5.25** (-20.67)	-0.02 (-0.66)	0.11	-5.14** (-21.61)	-0.03 (-1.04)	0.14
	P-10	-4.91** (-21.84)	-0.05 (-1.51)	0.13	-5.17** (-19.68)	-0.03 (-1.15)	0.10	-5.02** (-20.21)	-0.05 (-1.59)	0.12	-5.03** (-20.62)	-0.04 (-1.24)	0.09
I-3 [53]	P-1	-5.10** (-16.46)	-0.03 (-0.65)	0.09	-4.89** (-24.23)	-0.05 (-1.37)	0.16	-5.00** (17.07)	-0.03 (-0.78)	0.10	-5.01** (-18.68)	-0.05 (-1.17)	0.08
	P-10	-4.84** (-21.83)	-0.05 (-1.29)	0.13	-4.92** (-15.61)	-0.03 (-0.75)	0.10	-4.97** (-21.41)	-0.05 (-1.41)	0.11	-5.00** (-22.91)	-0.05 (-1.24)	0.18
I-4[56]	P-1	-5.22** (-21.11)	-0.04 (-1.11)	0.16	-4.79** (-18.92)	-0.04 (-1.31)	0.11	-5.18** (-15.90)	-0.04 (-1.11)	0.10	-5.10** (-15.93)	-0.06 (-1.22)	0.12
	P-10	-4.72** (-17.09)	-0.04 (-1.25)	0.09	-5.21** (-19.43)	-0.05 (-1.33)	0.15	-4.72** (-13.30)	-0.11# (-2.15)	0.15	-5.04** (-15.25)	-0.06 (-1.37)	0.11
I-5 [57]	P-1	-5.08** (-17.00)	-0.05 (-1.10)	0.14	-4.92** (-23.02)	-0.04 (-1.18)	0.04	-5.06** (-18.09)	-0.03 (-0.80)	0.18	-4.87** (-17.64)	-0.04 (-1.19)	0.14
	P-10	-4.77** (-21.46)	-0.03 (-1.06)	0.03	-4.99** (-19.92)	-0.05 (-1.29)	0.18	-4.84** (-18.91)	-0.04 (-1.65)	0.05	-5.06** (-19.73)	-0.06 (-1.56)	0.12
I-6[90]	P-1	-5.23** (-21.36)	-0.04 (-1.11)	0.16	-4.79** (-18.82)	-0.04 (-1.30)	0.10	-5.03** (-20.00)	-0.03 (-0.75)	0.13	-5.03** (-18.01)	-0.04 (-1.40)	0.15
	P-10	-4.71** (-17.11)	-0.04 (-1.25)	0.09	-5.19** (-19.65)	-0.05 (-1.36)	0.16	-4.93** (-17.57)	-0.06 (-1.61)	0.11	-5.01** (-19.41)	-0.02 (-0.63)	0.06
I-7[71]	P-1	-5.28** (-17.62)	-0.04 (-1.26)	0.09	-5.01** (-20.33)	-0.06** (-1.80)	0.14	-5.28** (-19.23)	-0.04 (-0.93)	0.08	-5.09** (-19.08)	-0.05** (-1.76)	0.12
	P-10	-5.03** (-21.16)	-0.06** (-1.94)	0.12	-5.21** (-18.33)	-0.05** (-1.84)	0.12	-4.97** (-21.87)	-0.06# (-2.60)	0.17	-5.26** (-20.53)	-0.06** (-1.85)	0.09
I-8 [50]	P-1	-5.01** (-18.07)	-0.03 (-0.75)	0.15	-4.81** (-18.22)	-0.04 (-1.22)	0.06	-5.02** (-17.42)	-0.02 (-0.39)	0.08	-4.93** (-19.38)	-0.05 (-1.39)	0.11
