Investor sentiment, stock market returns and volatility: evidence from National Stock Exchange of India
Pramod Kumar Naik* Department of Economics, Central University of Rajasthan, Bandarsindri 305817, Ajmer, Rajasthan, India Email: [email protected] *Corresponding author
Puja Padhi Department of Humanities and Social Sciences, Indian Institute of Technology Bombay, Mumbai 400076, India Email: [email protected]
Abstract: This study explores the relationship between investor sentiment and stock return volatility using monthly data from National Stock Exchange (NSE) of India over July 2001 to December 2013 period. Using seven market-related implicit indicators a sentiment index has been constructed with the help of principal component analysis. Then the analysis has been done by employing ordinary least squares methods, vector autoregression, Granger causality and EGARCH-M models. Findings show that sentiment index significantly influences market excess returns. At the first glance it was found that sentiment has negative influence on the conditional volatility. However, when the sentiment index is decomposed into positive sentiment and negative sentiment changes, the study reveals that positive and negative sentiments have asymmetric impacts on excess return volatility. The Granger causality results suggest a bi-directional causality between excess return and investor sentiment at the third lags.
Reference to this paper should be made as follows: Naik, P.K. and Padhi, P. (2016) ‘Investor sentiment, stock market returns and volatility: evidence from National Stock Exchange of India’, Int. J. Management Practice, Vol. 9, No. 3, pp.213–237.
Biographical notes: Pramod Kumar Naik is an Assistant Professor in the Department of Economics in Central University of Rajasthan. He obtained his MPhil and PhD from Indian Institute of Technology Bombay. He has published several articles in both international as well as national journals such as, Review of Accounting and Finance, Global Business Review, Indian Journal of Finance etc. His research interest includes financial economics, stock market volatility, corporate finance, applied econometrics time-series and panel data, industrial economics, money and banking.
214 P.K. Naik and P. Padhi
Puja Padhi is an Associate Professor of Economics in the Department of Humanities and Social Sciences in Indian Institute of Technology Bombay. She has published a number of journal articles in international as well as national journals, such as, Journal of Multinational Financial Management, Journal of Business Economics and Management, Journal of Empirical Finance, Transition Studies Review, Review of Accounting and Finance, Journal of Quantitative Economics etc. Her research area includes financial economics, stock market volatility, applied econometrics, money and banking.
This paper is a revised and expanded version of a paper entitled ‘Investor sentiment, stock market returns and volatility: evidence from National Stock Exchange India’ presented at the ‘International Conference on Evidence Based Management (ICEBM2015)’, BITS Pilani, India, 20–21 March 2015.
1 Introduction
The relationship between investor sentiment1 and stock returns has long been a considerable debate in the field of empirical and behavioural finance. Literature provides contradicting view on the sentiment-based traders’ impact on stock prices. The Efficient Market Hypothesis (EMH) introduced by Fama (1965) believes that financial markets are information efficient. Further, the Capital Asset Pricing Model (CAPM) holds that, in an efficient financial market, the prices of traded assets reflect all available information about market fundamentals. The argument is that investors in such markets are rational who always force the market prices to equal the present value of expected future cash flow; even if some investors are irrational their demands are offset by the arbitrageurs.2 Thus, according to this theory, investor sentiment has no role in determining stock prices volatility. The EMH was popular and dominant in the academic circles during the 1970s (Shiller, 2003). However, this classical finance theory failed to explain various market anomalies, such as the 1987 crash, the Dot.com bubble of 1990s, etc.
The belief of rationality in EMH has been challenged by modern empirical and behavioural finance literature. Under the behavioural finance theory, it is argued that investors are not necessarily rational, and they may prone to exogenous sentiment waves. Their overly optimism and pessimism about the market lead them to bias of irrationalities in investment decision. Thus, investor sentiment can induce systematic risk and affect asset price in equilibrium. Amongst many, Shiller (1981) argues that investors are not fully rational and thus stock prices may be affected by factors other than fundamentals. Similarly, Black (1986) and De Long et al. (1990) believe some of the market anomalies can be caused by the traders who trade based on noise or sentiment. According to Black (1986) such investors are known as noise traders since they do not have access to inside information and act irrationally on noise. If the uninformed noise traders make their trading decision on the basis of sentiment then measure of it may have predictive power for stock price behaviour (Wang et al., 2006). De Long et al. (1990) and Shefrin and Statman (1994) theoretically show that investor sentiment is an intrinsic factor that affects the asset prices in equilibrium. In literature it is termed as the ‘noise traders’ theory which suggests that if some investors trade on a noisy signals, that are unrelated to fundamentals, then asset prices would deviate from their intrinsic value. Thus, investor sentiment should influence the stock returns as well as the market volatility.
Investor sentiment, stock market returns and volatility 215
Supporting the ‘noise traders’ theory of De Long et al. (1990), Lee et al. (2002) empirically show that sentiment is a systematic risk that is priced. They document a positive relationship between excess returns and change in sentiment, and a negative relationship between the shifts in sentiment and the conditional volatility. Fisher and Statman (2000) analyse the relationship between level as well as change in sentiment and future stock returns considering three groups of investors such as the Wall Street strategists (large), the writers of investment newsletters (medium), and individual investors (small). They document a negative relationship between future returns and sentiment of both small and large groups. They also document that returns generate sentiment of small and medium groups of investors. Brown and Cliff (2004) find that investor sentiment (both in level and changes) are significantly correlated with the contemporaneous returns. They conclude that past returns are the important determinants of investor sentiment. Brown and Cliff (2005) indicate that investor sentiment predicts market returns over one to three years. Kumar and Lee (2006) examine whether the retail investor sentiment can explain the return co-movement and find that it has a significant ability to explain the return co-movements. Wang et al. (2006) investigate the causal relationship between sentiment, returns and volatility, and find that investor sentiments caused by market returns and volatility rather than vice versa.
The most influential study of Baker and Wurgler (2006, 2007) find that when the sentiment is low the returns are relatively high, whereas when sentiment is high the subsequent returns are low. Following their study and their sentiment index construction, several subsequent studies investigate the relationship between sentiment index and stock returns in various ways. For example, Yu and Yuan (2011) examine the impact of investor sentiment on market returns and volatility using a similar index of investor sentiment and find that the expected returns are positively related to volatility in the low sentiment periods but not significantly related in high sentiment periods. Dergiades (2012) also uses the similar index and report that investor sentiment has a significant influence on stock returns. Qiang and Shu-e (2009) find that the impacts due to positive and negative changes in sentiments are different in affecting stock price fluctuations. Chuang et al. (2010) conclude that the volatility that generated from investor sentiment gives rise to idiosyncratic risk rather than systematic risk.
Other studies, such as, Verma and Verma (2007) indicate that investor sentiment has a positive effect on stock returns but has a negative effect on market volatility for both individual and institutional investors. Verma and Soydemir (2009) find that individual and institutional investor sentiments are driven by both rational and irrational factors. Their findings indicate that when the noise traders are bullish the rational investors are bearish and when the noise traders are bearish the rational investors are bullish. Lux (2011) finds a feedback relationship between the stock returns and sentiment. Zhu (2012) shows a strong correlation between sentiment index and Shanghai stock market index. Changsheng and Yongfeng (2012) also show that investor sentiment has incremental power to explain return co-movement indicating that when investors are bullish the stock return is high and it is low when the investors are bearish. Li (2014) shows that the sentiment index has a good predictive power of Chinese stock market return. Huang et al. (2014) find that sentiment has positively related to current periods stock returns, whereas negatively related with lagged stock returns. Perez-Liston et al. (2014) estimate GARCH-in-mean model and VAR model to find that changes in investor sentiment have a positive influence on excess returns. Their analysis also shows that the bullish shift in investor sentiment has negative effect on conditional volatility. Yang and Copeland (2014) find
216 P.K. Naik and P. Padhi
that sentiment index has an asymmetric effect on both short-run and long-run volatility. They conclude that the bearish sentiment leads to a lower excess returns and the bullish sentiment leads to higher market returns; the bullish sentiment has a positive influence on short-run volatility, whereas it has a negative impact on long-run volatility.
Owing to these diverse explanations several studies tried to explore the impact of investor sentiment on stock returns, using several different proxies since the sentiment is not directly observable and quantifiable. Most studies discussed the positive and negative impacts of investor sentiments on the current and expected future stock returns. However, as Wang et al. (2006) note, if the investors are sensitive to the changes in sentiment, then, it would affect both return and volatility. Alternatively, if trading of noise traders is on the basis of extreme sentiment, i.e. they only trade when the sentiment is extremely high or low relative to previous levels, then the sentiment level is expected to affect returns and volatility. Moreover, one can also argue that investor sentiments, generally, generated through the market behaviour, and therefore returns should influence the investor sentiment change. Despite the well-documented literature of investor irrationality, only a handful of studies have been devoted to investigate how different states of sentiment are related to stock return and its volatility (see, e.g., Lee et al., 2002; Qiang and Shu-e, 2009; Chuang et al., 2010; Yu and Yuan, 2011; Huang et al., 2014; Yang and Copeland, 2014) and also the dynamic interaction as well as the direction of causality between them (e.g. Wang et al., 2006; Lux, 2011; Verma and Soydemir, 2009; Perez-Liston et al., 2014).
In the present study, we explore the role of investor irrational sentiment in influencing market excess returns and volatility considering an Indian equity market NSE. The basic objective is to investigate whether Indian equity market is driven by the investor sentiment or the market excess returns influence investor sentiment. In order to accomplish the objective we closely follow the approaches similar with Lee et al. (2002), Qiang and Shu-e (2009), Chuang et al. (2010), and Perez-Liston et al. (2014). More specifically, we examine the impact of positive and negative sentiment that has on market excess returns by following similar approach with Qiang and Shu-e (2009). The effects of different states of sentiment (i.e. positive and negative changes in sentiment) on market excess return are examined by augmenting the sentiment change in the conditional volatility model, similar with Lee et al. (2002), Qiang and Shu-e (2009), Chuang et al. (2010), and Perez-Liston et al. (2014). The dynamic interaction between investor sentiment and excess return has been examined by adopting approaches similar with Verma and Soydemir (2009), Lux (2011) and Perez-Liston et al. (2014). From the best of our knowledge this type of analysis has not yet been conducted using Indian data. Recently, Bennet et al. (2012), Dash and Mahakud (2012, 2013a, 2013b), and Chandra and Thenmozhi (2013) examined the impact of investor sentiment using Indian data. However, these studies primarily dealt with one part of the analysis, i.e. the overall impact of investor sentiment on stock returns, and the role of different states of investor sentiment, if any, in generating market volatility has been neglected.
It is expected that investigation of our kind would extend the growing literature in the context of emerging market economy by providing more ideas about the market behaviour, and discussing how a set of investors, i.e. the noise traders with their different states of sentiment, affect market excess return. As Fisher and Statman (2000) note, analysis in this kind would enable us to know about the biases in stock market forecast of investors; and it teaches us about the opportunities to earn extra returns by exploiting those biases.
Investor sentiment, stock market returns and volatility 217
The present work has been organised as follows. Section 2 deals with the data and empirical methodology, in which the construction of investor sentiment index, the data sources and the empirical framework have been described. The empirical findings are discussed in Section 3 and finally Section 4 concludes the study.
2 Empirical methodology and data
2.1 Construction of investor sentiment index
As there is no definitive indicator available, many previous studies adopted different proxies for investor sentiment. This is due to the fact that investor sentiment is not directly and physically observable. Existing literature establishes several different measures to represent the unobservable sentiment index, which can broadly be classified, at least, as follows. The first is the direct survey from the individual and institutional investors for their anticipated movement of stock market and aggregate economy (e.g. Fisher and Statman, 2000; Brown and Cliff, 2005; Verma and Soydemir, 2009; Schmeling, 2009; Lux, 2011). Studies also use investors’ intelligence sentiment index as a proxy variable (e.g. Lee et al., 2002; Perez-Liston et al., 2014). The second is the market-related implicit sentiment proxies (e.g. Baker and Wurgler, 2006; Baker and Wurgler, 2007; Wang et al., 2006; Brown and Cliff, 2004; Baker et al., 2012; Li, 2014).
Since the direct survey measure of investor sentiment has several limitations, such as likelihood of errors in the stage of data collection and processing, its limited scope of generalisation, unscalability, etc., many recent studies used the market-related implicit proxies as the indicators of investor sentiment. Unlike the direct survey, the market-related proxies have advantages of representing the mood of the economy, easily generalised and often available from the most authentic sources. However, there are no specific numbers of factors to represent these market-related implicit proxies. Different studies use different proxies according to the nature of the analysis. For example, the behavioural finance literature in the 1990s considered closed-end fund discounts to be the appropriate variable to represent the investment sentiment indicator (e.g. Lee et al., 1991; Swaminathan, 1996; Neal and Wheatley, 1998). Some studies also considered the trend in trade variables, such as trading volume and turnover ratios (e.g. Jackson, 2003; Kumar and Lee, 2006; Chuang et al., 2010). The contemporary studies modify the approach and construct a conglomerate sentiment index by combining several market-related implicit proxies. The most influential study in this regard is Baker and Wurgler (2006, 2007). They use six variables, namely closed-end fund discounts, number of IPO (NIPO), IPOs first day returns, turnover ratio, equity-debt ratios, and dividend premium to construct a composite sentiment index employing Principal Component Analysis (PCA). Following Baker and Wurgler (2006, 2007) many subsequent studies use several numbers of proxies and their composite sentiment index. Table 1 represents some of the previous studies in this regard.
Based on the literature and the availability of the data we select Advance Declining Ratios (ADRs), Put-Call Ratios (PCR), NIPO, PE ratios (PER), turnover rates (TURN), trading volumes (TV), and Mutual Funds Net Flow (MFNF) to construct a composite investor sentiment index. ADR represents the ratio of the number of advancing and declining stock prices. It helps to know the recent trend of the stock market performance. A rising values of ADR means the upward trend of the market; and a lower value shows
218 P.K. Naik and P. Padhi
the downward trend of the market. Brown and Cliff (2004), Wang et al. (2006), and Dash and Mahakud (2013a) use this variable as the indicator of market sentiment. The variable PCR represents the derivative trading activities. PCR is the ratio of trading volumes of put-call options. This measure is widely accepted as a bearish indicator, i.e. in a bear market period the PCR is high and in a bull market period it is low (Brown and Cliff, 2004). NIPO is the commonly used variable in the recent studies (e.g. Brown and Cliff, 2004; Baker and Wurgler, 2006; Baker and Wurgler, 2007; Kurov, 2010; Yu and Yuan, 2011; Changsheng and Yongfeng, 2012; Baker et al., 2012; Corredar et al., 2013; Dash and Mahakud, 2013a; Li, 2014). It is argued that NIPO can be considered as the sentiment indicator since demand for IPO is often sensitive towards the market condition.