	P-10	-4.61** (-18.48)	-0.05 (-1.26)	0.04	-4.93** (-17.31)	-0.05 (-1.39)	0.10	-4.69** (-18.25)	-0.06** (-1.91)	0.06	-4.86** (-18.65)	-0.05 (-1.06)	0.13
I-9[12]	P-1	-4.55** (-12.06)	-0.08# (-2.39)	0.10	-4.96** (-15.10)	-0.09** (-1.87)	0.15	-4.53** (-15.38)	-0.05# (-2.08)	0.13	-4.45** (-15.42)	-0.05 (-1.41)	0.10
	P-10	-4.68** (-15.46)	-0.08 (-1.46)	0.07	-4.60** (-14.15)	-0.07# (-2.41)	0.06	-4.88** (-14.43)	-0.09** (-1.80)	0.06	-5.02** (-14.43)	-0.12** (-3.75)	0.16
I-10 [08]	P-1	-5.39** (-16.54)	-0.06 (-1.46)	0.09	-5.10** (-19.45)	-0.04 (-1.27)	0.02	-5.40** (-15.88)	-0.06** (-1.78)	0.11	-5.14** (-18.88)	-0.05** (-1.75)	0.01
	P-10	-4.94** (-18.88)	-0.06# (-2.46)	0.08	-5.30** (-15.24)	-0.05 (-1.33)	0.05	-4.95** (-19.19)	-0.06# (-2.39)	0.09	-5.18** (-16.67)	-0.06** (-1.72)	0.05
I-11 [40]	P-1	-5.59** (-17.40)	-0.04 (-1.17)	0.10	-4.92** (-17.93)	-0.05 (-1.60)	0.15	-5.40** (-17.80)	-0.05 (-1.53)	0.07	-5.54** (-20.86)	-0.05 (-1.49)	0.16
	P-10	-4.81** (-20.60)	-0.04 (-1.61)	0.11	-5.15** (-17.31)	-0.04 (-1.17)	0.06	-5.03** (-17.44)	-0.05 (-1.62)	0.14	-5.22** (-19.14)	-0.06 (-1.63)	0.09
I-12 [71]	P-1	-5.26** (-18.83)	-0.06** (-1.65)	0.21	-4.91** (-17.41)	-0.06 (-1.40)	0.08	-5.13** (-19.40)	-0.06** (-1.70)	0.21	-5.21** (-18.80)	-0.07** (-1.71)	0.14
	P-10	-4.93** (-17.33)	-0.07** (-1.65)	0.06	-5.15** (-18.77)	-0.06* (-1.71)	0.17	-4.97** (-16.79)	-0.08* (-1.86)	0.10	-5.03** (-20.16)	-0.06* (-1.69)	0.15
I-13 [63]	P-1	-5.22** (-18.86)	-0.04 (-1.12)	0.19	-4.69** (-19.06)	-0.06* (-1.92)	0.15	-5.16** (-16.62)	-0.02 (-0.69)	0.18	-5.16** (-19.33)	-0.06# (-2.27)	0.24
	P-10	-4.78** (-18.30)	-0.06# (-1.82)	0.11	-5.01** (-18.34)	-0.05 (-1.56)	0.17	-4.80** (-21.38)	-0.07# (-2.11)	0.12	-4.83** (-20.97)	-0.04 (-1.16)	0.19
I-14 [36]	P-1	-4.87** (-16.68)	-0.03 (-0.74)	0.08	-4.74** (-15.50)	-0.08# (-2.66)	0.09	-4.64** (-17.48)	-0.06* (-1.81)	0.10	-4.84** (-17.77)	-0.07** (-3.24)	0.12
	P-10	-4.71** (-15.77)	-0.04 (-0.78)	0.05	-4.87** (-20.41)	-0.08** (-3.57)	0.13	-4.77** (18.25)	-0.08** (-3.51)	0.13	-4.61** (-17.67)	-0.07** (-2.97)	0.11