Table 1 List of investor sentiment proxies used in related studies
Studies Measure of sentiment
Lee et al. (2002) Investor intelligent index
Brown and Cliff (2004)
Advance and declining ratio, high and low ratio, margin borrowings, short interest, short sales, odd lot sales to purchase, put-call ratio, SPX future (institutional sentiment, activity of small traders), monthly forecast of commodity market returns, expected volatility relative to current volatility, closed-end fund discounts, mutual fund flows, fund cash, first-day IPO returns and number of IPO
Brown and Cliff (2005) Survey data of American Association of Individual investors
Kumar and Lee (2006) Buy–Sell imbalance ratio
Wang et al. (2006) Put-call trading volume ratio, put-call open interest ratio, ARMS index (advance decline ratio), survey data of American Association of Individual investors, investor intelligence index
Baker and Wurgler (2006, 2007)
closed-end fund discounts, number of IPO, IPOs first-day returns, turnover ratio, equity-debt ratios, and dividend premium
Qiang and Shu-e (2009) Turnover, closed-end fund discounts, growth rate of investors account
Schmeling (2009) Consumer confidence index
Verma and Soydemir (2009)
Survey data of individual and institutional sentiment similar with Brown and Cliff (2005)
Chuang et al. (2010) Trading volume
Yu and Yuan (2011) Baker and Wurgler (2006, 2007) measures
Lux (2011) Survey data
Dergiades (2012) Baker and Wurgler (2006, 2007) measures
Zhu (2012) PE ratio, trading volume, turnover, closed-end fund discount, new account amounts, VIX index
Changsheng and Yongfeng (2012)
IPO, closed-end fund discounts, turnover, number of new stock accounts
Investor sentiment, stock market returns and volatility 219
Table 1 List of investor sentiment proxies used in related studies (continued)
Studies Measure of sentiment
Rehman (2013) Baker and Wurgler (2006, 2007) measures
Dash and Mahakud (2013a)
Turnover volatility ratio, share turnover velocity, advance declining ratio, margin borrowings, buy–sell imbalance ratio, put-call ratio, number of IPO, equity issue in total issue, dividend premium, mutual fund flow, cash to total asset in mutual fund market, price-to-earnings high–low ratio difference
Perez-Liston et al. (2014) Investor intelligence
Li (2014) Closed-end fund discounts, turnover, number of IPO, first-day return of IPO, number of Chinese A shares net-added accounts, relative degree of active trading in equity market
We also use the Price to Earnings Ratio (PER) as it reflects both the price of the stock market and the financial situation of the listed companies in the macroeconomic environment. The PER is often positively correlated with the market index and thus a higher value of PER represents higher market sentiments (see also Sehgal et al., 2009; Zhu, 2012). TURN and TV represent the market liquidity. In a highly liquid market the irresistible investors cause the frequent trading so that the trading volume is high so as the turnover rates. Thus, previous studies such as Qiang and Shu-e (2009), Zhu (2012) and Li (2014) use turnover rate as a sentiment indicator. Chuang et al. (2010) uses trading volume as the investor sentiment index. Following Brown and Cliff (2004), Chi et al. (2012) and Dash and Mahakud (2013a) we use the mutual fund net flow as another sentiment proxy since fund flow suggests the importance of a preferred asset class, and as an economic substitute by the market participants. The mutual fund investors are well known to chase investment in high returns.
It should be noted that the aforementioned sentiment indicators may partially contain the rational expectation-based risk factors and hence likely to be contained fundamental (rational) and non-fundamental (irrational) components (see Shleifer and Summers, 1990; Brown and Cliff, 2004; Brown and Cliff, 2005; Baker and Wurgler, 2006; Baker and Wurgler, 2007; Verma and Soydemir, 2009; Dash and Mahakud, 2012). In the present study, our goal is to test the impact of investor irrational sentiment on stock market return and volatility. Thus, we follow Baker and Wurgler (2006), Verma and Soydemir (2009) and Dash and Mahakud (2012) approaches to obtain the irrational sentiment component. For this purpose, each of the underlying market-related implicit proxies have been regressed on the six important macroeconomic fundamentals, namely the growth rate of industrial production index (IIP), the rate of inflation (INF), exchange rate (EXRT), short-term interest rate (INT), term spread (TS),3 and net investment of foreign institutional investors (FINI). This process removes the business cycle variation from each of the seven sentiment proxies prior to the construction of final composite sentiment index (Baker and Wurgler, 2006). We estimate the following equation:
0 t1
= + Funda +k
t k k tk
Y (1)
where Yt represents each of the market-related implicit sentiment proxies undertaken in the study, α0 is the constant, βk is the parameter to be estimated, Funda represents the aforementioned macroeconomic fundamentals, K represents the number of
220 P.K. Naik and P. Padhi
macroeconomic fundamentals, and εt is the error term. The fitted values of equation (1) would provide the rational components, whereas the residuals capture the irrational component of sentiment (see Verma and Soydemir, 2009). We then obtain the residual from equation (1) to have the orthogonal implicit sentiment proxies and used them in the subsequent analysis. After obtaining each of the orthogonal sentiment proxy from equation (1) we then construct a conglomerate market sentiment index to represent the aggregate investor sentiment employing the PCA. The PCA transforms the original set of variables into a smaller set of linear combinations that account for the most of the variance of the original set. The benefit of PCA is that it filters out idiosyncratic noise in the orthogonal sentiment measures and captures their common components. The aim here is to extract the factors loadings for our original sentiment proxy indicators. After obtaining the factor loadings for each orthogonal sentiment indicators a composite sentiment index has been constructed using the following formula.
1
SentIndexk
tt
j tj
Ya
Y
(2)
where aj is the factor loadings for j-th item derived by the PCA, K is the number of sentiment proxies used, and Yt and σYt are the sentiment proxies and their standard deviation, respectively.
It is argued in literature that some proxies take longer time to reveal the sentiment. For this reason we follow the approach similar with Baker and Wurgler (2006) and estimate the PCA for our essential variables with levels as well as with their lags. For example, we have seven sentiment proxies and with their lags leaving us to have 14 loadings. In the spirit of Baker and Wurgler (2006) the final sentiment index has been identified by adopting the following steps. First, a raw sentiment index has been calculated using the seven orthogonal sentiment proxies and their lags that provide us 14 factor loadings. Second, compute the correlation coefficient between this first stage raw sentiment index and the seven orthogonal proxies and their lags. Third, we select the variables with or without lags (whichever has higher correlation with the first stage raw sentiment index) and then construct the final sentiment index. The first principal component having 28% sample variance gives the following measure of sentiment index.
1
1 1
SentIndex 0.003 0.575 0.536 0.530
0.112 0.120 0.286t t t t t
t t t
ADR PCR NIPO PER
TURN TV MFNI
The correlation coefficient between the 14-terms first stage sentiment index and the seven-terms final sentiment index is 0.955, suggesting that the risk of losing substantial information by dropping the seven implicit sentiment proxies is very less. Figure 1 depicts the trend and the relationship between the first stage raw sentiment index and the final sentiment index. A close correlation between these two stages sentiment index is clearly visible. Figure 2 depicts the trend of sentiment index and the Nifty closing index for the period from July 2001 to December 2013. It can be observed from the figure that these two indices show a similar trend. It seems that investor sentiment gradually increased with the increase in Nifty index, but SentIndex fluctuates more than the Nifty index. It also seems that during the US sub-prime crisis both the indices fall together and then recovered gradually. Figure 3 shows the trend of sentiment index and the trend of change in sentiment index, comparing them with the trend of market excess return. It
Investor sentiment, stock market returns and volatility 221
seems from the figure that all the three lines on an average are approaching each other. It can be observed that when the stock market index, as well as excess market return, goes upward the sentiment index also goes up, but when the market index and excess return approach to a downward trend the investor tends to become more bearish. However, the figures only tell about the trend and not the cause and effect relationship between these two. Statistical test therefore is necessary.