Source: Authors.

Notes: The 14 industry groups are: Cement (I-1), Chemical (I-2), Pharmaceutical (I-3), Construction and real estate (I-4), Food and Beverages (I-5), Machinery Manufacturing (I-6), Textile (I-7), Automobile (I-8), Mining (I-9), Power and Energy (I-10), Retail (I-11), IT Services (I-12), Financial Services (I-13) and Banking (I-14).

We have opted for the top 30 per cent (P-1, i.e., small) and bottom 30 per cent (P-10, i.e., large) as the break points for portfolio construction for all four characteristics. In the industry group portfolio column, figures in the square brackets [ ] represent the annual average number of companies available under each industry groups. All t-statistics in parentheses are hetroskedasticity-consistent standard errors of White (1980). *, # and ** represent significance at the level of 0.1, 0.05, and 0.01 per cent, respectively.

Table 4 suggests that the sentiment risk pricing for different characteristic-based portfolios across several industry groups conforms to the theoretical argument. By considering all the sentiment pricing evidence it can be generalised that for Indian stock market, industries such as Cement (I-1), Textiles (I-7), Mining (I-9), Power and Energy (I-10), IT Services (I-12), Financial Services (I-13) and Banking (I-14) are more prone to sentiment mispricing than other industries. To be more specific, in respective equal weighted portfolio analysis, large SZ-low BM-high LQ-loser stocks for I-1, large SZ-low and high BM-high LQ-loser and winner stocks for I-7, large SZ-low and high BM-low and high LQ-winner stocks for I-9, large SZ-low and high LQ-loser and winner stocks for I-10, small and large SZ-high BM-low and high LQ-loser and winner stocks for I-12, large SZ-low BM-high LQ-loser stocks for I-13, low and high BM-low and high LQ-loser and winner stocks for I-14 are more sensitive for a broad wave of market-wide investor sentiment fluctuations.

Table 5

Estimation Results of Granger-causality Test for Decile-Shorted Portfolios

SZ			BM			LQ			MM
Decile Portfolio	Sent. R_p	R _p Sent	Decile Portfolios	Sent. R_p	R _p Sent	Decile Portfolios	Sent. R_p	R _p Sent	Decile Portfolios	Sent. R_p	R _p Sent
SZ-1	0.87{0.35}	0.31{0.58}	BM-1	2.27 {0.10}	0.47{0.49}	LQ-1	0.01{0.95 }	2.58 {0.01}	MM-1	2.95 {0.08}	1.00{0.31}
SZ-2	1.01{0.31}	0.82{0.36}	BM-2	1.28{0.25}	0.13{0.71}	LQ-2	0.06{0.80}	1.08{0.30}	MM-2	5.20 {0.02}	0.33{0.56}
SZ-3	1.41{0.23}	0.51{0.47}	BM-3	1.91 {0.10}	0.92{0.33}	LQ-3	0.17{0.67}	0.73{0.39}	MM-3	1.68{0.19}	0.94{0.33}
SZ-4	0.97{0.32}	0.81{0.36}	BM-4	2.41 {0.08}	0.64{0.42}	LQ-4	0.05{0.81}	0.31{0.57}	MM-4	2.25 {0.10}	1.48{0.22}
SZ-5	1.59{0.20}	0.77{0.38}	BM-5	2.56 {0.10 }	0.77{0.38}	LQ-5	0.27{0.60}	0.20{0.65}	MM-5	3.50 {0.06}	0.09{0.76}
SZ-6	1.45{0.23}	0.97{0.32}	BM-6	1.12{0.29}	0.73{0.39}	LQ-6	0.14{0.70}	0.42{0.51}	MM-6	1.57{0.21}	0.35{0.55}
SZ-7	1.13{0.28}	1.25{0.26}	BM-7	0.47{0.49}	0.68{0.40}	LQ-7	0.06{0.80 }	2.76 {0.09}	MM-7	2.35{0.10}	0.01{0.99}
SZ-8	1.35{0.24}	1.17{0.28}	BM-8	0.81{0.37}	0.89{0.34}	LQ-8	0.02{0.88}	0.85{0.35}	MM-8	0.86{0.36}	0.87{0.35}
SZ-9	1.97 {0.10}	1.43{0.23}	BM-9	2.11 {0.10}	0.17{0.67}	LQ-9	0.17{0.67}	2.43{0.12}	MM-9	0.92{0.33}	1.56{0.21}
SZ-10	2.06 {0.10}	0.45{0.50}	BM-10	1.37{0.24 }	0.86{0.35}	LQ-10	0.01{0.99 }	1.93 {0.10}	MM-10	0.35{0.55}	11.48 {0.00}

Source: Authors.

Notes: The portfolios SZ-1, BM-1, LQ-1 and MM-1 represent the smallest portfolio (P-1); portfolios SZ-10, BM-10, LQ-10, MM-10 represent the largest portfolio (P-10). Figures in the curly brackets represent the corresponding P-values of the Granger causality (1969, 1988) test statistics. “ ” implies the null hypothesis ‘does not Granger-cause’. R_P indicates raw portfolio returns. Sent. indicates the sentiment index. The lag period has been taken as one month. The sample period includes 98 monthly observations from February 2003 to March 2011.

Overall, the results reported in Tables 2, 3 and 4 suggest that sentiment pricing is not only significant for small size, value and high growth stocks, but also for large size stocks. In particular, Baker and Stein (2004) argue that in the presence of short-sell constraints, high liquidity is a symptom of over-valuation by irrational investors, which in turn make such stocks hard to value and difficult to arbitrage. Given such arguments, in our reported results, the high liquid stocks sensitivity to sentiment risk seems to be followed from the initial mispricing caused by irrational overconfident investors.