Figure 1 Trend of investor sentiment index (final) and sentiment index (first stage) from July 2001 to December 2013 (see online version for colours)
Figure 2 Trend of investor sentiment index and Nifty index from July 2001 to December 2013 (see online version for colours)
Figure 3 Trend of investor sentiment index, change in investor sentiment index and Nifty excess returns from July 2001 to December 2013 (see online version for colours)
222 P.K. Naik and P. Padhi
Table 2 Descriptive statistics of market-related implicit sentiment proxies and other macroeconomic variables
Var
iabl
es
Mea
n M
ax
Min
St
d. D
ev.
Skew
ness
K
urto
sis
Jarq
ue–B
era
Pro
b.
Mar
ket-
rela
ted
impl
icit
pro
xies
for
sent
imen
t
AD
R
0.93
3 1.
800
0.55
0 0.
219
0.73
8 4.
006
19.9
37
0.00
0
PC
R
0.88
4 1.
462
0.33
0 0.
204
–0.4
45
2.94
3 4.
972
0.08
3
NIP
O
3 18
0
3.35
7 1.
535
5.97
3 11
4.10
7 0.
000
PE
R
18.0
24
27.6
00
11.7
00
3.30
0 0.
218
2.55
0 2.
453
0.29
3
TU
RN
18
8410
.7
4965
89.0
28
572.
0 10
3118
.5
0.42
3 2.
697
5.05
6 0.
080
TV
81
5.06
7 18
19.0
00
97.0
00
446.
151
–0.0
90
1.61
7 12
.155
0.
002
MF
NI
–116
.920
78
93.0
00
–723
6.00
0 20
21.8
26
0.27
8 6.
489
78.0
23
0.00
0
Mac
roec
onom
ic fa
ctor
s
IIP
Gro
wth
0.
789
14.9
40
–14.
264
5.53
6 –0
.194
3.
697
3.97
5 0.
137
WP
IGro
wth
0.
468
2.57
9 –2
.500
0.
693
–0.4
67
5.76
0 53
.066
0.
000
EX
RT
47
.313
63
.752
39
.374
4.
802
1.27
7 5.
085
67.9
52
0.00
0
TB
ILL
91
6.39
5 11
.510
3.
227
1.68
3 0.
185
2.68
1 1.
488
0.47
5
TS
0.
185
2.59
2 –1
.840
0.
498
0.41
9 8.
429
188.
610
0.00
0
FII
NI
4913
.647
35
228.
000
–441
62.0
00
1036
0.26
0 –0
.005
6.
360
70.5
54
0.00
0
Stoc
k m
arke
t ind
ex
NIF
TY
36
16.8
50
6246
.900
94
9.40
0 17
92.1
74
–0.1
81
1.50
7 14
.747
0.
001
LnR
t 0.
012
0.18
1 –0
.270
0.
062
–0.7
77
5.59
5 57
.164
0.
000
Investor sentiment, stock market returns and volatility 223
2.2 Data set and summary statistics
National Stock Exchange (NSE) of India has been considered to represent the Indian stock market4 and an aggregate investor sentiment index has been constructed based on the market-related implicit proxies. We choose seven market-related implicit sentiment indicators based on their use in the literature and relevance with the Indian stock market behaviour. The details of these variables are already discussed in the above subsection. This study utilised monthly data for the period from July 2001 to December 2013. The sample period is based on the availability of data for all the variables used in this study. All the stock market-related data are extracted from a single source, i.e. Handbook of Statistics on Indian Security Market 2012 and 2013 provided by SEBI year books. Monthly closing prices of S&P CNX Nifty are converted into compounded log return as
1
ln tt
t
Pr
P
where rt is the compounded return at time t and Pt and Pt–1 are the monthly
stock index at the two successive months t and t – 1, respectively. Then the risk-free interest rate has been subtracted to obtain the excess return. Monthly data of macroeconomic variables are obtained from Handbook of Statistics on Indian Economy provided by Reserve Bank of India.
Table 2 reports the descriptive statistics of all the variables used in the first stage analysis. From Table 2 it can be observed that during the period of investigation the mean values of all the sentiment indicators except the net investment of mutual funds are positive. The negative mean value of mutual fund net flow indicates that on an average the mutual fund investor sells more than they purchase stocks. While the positive value of put-call option ratio shows the market as bearish sentimental the mean values of ADR, price-to-earnings ratio and the market liquidity measures indicate the market has been performing well. This is also evident from the mean value of Nifty index and returns. The average number of IPO issues in the study period 3 per month with a minimum of 0 and maximum of 18.
Table 3 Correlation matrix of market-related orthogonal sentiment proxies
ADR PCR NIPO PER TURN TV MFNI
ADR 1
PCR 0.011 1
NIPO –0.055 0.533 1
PER 0.124 0.486 0.376 1
TURN 0.139 0.077 0.059 0.018 1
TV 0.125 –0.045 –0.041 –0.072 0.868 1
MFNI –0.012 –0.144 –0.096 –0.249 0.026 0.076 1
Table 3 reports the cross-correlation among the investors irrational sentiment indicators, i.e. the orthogonal sentiment proxies. It can be seen from this table that on an average all the sentiment proxies are positively correlated with each other but the magnitudes are lesser. Table 4 reports the descriptive statistics of the final variables of interest. The mean value of the irrational sentiment is positive but very less in magnitude which is not sufficient to justify a bullish sentiment. It may be noted that during the study period the average monthly return is 0.012 (see Table 2), while that of the excess return is –0.052.
224 P.K. Naik and P. Padhi
The correlation between the sentiment index and the excess returns, however, is positive. This suggests that Indian investors became more conservative when the market shows downward trend and the market sentiment decreases. It can therefore be hypothesised that when sentiment changes to positive the excess return will be higher and when the sentiment turns to negative the excess return will be lower.
Table 4 Descriptive statistics of final sentiment index and market excess return
SentIndex_Final SentIndex_Raw Excess Return
Mean 0.010 0.004 –0.052
Median 0.050 0.082 –0.046
Max 4.700 5.139 0.142
Min –2.940 –3.597 –0.351
Std. Dev. 1.401 1.888 0.068
Skewness 0.291 0.207 –0.604
Kurtosis 2.847 2.462 5.192
Jarque–Bera 2.243 2.861 38.887
Prob. 0.326 0.239 0.000
Obs. 149 149 149
Correlation matrix
SentIndex_Final 1
SentIndex_RAW 0.955 1
ExcessReturn 0.257 0.175 1
ADF test statistics
ADF t-stat –4.209**
(0.000)
–2.735*
(0.070)
–4.738**
(0.000)
Notes: The probability values for ADF statistics are in parenthesis.
** indicates statistical significant at 1% level and * indicates statistical significant at 10% level.
2.3 Empirical methodology
The analysis has been done by employing various econometric techniques, such as the Ordinary Least Squares (OLS), vector autoregression (VAR), Granger Causality, and
the Exponential Generalised Autoregressive Conditional Heteroskedasticity-in-mean (EGARCH-M) methods. After constructing the aggregate investor sentiment index, its impact on excess stock market return and volatility has been tested. Before employing these econometrics techniques, the stationary properties of both the time-series have been confirmed by the Augmented Dickey–Fuller (ADF) unit root tests.
We start the analysis by employing the OLS method. The regression equation for the purpose is described as follows.
1SentIndex ;ft t t t t t tr r (3)
Investor sentiment, stock market returns and volatility 225
where rt stands for monthly log return of market index, ftr represents the risk-free
interest rate, α is the constant, β is the parameter to be estimated, SentIndex represents the investor sentiment index, and μt is the error term. The first-order autoregressive residual is added to the equation to control for serial correlation of residuals. It has often been argued in the literature that different impacts of positive and negative sentiment as well as the changes in positive and negative sentiment have likely to influence differently on the excess stock returns. To test this belief, a dummy variable Dt has been added. We define Dt is equal to 1 when the sentiment index value is positive, and 0 otherwise. Similar definition has been used to describe the change in sentiment index. Therefore, the influence of positive and negative sentiment on excess return has been tested by estimating the following equation.5
2 2
0 1 2
1
SentIndex SentIndex 1t t
ft t t t t
t t t
r r D D
(4)
where β1 and β2 represent the coefficient of positive and negative sentiment, respectively, representing the bullish and bearish shift in investor irrational sentiment. Similarly, the impacts of positive sentiment changes and negative sentiment changes have been tested by estimating the following equation. However, in this case the dummy Dt is equal to 1 when the changes in sentiment index are positive, and 0 otherwise. The regression equation is specified as follows.
2 2
0 1 2
1
SentIndex SentIndex 1t t
ft t t t t
t t t
r r D D
(5)
where ΔSentIndext represents the change in sentiment index, i.e. SentIndext – SentIndext–1. The OLS analysis will provide us the direct impact of investor sentiment on excess return. It does not provide sentiment impact on market volatility, i.e. whether the fluctuation in investor sentiment is a systematic risk and obtains risk premium (Qiang and Shu-e, 2009). For this purpose previous studies, such as Lee et al. (2002), Verma and Verma (2007), Qiang and Shu-e (2009), Chuang et al. (2010) and Yang and Copeland (2014), augmented sentiment index in the volatility models like GARCH-in-mean and EGARCH-in-mean. Following these literatures and assuming that Indian stock market exhibits volatility asymmetry, we test the impact of positive and negative investor sentiment on returns as well as volatility using the EGARCH-M model. Accordingly the testable EGARCH (1, 1)-M model has been represented as follows.
1 11
1 1 1 1 1 1 1 1 1
ln SentIndex
ln ln SentIndex
f ft t t t t t tt
t t t t t t t
r r a h b r r c
h h h h
(6)
where ft tr r is the excess return, the δht represents the market risk premium for the
expected volatility, and εt is the error term. The trade-off parameter δ can take positive, negative and a zero value. The log of the conditional variance in the left-hand side implies that the asymmetric effect is exponential rather than quadratic and that forecast of
226 P.K. Naik and P. Padhi
conditional variance is guaranteed to be non-negative. The terms ω, α1, β1, γ1, and θ1 are the parameters to be estimated. In order to incorporate the impact of positive change and negative change of sentiment, equation (6) is modified as follows.
1 11
2
1 1 1 1 1 1 1 1 1
2
2 1
ln SentIndex
ln ln SentIndex
SentIndex 1
f ft t t t t t tt
t t t t t t tt
ft tt
r r a h b r r c
h h h h D
D r
(7)
In the spirit of Lee et al. (2002), Chuang et al. (2010) and Perez-Liston et al. (2014) we add the risk-free interest rate in the conditional volatility in equation (7).
Further, some of the previous studies suggest that investor sentiment and stock market return may act as a system (see Brown and Cliff, 2004; Brown and Cliff, 2005; Verma and Soydemir, 2009; Perez-Liston et al., 2014). OLS and the conditional volatility model do not able to control for the endogeneity bias, and they unable to provide the direction of causality. Thus, a VAR model has also been estimated which is free from endogeneity problem. The aim here is to test the dynamic interaction of sentiment and market excess returns and any statistical causality exists between them.
The causality equations are expressed in a VAR framework as follows.
1 1 1 11 1
Excessreturn = Excessreturn SentIndexp p
Rt i t i t i t
i i
(8)
2 2 1 2 11 1
SentIndex = SentIndex Excessreturnp p
St i t i t t
i i
(9)
where α1 and α2 are the intercepts, β and γ are the parameters to be estimated, and Rt and
St are the white noise error terms, and p denotes the lag lengths. In equation (8),
sentiment index Granger cause market excess return if either γ1i are jointly significant by testing null hypothesis of H0: γ11 = γ12 = … = γ1p = 0. Similarly, in equation (9) market excess return Granger sentiment index if either β1i are jointly significant.
3 Empirical results
The stationary properties of the variables SentIdex and ExcessReturn have been tested through the ADF test and are reported in Table 4. The results clearly show that both SentIndex and ExcessReturns are stationary at level. The analysis has been started with estimating the OLS regression. The OLS regression has been conducted in three different specifications. The results are reported in Table 5. Specification1, specification2 and specification3 represent the results of regression equations (3), (4) and (5), respectively. It can be observed from Table 5 that sentiment index positively influences market excess returns. When the sentiment index is decomposed into positive sentiment and negative sentiment, an asymmetric relationship has been observed. It is evident that while the positive sentiment index has a positive impact on market excess return the negative sentiment index has negative impact. Similar findings are obtained when considered the
Investor sentiment, stock market returns and volatility 227
changes in sentiment index. The results indicate that during the period of investigation, when the change in investor sentiment turns to positive (negative) the market excess return also moves positive (negative) as well. The magnitude of negative changes in sentiment is more dominant than the changes in positive sentiment. The adjusted R-squared values for the regression specifications1, specification2 and specification3 are 22%, 23% and 27%, respectively, indicating that around 27% of the excess returns have been explained by the sentiment index. The explanatory power of the equations has been increased when the sentiment index decomposed to positive and negative sentiment changes. The regression results free from heteroskedasticity and serial correlations.
From the OLS regression results, it can be said that investor irrational sentiment or the impact of noise traders and the excess returns in Indian market are positively related, but the impact of negative sentiment seems higher impact on marker excess return than the positive sentiment. It implies that when investor irrational sentiment is positive the investors are more optimistic about the market, earning more excess return and their speculative motive are stronger, tempting them to invest more. Eventually, they tend to lose when the sentiment goes bearish. In short, when investors are more optimistic or bullish about the market they earn higher excess return and when they are more pessimistic or bearish they earn lower excess return.
Notes: Specifications1, 2 and 3 represent the OLS regression in equations (3), (4) and (5), respectively. AR(1) is the first-order autocorrelation in the residuals. D-W stat is the Durbin–Watson d-statistics.
*** implies statistical significance at 1% level; ** implies statistical significance at 5% level; and * implies statistical significance at 10% level.
We also estimated the sentiment augmented EGARCH-in-mean model in order to investigate if the investor irrational sentiment impacts the market excess return and volatility. Table 6 reports the results of the relationship between the investor sentiment, market excess returns and conditional volatility based on the EGARCH(1, 1)-M model. Again, the analysis has been conducted in two different specifications. In specification1 we consider the aggregate sentiment index described in equation (6), and in specification2 we consider the changes in investor sentiment described in equation (7).
228 P.K. Naik and P. Padhi
The risk-return trade-off parameter δ from the mean equation of Table 6 is found to be positive and statistically significant at 1% level which supports the argument of CAPM. The conventional CAPM states that investor should reward by taking the systematic risk. Our results imply that the noise trader’s risks are systematic. The parameter b1 representing the coefficient for lagged excess returns is positive and highly significant. The coefficient of investor irrational sentiment, c1, is positive and statistically significant at 10% in specification1 and 1% level in specification2 implying that when investor sentiment is bullish investors are optimistic about the market and they earn higher excess return. This result is consistent with findings in Table 5. Similar results have also been documented in the previous studies, such as Lee et al. (2002), Chuang et al. (2010) and Perez-Liston et al. (2014).
The variance equation of specification1 in Table 6 indicates that all the GARCH terms α1 and β1 have the expected sign. The asymmetric parameter γ1 has a negative sign but statistically insignificant. The coefficient of investor irrational sentiment θ1 appears with a negative sign and statistically significant at 10% level. This result indicates that in aggregate the investor sentiment has negative effect on the conditional volatility. The positive and negative shift in investor sentiment has been augmented in specification2. Results show that the asymmetric parameter γ1 has now statistically significant at 1% level with the expected negative sign. This implies that the market exhibits volatility asymmetry.