4.3 Causality Test Between Investor Sentiment and Portfolio Returns

The reported results in Table 5 suggest that across the decile portfolios of size characteristics, there is a unidirectional causality from investor sentiment to portfolio returns. The null hypothesis that suggests that sentiment does not Granger-cause variations in the portfolio returns of BM-1, BM-3, BM-4, BM-5 and BM-9 was found to be rejected. This supports the sentiment risk pricing results reported in Table 2 for the BM-3, BM-5 and BM-9 portfolios, however, the sentiment risk pricing evidence of such portfolio groups results in the fact that sentiment Granger-causes subjective valuation of such stocks. The liquidity decile-shorted portfolio causality test results were found to be different for the SZ and BM-shorted portfolios.

The reported results suggest that there exists a unidirectional causal relationship between portfolio returns and sentiment risk. More specifically, the results suggest that such a unidirectional causal relationship exists with respect to the LQ-1, LQ-7 and LQ-10 portfolios. The sentiment pricing evidence for the LQ-1 and LQ-10 portfolios and the unidirectional causal relationship from portfolio returns to sentiment risk for such portfolios could indicate the common assumption of liquidity as a proxy for measuring sentiment. Observing this argument in greater detail, one may argue that as stock liquidity is commonly associated with a market-wide positive sentiment for such stocks (Baker and Stein, 2004; Baker and Wurgler, 2006, 2007), the sentiment pricing evidence for such stocks may be arising because of the unidirectional causal link between stock returns and aggregate investor sentiment. As in the SZ and BM decile portfolios, in the momentum decile portfolios we found a unidirectional causal relationship between investor sentiment and portfolio returns for the ‘loser’ stock portfolios. However, the unidirectional relationship from portfolio returns to aggregate market-wide investor sentiment exists for the winner stock portfolios (MM-10). Overall, the reported results in Table 5 indicate that the causal relationship between investor sentiment and portfolio returns never appears to be bidirectional, but exists only in the unidirectional form and that the nature of the unidirectional causality depends on the portfolio characteristics.

Table 6 reports the Granger causality test between investor sentiment and portfolio returns of several industry groups. In contrast to the presence of a unidirectional causal relationship from portfolio returns to investor sentiment in certain portfolio categories in Table 5, the reported results in Table 6 suggest that although a causal relationship exists, it is always from investor sentiment to portfolio returns. Overall, the results in Table 6 indicate that the null hypothesis of sentiment does not Granger-cause portfolio returns fails to be rejected across all the test asset portfolios constructed for the 14 industry groups. The causal relationship from the aggregate market-wide investor sentiment to portfolio returns exists for the I-1, I-7, I-9, I-10, I-12 and I-13 industry groups for the SZ characteristics. The results for these industry groups are consistent with the sentiment risk pricing evidence of Table 4. Similarly, for the BM characteristics, the unidirectional causal relationship is consistent with the sentiment pricing evidence of the I-1, I-7, I-9, I-12, I-13 and I-14 industry groups. In contrast with the LQ decile-shorted portfolio results, for the industry-shorted portfolios we do not find any evidence of a unidirectional causal relationship from LQ portfolio returns to aggregate market-wide investor sentiment.

Table 6

Estimation Results of Granger-causality Test for Industry Portfolios

Industry Groups	Portfolio Shorts	SZ		BM		LQ		MM
Industry Groups	Portfolio Shorts	Sent. R_p	R _p Sent.	Sent. R_p	R _p Sent.	Sent. R_p	R _p Sent.	Sent. R_p	R _p Sent.
I-1[18]	P-1	3.45 {0.06}	0.98 {0.32}	0.84 {0.36}	0.59 {0.44}	4.07 {0.04}	1.28 {0.25}	7.16 {0.00}	0.61 {0.43}
	P-10	7.31 {0.00}	0.06 {0.80}	7.78 {0.00}	0.32 {0.57}	6.65 {0.01}	0.10 {0.74}	3.27 {0.07}	0.35 {0.54}
I-2 [110]	P-1	0.60 {0.44}	0.26 {0.61}	0.84 {0.36}	0.59 {0.44}	12.15 {0.00}	0.01 {0.90}	2.51 {0.10}	0.01 {0.95}
	P-10	11.24 {0.00}	0.18 {0.67}	7.78 {0.00}	0.32 {0.57}	2.01 {0.15}	0.01 {0.93}	5.07 {0.02}	0.03 {0.85}
I-3[53]	P-1	0.54 {0.46}	0.02 {0.86}	11.79 {0.00}	0.04 {0.83}	0.89 {0.34}	0.20 {0.65}	3.02 {0.08}	0.15 {0.69}
	P-10	7.85 {0.00}	0.17 {0.68}	0.73 {0.39}	0.01 {0.98}	5.86 {0.01}	0.47 {0.49}	3.86 {0.05}	0.90 {0.34}
I-4[56]	P-1	2.75 {0.10}	0.22 {0.63}	3.81 {0.05}	0.11 {0.73}	2.24 {0.10}	1.10 {0.27}	3.99 {0.44}	0.06 {0.79}
	P-10	4.10 {0.04}	0.20 {0.65}	4.43 {0.03}	0.01 {0.95}	12.88 {0.00}	0.55 {0.45}	5.50 {0.02}	0.02 {0.87}
I-5 [57]	P-1	2.14 {0.14}	0.07 {0.78}	3.75 {0.05}	0.49 {0.48}	1.40 {0.23}	2.33 {0.12}	2.23 {0.13}	0.98 {0.32}
	P-10	2.27 {0.09}	0.74 {0.38}	2.71 {0.10}	0.03 {0.84}	0.04 {0.83}	0.01 {0.98}	4.96 {0.02}	0.57 {0.44}
I-6[90]	P-1	2.62 {0.10}	0.17 {0.67}	3.70 {0.05}	0.15 {0.69}	1.70 {0.19}	0.56 {0.45}	2.21 {0.10}	1.25 {0.26}
	P-10	4.20 {0.04}	0.21 {0.64}	4.38 {0.03}	0.01 {0.94}	6.07 {0.01}	0.18 {0.67}	0.88 {0.34}	0.46 {0.49}
I-7[71]	P-1	1.99 {0.16}	0.01 {0.92}	7.54 {0.00}	0.54 {0.51}	3.22 {0.07}	0.01 {0.90}	8.89 {0.00}	0.42 {0.51}
	P-10	7.54 {0.00}	0.01 {0.92}	6.73 {0.01}	0.27 {0.60}	2.54 {0.10}	0.01 {0.93}	5.07 {0.02}	0.09 {0.75}
I-8 [50]	P-1	2.19 {0.14}	0.01 {0.96}	4.68 {0.03}	0.15 {0.69}	1.10 {0.29}	1.46 {0.22}	4.73 {0.03}	0.01 {0.90}
	P-10	6.88 {0.01}	0.83 {0.36}	3.59 {0.06}	0.01 {0.96}	8.14 {0.00}	1.41 {0.23}	5.86 {0.01}	0.01 {0.98}
I-9[12]	P-1	4.78 {0.03}	1.82 {0.18}	11.53 {0.01}	1.25 {0.26}	1.99 {0.16}	0.21 {0.64}	2.64 {0.10}	0.01 {0.92}
	P-10	7.98 {0.01}	0.27 {0.60}	3.45 {0.06}	1.21 {0.27}	8.35 {0.00}	0.80 {0.37}	17.69 {0.00}	1.12 {0.29}
I-10[08]	P-1	3.54 {0.06}	0.24 {0.62}	0.39 {0.87}	0.61 {0.84}	4.55 {0.03}	0.23 {0.63}	3.69 {0.05}	2.43 {0.10}
	P-10	6.75 {0.01}	0.78 {0.38}	2.15 {0.10}	0.71 {0.40}	5.67 {0.01}	0.95 {0.33}	3.01 {0.08}	0.52 {0.46}
I-11[40]	P-1	1.75 {0.18}	0.02 {0.87}	2.14 {0.14}	0.87 {0.35}	2.61 {0.10}	0.51 {0.47}	3.69 {0.05}	0.13 {0.71}
	P-10	6.30 {0.01}	0.01 {0.90}	2.89 {0.57}	0.09 {0.44}	6.78 {0.01}	0.04 {0.83}	5.94 {0.01}	0.17 {0.67}
I-12[71]	P-1	3.96 {0.04}	0.01 {0.97}	5.52 {0.02}	0.01 {0.89}	4.68 {0.03}	0.07 {0.78}	6.33 {0.01}	0.25 {0.61}
	P-10	8.06 {0.01}	0.37 {0.54}	4.28 {0.04}	0.11 {0.73}	9.62 {0.01}	0.55 {0.45}	6.36 {0.01}	0.92 {0.33}
I-13[63]	P-1	2.80 {0.09}	0.44 {0.50}	7.39 {0.00}	0.02 {0.87}	1.51 {0.22}	0.39 {0.53}	7.76 {0.00}	0.45 {0.50}
	P-10	8.97 {0.00}	0.75 {0.38}	3.85 {0.05}	0.05 {0.81}	8.53 {0.00}	0.07 {0.93}	4.12 {0.04}	0.01 {0.97}
I-14[36]	P-1	3.11 {0.08}	0.01 {0.97}	13.64 {0.00}	0.91 {0.34}	9.21 {0.01}	0.54 {0.46}	9.84 {0.01}	1.72 {0.19}
	P-10	3.41 {0.06}	1.43 {0.23}	10.76 {0.00}	0.68 {0.40}	11.71 {0.01}	0.86 {0.35}	10..77 {0.01}	1.61 {0.20}