The positive change in investor sentiment or the bullish shift and the negative change in investor sentiment or the bearish shift has been measured by the coefficient θ1 and θ2. It is evident from Table 6 that, while θ1 appears with a negative sign and significant at 1% level, θ2 appears with a positive sign and significant at 10% level. These results imply that when the change in investor sentiment is positive or bullish, the return volatility is lower; on the other hand, when the change in investor sentiment is negative volatility is higher. This finding is consistent with Lee et al. (2002) and Perez-Liston et al. (2014). However, it contradicts with the findings of Chuang et al. (2010) which reported that bullish shift is insignificant. Our findings confirm that the bullish and bearish shifts in investor sentiment have an asymmetric impact on conditional volatility in Indian equity market.
In line with Lee et al. (2002), Chuang et al. (2010) and Perez-Liston et al. (2014) we add the risk-free rate as an exogenous variable in the conditional volatility equation. The parameter Φ1 represents the coefficient of risk-free interest rate. We find a negative and statistically significant coefficient for Φ1 which is in contradict with Lee et al. (2002) but consistent with Chuang et al. (2010) and Perez-Liston et al. (2014).
Table 6 Investor sentiment, excess returns, and conditional volatility: EGARCH(1, 1)-M model
Parameters Specification1 Specification2
Mean equation
Coeff. z-stat Coeff. z-stat
a 6.510 14.828** 0.521 9.043**
δ 1.107 15.158** 0.090 9.487**
b1 0.577 3.864** 0.738 12.343**
c1 0.166 1.968* 0.041 5.771**
AR(1)
Investor sentiment, stock market returns and volatility 229
Table 6 Investor sentiment, excess returns, and conditional volatility: EGARCH(1, 1)-M model (continued)
Parameters Specification1 Specification2
Variance equation
ω –5.215 –12.153** –3.615 –34.988**
α1 0.010 1.716* 0.0009 0.029
β1 0.116 1.645* 0.342 390.289**
γ1 –0.012 –1.302 –0.382 –22.956**
θ1 –0.130 –1.716* –0.112 –2.656**
θ2 0.063 1.925*
Φ1 –0.038 –4.723**
Log likelihood 218.256 233.037
AIC –2.808 –2.980
SBC –2.627 –2.758
Q(12) 19.09 17.56
Q2(12) 17.47 28.84**
Notes: Specification1 and 2 represent equations (6) and (7), respectively.
*** indicate statistical significance at 1% level and * indicate statistical significance at 10% level.
Since it has also been argued in the literature that the market excess return and the investor sentiment may act as a system, and due to the fact that the OLS and the EGARCH model do not able to control for the problem of endogeneity, we also estimate a bi-variate vector autocorrelation (VAR) model. This will provide us the robustness check. However, in this section we have only consider the aggregate sentiment index and not decomposed into positive and negative changes. The aim here is to test the dynamic interaction between sentiment index and excess market returns, and identify the direction of causality, if any. Since the VAR model is likely to be sensitive towards different lags, we first checked for the optimum lags through the VAR lag selection criteria. Table 7 reports results of VAR lag selection criteria. While the SBC test suggests one lag for the VAR estimation, the HQ test suggests a lag order of three and the AIC test suggests a lag order of five. As it is almost impossible to judge which model would have to be preferred we proceed with all three criteria. Accordingly, we estimate the VAR model of the form of equations (8) and (9) with one, three, and five lags. The results are reported in Table 8.
Table 8 shows that both excess returns and investor sentiment have the powerful predictor of themselves. The VAR estimation results with one lag indicate that investor sentiment has insignificant in influencing the market excess return. However, the excess return has a positive and significant impact on investor sentiment. The lagged level of investor sentiments and market excess return explain about 62% of variation in sentiment. This is evident from the adjusted R-squared value of 0.629. In fact, in all the three specifications (i.e. VAR model with one lag, three lags, and five lags) the lagged levels of investor sentiment and market excess returns explain a substantial variation in sentiment. The adjusted R-squared is more than 60%.
When considered a VAR model with three lags, sentiment index turns significant and negative in influencing the market excess returns at lag three. The excess returns
230 P.K. Naik and P. Padhi
continued to be positive and significant at its first lag but significantly negative in its second lag in influencing the sentiment index. Similar results are found when consider the VAR model with five lags. Excess market returns seem to be influenced by the past sentiment although with relatively long lag. At its third lag sentiment index negatively and significantly influences market excess return. Once again, the excess return is significant and positive in its first lag while significantly negative with its second lag to influence the investor sentiment index. The VAR Granger causality/block exogeneity test suggests that with one lag and five lags, the excess return causes the sentiment index; but with three lags it shows a bidirectional causality between sentiment and excess return. These results imply that when the market performs well it throws signal of a bullish sentiment to the investor, but over optimistic tempt the noise traders to buy (sell) most of the risky stocks and more likely to suffer a capital loss (Friedman’s effect).
The predicted pattern of surprise changes or the innovation has been captured by the Impulse Response Function (IRF) generated from the VAR model. The results of which are plotted in Figure 4. This figure depicts the generalised IRF of estimated VAR (5) model where the response of sentiment index and excess market return to a one time unitary shocks in shocks of sentiment index and excess market return are captured. In other words, it let us know by what percentage a series increases or decreases in response to one unit shock due to some other series in the system. While this is measured in the y-axis of the IRF in Figure 4, the x-axis measures the time path of the response of a series to shock. Results clearly indicate that the response of sentiment index to the shocks excess return is positive and significant (see bottom of column 1 in Figure 4) and the impact continues for a long period. The response of market excess return to shocks of sentiment index is negatively insignificant in the first month; it turns to positive and insignificant in the second month and again turns negative and significant in the third month. The response of excess return to its own shock is significantly positive in the first and third month but insignificant afterwards. On the other hand, the response of sentiment to its own shock is highly significant and positive and also remains for a long time.
The proportion of forecast error of one variable due to other variables has been shown through the variance decomposition. Table 10 presents the results of variance decomposition. It indicates that the share of sentiment index in fluctuation of excess returns is only 6.5%. In fact, the forecast error variance of excess returns is almost accounted by its own shocks (95%). Sentiment index explains 5% impact on excess return. However, excess return explains around 32% on sentiment index, and around 68% of the forecast error variance of sentiment index is accounted by its own shocks.
Table 7 VAR lag order selection criteria
Endogenous variables: ExcessReturn and SentIndex
Lag LogL LR FPE AIC SC HQ
0 –58.194 – 0.008 0.872 0.914 0.889
1 23.485 159.809 0.0026 –0.253 –0.126* –0.201
2 28.678 10.009 0.0026 –0.270 –0.058 –0.184
3 37.464 16.680 0.0024 –0.340 –0.043 –0.219*
4 38.678 2.269 0.0025 –0.299 0.082 –0.144
5 46.138 13.730 0.0024* –0.349* 0.116 –0.160
Investor sentiment, stock market returns and volatility 231
Table 7 VAR lag order selection criteria (continued)
Endogenous variables: ExcessReturn and SentIndex
Lag LogL LR FPE AIC SC HQ
6 47.027 1.611 0.0025 –0.304 0.246 –0.080
7 48.644 2.882 0.0026 –0.270 0.366 –0.011
8 49.955 2.297 0.0027 –0.231 0.489 0.061
9 54.151 7.237 0.0027 –0.234 0.571 0.093
10 54.697 0.924 0.0028 –0.184 0.706 0.178
11 61.139 10.737* 0.0027 –0.219 0.756 0.177
12 64.959 6.256 0.0027 –0.216 0.843 0.214
Note: LR: sequential modified LR test statistic (each test at 5% level); FPE: final prediction error; AIC: Akaike information criterion; SC: Schwarz information criterion; HQ: Hannan–Quinn information criterion.