Source: Authors.

Notes: The 14 industry groups considered are similar to those in Table 4. We have opted for the top 30% (P-1, i.e., small) and bottom 30% (P – 10, i.e., large) as the breakpoints for portfolio construction for the four characteristics. In the industry group portfolio column, figures in the square brackets [ ] represent the annual average number of companies available under each industry group. Figures in the curly brackets represent the corresponding P-values of the Granger causality (1969, 1988) test statistics. “” implies that the null hypothesis ‘does not Granger-cause’. R_P indicates raw portfolio returns. Sent. indicates the sentiment index. The lag period has been taken as one month. The sample period includes 98 monthly observations from February 2003 to March 2011.

For the LQ characteristics across all industry groups, the unidirectional causal relationship between investor sentiment and portfolio returns and the sentiment risk pricing evidence of Table 4 are found to be consistent for the I-1, I-4, I-7, I-9, I-10, I-12 and I-13 portfolios. However, although there is supportive pricing evidence of sentiment risk evident for the I-14 group (Table 4), in Table 6 we do not observe any causal link for the I-14 industry group. For the MM characteristics across all the industry-shorted portfolios, we find a unidirectional causal relationship from investor sentiment to portfolio returns for both the loser (P-1) and winner (P-10) stocks. The sentiment-pricing evidence (Table 4) and unidirectional causal relationship between sentiment and portfolio returns found common ground for the I-1, I-7, I-9, I-10, I-12, I-13 and I-14 industry groups. Considering the maximum appearance of the unidirectional aggregate investor sentiment and portfolio return evidence, our results appear to be inconsistent with the findings of Schmeling (2009), which suggest a bidirectional causal relationship between investor sentiment and market index returns for most industrial countries.

5. CONCLUSIONS AND IMPLICATIONS

This article attempts to analyse the implications of the sentiment risk effect on cross-sectional return variations across industry groups, after controlling all five major market-wide risk factors. After controlling the effects of market-wide risk factors we find that sentiment fluctuations by one standard deviation can induce a negative effect on subsequent stock returns. With single-shorted portfolios, after controlling the effects of market-wide risk factors (market risk premium, size, value, momentum and liquidity), we find that stocks with small size, high and low book-to-market equity, low and high liquidity, extreme loser and winner portfolios are more sensitive to fluctuations in sentiment that are unwarranted by changes in fundamentals. Such characteristics can be more profoundly considered to be the replication of hard-to-value and difficult-to-accurately-price in the Indian stock market. Since such categories of stocks are prone to misevaluations and are subsequently associated with high arbitrageur costs, rational arbitrageurs may fail to counterbalance the uninformed demand shocks for such stocks. Our findings are consistent with the existing related literature (Baker and Wurgler, 2006, 2007; Changsheng and Yongfeng, 2012; Dash and Mahakud, 2012; Finter et al., 2011; Schmeling, 2009).

In our subsequent analysis for several industry groups we find similar sentiment sensitivity of four selected risk characteristics. In aggregate our results suggest that, among the 14 industry groups, seven groups are more prominent for sentiment sensitivity. The existing literature assumes that the commonality of the sentiment effect across industry is similar to the aggregate market and that the sources of sentiment impact variations across industry have never been a source of empirical validation. In our results, although the impact of sentiment risk on stock returns for industry-shorted portfolios is consistent with the theoretical argument, cross-sectional variations with respect to different industries have been heavily dependent on the availability of stocks in that particular industry. The results suggest that the generalisation of hard-to-value and difficult-to-arbitrage argument must be judged with caution, keeping the industry effects in mind. The causality test results suggest a unidirectional causal relationship between aggregate market-wide investor sentiment and characteristics-based test asset portfolio returns for almost all the portfolio returns, with minor exceptions only among the decile-shorted portfolios. The causality test results also reveal that, although at certain points of time it appears that there exists a unidirectional causal relationship between aggregate market-wide investor sentiment and portfolio returns, it does not draw a significant sentiment risk pricing evidence for such category portfolio returns.

Our results can be more useful from the practitioners’ perspective for tactical asset allocation, by following an active investment strategy. Such sentiment insensitive industry groups may be used for contrarian investment strategy exploration when sentiment is high in the market. Fund managers who want to hedge against highly volatile market behaviour can use the non-sentiment prone industry groups to protect their fund valuation.

Footnotes

Appendix

Construction of Market-wide Systematic Risk Factors
MRKT	[Market Return - Risk Free Rate of Return]
SMB	[(Small Low) - (Big Low)] + [(Small Medium) - (Big Medium)] + [(Small High) - (Big High)]/3
HML	[(Small High) - (Small Low)] + [(Big High) -(Big Low)]/2
UMD	[(Small Winner) - (Small Loser)] + [(Big Winner) - (Big Loser)]/2
LMHL	[(Small Low liquid) - (Small High liquid)] + [(Big Low liquid) - (Big High liquid)]/2

References

Amihud

(2002). Illiquidity and stock returns: Cross-section and time-series effects. Journal of Financial Markets, 5(1), 31–56.

Baker

Wurgler

(2006), Investor sentiment and the cross-section of stock returns. Journal of Finance, 61(4), 1645–80.

Baker

Wurgler

(2007). Investor sentiment in the stock market. Journal of Economic Perspective, 21(2), 129–51.

Baker

Wurgler

Yuan

(2011). Global, local, and contagious investor sentiment. Journal of Financial Economics, 104(2), 272–87.

Baker

Stein

J.C.

(2004). Market liquidity as a sentiment indicator. Journal of Financial Markets, 7(3), 271–99.

Baker

H.K.

Nofsinger

J.R.

(2002). Psychological biases of investors. Financial Services Review, 11(2), 97–116.

Barberis

Shleifer

Vishny

(1998). A model of investor sentiment. Journal of Financial Economics, 49(3), 307–43.

Barberis

N.R.

Thaler

R.H.

(2003). A survey of behavioral finance (Chapter-18). In Constantinides

G.M.

Harris

Stulz

R. M.

(Eds), Handbook of the Economics of Finance, Vol. 1, (pp. 1053–128), North Holland, Netherlands: Elsevier

Black

(1986). Noise. Journal of Finance, 41(3), 529–43.

10.

Brown

G.W.

Cliff

M.T.

(2004). Investor sentiment and the near-term stock market, Journal of Empirical Finance, 11(1), 1–27.

11.

Brown

G.W.

Cliff

M.T.

(2005). Investor sentiment and asset valuation. Journal of Business, 78(2), 405–40.

12.

Carhart

M.M.

(1997). On persistence in mutual fund performance. Journal of Finance, 52(1), 57–82.

13.

Changsheng

Yongfeng

(2012). Investor sentiment and asset valuation. Systems Engineering Procedia, 3, 166–71.

14.

Chui

C.W.A.

Wei

K.C.J

(1998). Book-to-market, firm size, and the turn of the year effect: Evidence from Pacific-Basin emerging markets. Pacific-Basin Finance Journal, 6(3–4), 275–93.

15.

Classens

Dasgupta

Glen

(1995). Return behaviour in emerging stock markets. World Bank Economic Review, 9(1), 131–51.

16.

Classens

Dasgupta

Glen

(1998). The cross section of stock returns: Evidence from the emerging markets. Emerging Market Quarterly, 9(1), 4–13.

17.

Dash

S.R.

Mahakud

(2012). Investor sentiment, risk factors and stock return: Evidence from Indian non-financial companies. Journal of Indian Business Research, 4(3), 194–218.

18.

Daniel

Hirshleifer

Subrahmanyam

(1998). Investor psychology and security market under and overreactions. Journal of Finance, 53(6), 1839–85.

19.

Daniel

Hirshleifer

Teoh

(2002). Investor psychology in capital markets: Evidence and policy implications. Journal of Monetary Economics, 49(1), 139–209.