* indicates lag order selected by the criterion.
Table 8 Results of VAR models
VAR with one lag VAR with three lags VAR with five Lags
Notes: The standard errors and t-statistics are shown in () and [], respectively.
*** represents statistical significance at 1%; ** represents statistical significance at 5%; and * represents statistical significance at 10%, respectively.
In order to get the closer insight of the direction of causality, i.e. whether sentiment cause excess market return or market return cause sentiment we also apply the Granger causality test. If the noise trader explanation is expected then causality must run from investor sentiment to market excess return. However, if asked how the sentiment might be generated then it is reasonable to expect that the market behaviour will generate the bullish trend and bearish trend which ultimately influence the investor sentiment (see Wang et al., 2006). Table 9 reports the results of pair wise Granger causality test starting from lag one to lag five. In all the cases the null hypothesis ‘ExcessReturn does not Granger cause SentIndex’ has been rejected at one and 5% level of significance. Except in lag three, we do not able to reject the null hypothesis that ‘SentIndex does not Granger cause ExcessReturn’ at the usual 5% level. Thus, it can be concluded from this result that market excess return Granger cause investor sentiment and not vice versa is found. This finding is consistent with the previous findings, such as Brown and Cliff (2004) and Wang et al. (2006) who document the empirical support for return causes sentiment and not vice versa.
Investor sentiment, stock market returns and volatility 233
Table 9 Pair wise Granger causality test
Lags Obs. H0: SentIndex does not
Granger cause excess return H0: excess return does not Granger cause SentIndex
F-statistic Prob. F-statistic Prob.
1 149 1.675 0.197 3.943 0.048
2 148 1.287 0.279 7.053 0.001
3 147 2.772 0.043 4.866 0.003
4 146 2.097 0.084 3.334 0.012
5 145 1.638 0.154 3.341 0.007
Table 10 Variance decomposition
Variance decomposition of excess returns:
Period S.E. ExcessReturns SentIndex
1 0.058 100 0.000
2 0.065 99.865 0.134
3 0.065 99.067 0.932
4 0.066 98.936 1.063
5 0.068 96.502 3.497
6 0.068 96.078 3.921
7 0.068 95.530 4.469
8 0.068 95.072 4.927
9 0.068 94.899 5.100
10 0.069 94.588 5.411
Variance decomposition of SentIndex:
Period S.E. ExcessReturns SentIndex
1 0.800 12.023 87.976
2 1.028 25.290 74.709
3 1.138 25.695 74.304
4 1.215 25.706 74.293
5 1.240 27.855 72.144
6 1.267 28.660 71.339
7 1.296 29.719 70.280
8 1.320 30.724 69.275
9 1.351 31.193 68.806
10 1.368 31.931 68.068
234 P.K. Naik and P. Padhi
Figure 4 Impulse response functions (see online version for colours)
-.04
-.02
.00
.02
.04
.06
.08
1 2 3 4 5 6 7 8 9 10
Response of Excess Return to Excess Return
-.04
-.02
.00
.02
.04
.06
.08
1 2 3 4 5 6 7 8 9 10
Response of Excess Return to SentIndex
-0.2
0.0
0.2
0.4
0.6
0.8
1.0
1 2 3 4 5 6 7 8 9 10
Response of SentIndex to Excess Return
-0.2
0.0
0.2
0.4
0.6
0.8
1.0
1 2 3 4 5 6 7 8 9 10
Response of SentIndex to SentIndex
Response to Cholesky One S.D. Innovations ± 2 S.E.
4 Conclusions
The present study investigated whether Indian equity market driven by the irrational investor sentiment or the excess returns impact investor sentiment, by considering the NSE of India. The method of PCA has been employed to construct the investor sentiment index using several market-related implicit sentiment proxies such as trading volume, turnover ratio, ADR, the ratio of put and call option, NIPOs, price earnings ratio, and mutual fund net investment. The irrational component of sentiment has been generated by regressing each of the sentiment indicators on macroeconomic fundamentals such as growth rate of IIP, rate of inflation, exchange rate, risk-free rate of interest, term spread, and net flow of foreign institutional investors. After constructing the sentiment index the study also decomposed it into the positive changes and negative changes in investors’ sentiment to represent the bullish and bearish sentiments, respectively. The analysis has been done using the regression methods of OLS, the vector autoregression (VAR) and EGARCH-M models.
The main findings from the study may be summarised as follows. The results from OLS estimation indicate that sentiment index significantly influences market excess returns. When the sentiment index decomposed into positive and negative sentiment the study finds an asymmetric relationship. It is found that, while the positive sentiment index has a positive impact on market excess return, the negative sentiment index has negative impact. These results imply that when investors are more optimistic about the market they earn more excess return, and their excessive optimism leads them to speculate more which tempt them to invest even more. Subsequently, they tend to lose when the sentiment goes bearish.
Investor sentiment, stock market returns and volatility 235
The results of EGARCH-M model suggest that sentiment index positively influences the market excess returns, whereas it has a negative impact on volatility. When the sentiment index is decomposed into positive sentiment changes and negative sentiment changes, the results indicate that the changes in investor sentiment continue to have a positive impact on contemporaneous returns. However, an asymmetric impact of the positive and negative sentiment changes on the excess return volatility is found, i.e. while the positive sentiment change negatively related to market volatility, the negative sentiment change influences the market volatility positively. Thus, it can be concluded from these results that when the change in investor sentiment takes a bullish shift excess return volatility goes down, and when it takes a bearish shift excess return volatility goes up during the study period.
The results from VAR estimation suggest that excess market returns are negatively influenced by the lagged sentiment although at its third lag. Secondly, we find that the lagged excess return starts with a positive influence on investor sentiment in the first lag, but turns to negatively influence on sentiment in the second lag. Finally, the Granger causality results suggest a bidirectional causality between excess return and investor sentiment at the third lag. However, causality runs from excess return to sentiment at the one and fifth lags. The findings from this study may helpful for the policy makers, retail investors and other decision makers in the Indian stock market. However, the findings are limited with seven market-related implicit factors for investor sentiment. Inclusion of more such factors or using a direct survey data on investor sentiment may extend the study.
References
Baker, M. and Wurgler, J. (2006) ‘Investor sentiment and cross-section of stock returns’, Journal of Finance, Vol. 61, pp.1645–1680.
Baker, M. and Wurgler, J. (2007) ‘Investor sentiment in the stock market’, Journal of Economic Perspectives, Vol. 21, pp.129–151.
Baker, M., Wurgler, J. and Yuan, Y. (2012) ‘Global, local, and contagious investor sentiment’, Journal of Financial Economics, Vol. 104, pp.272–287.
Bennet, E., Selvam, M., Vivek, N. and Shalin, E.E. (2012) ‘The impact of investors’ sentiment on the equity market: evidence from Indian stock market’, African Journal of Business Management, Vol. 6, No. 32, pp.9317–9325.
Black, F. (1986) ‘Noise’, Journal of Finance, Vol. 41, pp.529–543.
Brown, G.W. and Cliff, M.T. (2004) ‘Investor sentiment and the near-term stock market’, Journal of Empirical Finance, Vol. 11, pp.1–27.
Brown, G.W. and Cliff, M.T. (2005) ‘Investor sentiment and asset valuation’, Journal of Business, Vol. 78, pp.405–440.
Chandra, A. and Thenmozhi, M. (2013) Investor Sentiment, Volatility and Stock Return Comovements. Available online at: http://ssrn.com/abstract=2353073
Changsheng, H. and Yongfeng, W. (2012) ‘Investor sentiment and assets valuation’, Systems Engineering Procedia, Vol. 3, pp.166–171.