20.

De Long

Bradford

A.S.

Lawrence

H.S.

Robert

J.W.

(1990). Noise trader risk in financial markets. Journal of Political Economy, 98(4), 703–38.

21.

Fama

MacBeth

(1973). Risk, return, and equilibrium: empirical tests. Journal of Political Economy, 81(3), 607–36.

22.

Fama

E.F.

French

K.R.

(1992). The cross-section of expected stock returns. Journal of Finance, 47(2), 427–65.

23.

Fama

E.F.

French

K.R.

(1993). Common risk factors in the returns on stocks and bonds. Journal of Financial Economics, 33(1), 3–56.

24.

Fisher

K.L.

Statman

(2000). Investor sentiment and stock returns. Financial Analysts Journal, 56(2), 16–23.

25.

Finter

Niessen-Ruenzi

Ruenzi

(2011). The impact of investor sentiment on the German stock market. CFR-Working Paper No. 10–03, Centre for Financial Research (CFR), University of Cologne, Germany, October 2011.

26.

Granger

C.W.J

(1969). Investigating causal relations by econometric models and cross spectral methods. Econometrica, 37(3), 424–38.

27.

Granger

C.W.J

(1988). Some recent developments in the concept of causality. Journal of Econometrics, 39(1), 199–211.

28.

Her

L.J.F.

Masmoudi

Suret

M.J.

(2004). Evidence to support the four factor pricing model from the Canadian stock market. Journal of International Financial Markets, 14(4), 313–28.

29.

Hirshliefer

(2001). Investor psychology and asset pricing. Journal of Finance, 56(4), 1533–98.

30.

Hong

Stein

J.C.

(1999). A unified theory of underreaction, momentum trading, and overreaction in asset markets. The Journal of Finance, 54(6), 2143–84.

31.

Jegadeesh

Titman

(1993). Returns to buying winners and selling losers: Implications for stock market efficiency. Journal of Finance, 48(1), 65–91.

32.

Keene

A.M.

Peterson

R.D.

(2007). The importance of liquidity as a factor in asset pricing. Journal of Financial Research, 30(1), 91–109.

33.

Keynes

J.M.

(1936). The general theory of employment, interest, and money. New York: Harcourt Brace.

34.

Kumar

Lee

C.M.C

(2006). Retail investor sentiment and return comovements. The Journal of Finance, 51(2), 2451–86.

35.

Lam

K.S.K.

Tam

K.H.L

(2011). Liquidity and asset pricing: Evidence from the Hong Kong stock market. Journal of Banking and Finance, 35(9), 2217–30.

36.

Misra

A.K.

Mahakud

(2009). Emerging trends in financial markets integration. International Journal of Emerging Markets, 4(3), 235–51.

37.

Rabin

(1998). Psychology and economics. Journal of Economic Literature, 36(1), 11–46.

38.

Ritter

J.R.

(2003). Behavioral finance. Pacific-Basin Finance Journal, 11(4), 429–37.

39.

Rouwenhorst

K.G.

(1999). Local return factors and turnover in emerging stock markets. Journal of Finance, 54(4), 1439–64.

40.

Schmeling

(2009). Investor sentiment and stock returns: Some international evidence. Journal of Empirical Finance, 16(3), 394–408.

41.

Sehgal

Jain

(2011). Short-term momentum patterns in stock and sectoral returns: Evidence from India. Journal of Advances in Management Research, 8(1), 99–122.

42.

Shefrin

(2005). A behavioural approach to asset pricing. New York: Elsevier Academic Press.

43.

Shiller

(2000). Irrational exuberance. New Jersey: Princeton University Press.

44.

Shiller

R.J.

Fischer

Friedman

B.M.

(1984). Stock prices and social dynamics. Brookings Papers on Economic Activity, 1984(2), 457–98.

45.

Shleifer

Vishny

(1997). The limits of arbitrage. Journal of Finance, 52(1), 35–55.

46.

Shleifer

Summers

L.H.

(1990). The noise trader approach to finance. Journal of Economic Prospective, 4(2), 19–33.

47.

Stracca

(2004). Behavioral finance and asset prices: Where do we stand? Journal of Economic Psychology, 25(3), 373–405.

48.

Verma

Soydemir

(2009). The impact of individual and institutional investor sentiment on the market price of risk. The Quarterly Review of Economics and Finance, 49(3), 1129–45.

49.

White

(1980). A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity. Econometrica, 48(4), 817–38.