Chi, L., Zhuang, X. and Song, D. (2012) ‘Investor sentiment in the Chinese stock market: an empirical analysis’, Applied Economic Letters, Vol. 19, pp.345–348.
Chuang, W-J., Ouyang, L-Y. and Lo, W-C. (2010) ‘The impact of investor sentiment on excess returns: a Taiwan market cases’, International Journal of Information and Management Sciences, Vol. 21, pp.13–28.
236 P.K. Naik and P. Padhi
Dash, S.R. and Mahakud, J. (2012) ‘Impact of investor sentiment on stock return: evidence from India’, Journal of Management Research, Vol. 13, No. 3, pp.131–140.
Dash, S.R. and Mahakud, J. (2013a) ‘Investor sentiment and stock returns: do industries matter?’, Margin: The Journal of Applied Economic Research, Vol. 7, pp.315–349.
Dash, S.R. and Mahakud, J. (2013b) ‘Investor sentiment, risk factors, and stock returns: evidence from Indian non-financial companies’, Journal of Indian Business Research, Vol. 4, pp.194–218.
De Long, J., Shleifer, A., Summers, L. and Waldmann, R. (1990) ‘Noise trader risk in financial markets’, Journal of Political Economy, Vol. 10, pp.407–432.
Dergiades, T. (2012) ‘Do investors’ sentiment dynamics affect stock returns? Evidence from the US economy’, Economics Letters, Vol. 116, pp.404–407.
Fama, E. (1965) ‘The behavior of stock market prices’, Journal of Business, Vol. 38, pp.34–105.
Fisher, K.L. and Statman, M. (2000) ‘Investor sentiment and stock returns’, Financial Analysts Journal, Vol. 56, pp.16–23.
Huang, C., Yang, X., Yang, X. and Sheng, H. (2014) ‘An empirical study of the effect of investor sentiment on returns of different industries’, Mathematical Problems in Engineering, doi:org/10.1155/2014/545723.
Jackson, A. (2003) The Aggregate Behaviour of Individual Investors. Available online at: http://ssrn.com/abstract=536942
Kumar, A. and Lee, C.M.C. (2006) ‘Retail investor sentiment and return comovements’, Journal of Finance, Vol. 51, pp.2451–2486.
Kurov, A. (2010) ‘Investor sentiment and the stock market’s reaction to monetary policy’, Journal of Banking & Finance, Vol. 34, pp.139–149.
Lee, W.Y., Jiang, C.X. and Indro, D.C. (2002) ‘Stock market volatility, excess returns, and the role of investor sentiment’, Journal of Banking and Finance, Vol. 26, pp.2277–2299.
Lee, C.M.C., Shleifer, A. and Thaler, R.H. (1991) ‘Investor sentiment and the closed-end fund puzzle’, The Journal of Finance, Vol. 46, No. 1, pp.75–109.
Li, B.H. (2014) ‘Does investor sentiment predict stock returns? The evidence from Chinese stock market’, Journal of System Science Complex, Vol. 27, pp.130–143.
Lux, T. (2011) ‘Sentiment dynamics and stock returns: the case of the German stock market’, Empirical Economics, Vol. 41, pp.663–679.
Neal, R. and Wheatley, S. (1998) ‘Do measures of investor sentiment predict stock returns?’, Journal of Financial and Quantitative Analysis, Vol. 34, pp.523–547.
Perez-Liston, D., Huerta, D. and Haq, S. (2014) ‘Does investor sentiment impact the returns and volatility of Islamic equities?’, Journal of Economics and Finance, doi:10.1007/s12197-014-9290-6.
Qiang, Z. and Shue-e, Y. (2009) ‘Noise trading, investor sentiment volatility, and stock returns’, System Engineering: Theory and Practice, Vol. 29, No. 3, pp.40–47.
Rehman, M.U. (2013) ‘Investor’s sentiments and stock market volatility: an empirical evidence from emerging stock market’, Pakistan Journal of Commerce and Social Sciences, Vol. 7, No. 1, pp.80–90.
Schmeling, M. (2009) ‘Investor sentiment and stock returns: some international evidences’, Journal of Empirical Finance, Vol. 16, pp.394–408.
Sehgal, S., Sood, G.S. and Rajput, N. (2009) ‘Investor sentiment in India: a survey’, Vision: The Journal of Business Perspective, Vol. 13, pp.13–23.
Shefrin, H. and Statman, M. (1994) ‘Behavioral capital asset pricing theory’, Journal of Financial and Quantitative Analysis, Vol. 29, No. 3, pp.323–349.
Shiller, R.J. (1981) ‘Do stock prices move too much to be justified by subsequent changes in dividends?’, American Economic Review, Vol. 71, No. 3, pp.421–436.
Investor sentiment, stock market returns and volatility 237
Shiller, R.J. (2003) ‘From efficient markets theory to behavioral finance’, Journal of Economic Perspective, Vol. 17, pp.83–104.
Shleifer, A. and Summers, L.H. (1990) ‘The noise trader approach to finance’, Journal of Economic Perspectives, Vol. 4, pp.19–33.
Swaminathan, B. (1996) ‘Time-varying expected small firm returns and closed-end fund discounts’, Review of Financial Studies, Vol. 9, No. 3, pp.845–888.
Verma, R. and Soydemir, G. (2009) ‘The impact of individual and institutional investor sentiment on the market price of risk’, The Quarterly Review of Economics and Finance, Vol. 49, pp.1129–1145.
Verma, R. and Verma, P. (2007) ‘Noise trading and stock market volatility’, Journal of Multinational Financial Management, Vol. 17, pp.231–243.
Wang, Y-H., Keswani, A. and Taylor, S.J. (2006) ‘The relationships between sentiment, returns and volatility’, International Journal of Forecasting, Vol. 22, pp.109–123.
Yang, Y. and Copeland, L. (2014) The Effects of Sentiment on Market Return and Volatility and the Cross-Sectional Risk Premium of Sentiment-Affected Volatility, Cardiff Economics Working Papers, No. E2014/12, Cardiff Business School, Cardiff.
Yu, J. and Yuan, Y. (2011) ‘Investor sentiment and the mean-variance relation’, Journal of Financial Economics, Vol. 100, pp.367–381.
Zhu, X. (2012) ‘Investor sentiment and volatility of stock index an empirical analysis from the perspective of behavioral finance’, Advances in Applied Economics and Finance, Vol. 3, No. 4, pp.627–629.
Notes
1 Investor sentiment represents the expectation of market participants based on the market behaviour. For example, a bullish investor expects the returns to be above average and a bearish investor expects it to be below average (Brown and Cliff, 2004). Baker and Wurgler (2006) define it as a belief about future cash flows and investment risks that is not justified by the facts at hand. Overall, investor sentiment is the propensity of the investors to believe the future trend of the market.
2 Irrational investors are met in the market by rational arbitrageurs who trade against them and in the process drive prices close to fundamental values. In the course of such trading those whose judgements of asset values are sufficiently mistaken to affect prices lose money to arbitrageurs and so eventually disappear from the market (De Long et al., 1990).
3 We consider 91 days T. bills rate as risk-free interest rate and the term spread has been computed as the difference between 364 days T. bills rate and 91 days T. bills rate.
4 These days NSE has gained utmost popularity among the trading members contributing 83% of total turnover in India as on 2012–2013. S&P CNX Nifty, as a benchmark index of the Indian equity markets, consists of 50 major stocks of Indian companies and covering 22 sectors of the Indian economy. It represents about 67% of the free float market capitalisation of the stocks listed in NSE of India.
5 Similar approach for positive and negative changes in sentiment has employed in the previous studies such as Lee et al. (2002), Qiang and Shu-e (2009), Chuang et al. (2010) and Yang and Copeland (2014).
