ANEXO VII Protocolo de intervención en el grupo experimental

This study intends to test whether investor sentiment is priced and has a role in the risk-return trade off. It should be made clear that there is no intension to evaluate the mean- variance model from the perspectives of different volatility models, mean-variance model specifications, sampling issues and linearity assumptions of the mean-variance trade-off as discussed in section 2.6.5. Specifically, I argue that noise traders dominate the market during high sentiment periods and introduce noise trader risk that should be priced. In addition, high investor sentiment attenuates the otherwise positive mean-variance relationship, or reverses it to become negative as explain by prospect theory. This study fundamentally incorporates

the role of sentiment and provides a basis for the discussion of mixed mean-variance relationships.

I first assume that investors are prone to cognitive biases instead of being fully rational. These biases may arise throughout the process of forming beliefs and preferences. People are subject to cognitive biases, for example, overconfidence, optimism, representativeness, conservatism, belief perseverance, anchoring, availability biases, mental accounting, and the framing effect (Barberis & Thaler, 2003; Ritter, 2003).

Second, the cognitive biases lead to a bias expectation for risk and return among sentiment investors. Campbell and Kyle (1993) suggest that excessive volatility of stock prices do not attribute to fundamentals; IPOs (Ritter, 1991) and close-end funds are sold at discount (Chopra et al., 1993; Lee et al., 1991).

Third, there are more noise traders in the market who trade actively when the sentiment is relatively high or low. DeLong et al.(1990b) suggest that sophisticated investors turn passive when sentiment traders dominate the market. Prices become volatile due to lack of arbitrage activities against the noise traders.

The above assumptions imply that there is a role for investor sentiment in asset pricing and mean-variance relationships. During the period of high sentiment, sentiment investors overreact to trading-induced good news and bid the price even higher. They buy at a high price and thus lower the required return. Regardless of the impact of sentiment on risk, sentiment investors are simply willing to be compensated with lower returns. Hence, this study attempts to test the following hypotheses:

Hypothesis 8

H0: There is no mean-variance relationship

H1a: There is positive mean-variance relationship when create space effect outweigh

the Friedman effect.

H1b: There is negative mean-variance relationship when Friedman effect outweigh the

create space effect.

The Merton (1973) model suggests that investors demand higher returns for investing in riskier assets. A positive mean-variance relationship implies a risk-return trade off. However, there is mixed evidence for mean-variance relationships. DeLong et al. (1990b) attribute this to noise traders misperceiving the risks of holding the risky assets. A greater shift in sentiment is associated with greater future returns volatility (higher risk) and lower expected returns. On one hand, noise traders have poor market timing, follow the footsteps of other noise traders, and end up buyinghigh and-sellinglow, eventually earning poor returns. This is named as ‘Friedman’ effect; noise traders’ returns are negatively related to the variability of their beliefs. On the other hand, risk-averse arbitrageur will avoid betting on noise traders’ mispricing when there is high variability in noise trader beliefs. This is the so- called ‘create space’ effect. Volatility is positively related to expected returns. In conclusion, the net impact of volatility on excess returns is positive when the create space effect outweighs the Friedman effect and vice versa. I test Hypothesis 8 using the coefficient, 𝛽𝛽1, in

Equation 14a and 14b.

Hypothesis 9

H0: Sentiment does not predict excess returns

H1a: High sentiment positively (negatively) predicts the returns when hold-more effect

(price pressure) dominate the price pressure (hold-more) effect. 72

H1b: Low sentiment negatively predicts the excess returns because price pressure and

hold-more effect are negative.

DeLong et al. (1990b) explain the impact of noise traders on asset pricing. The direct short-term impact arises from ‘price pressure’ and the ‘hold more’ effect. Bullish noise traders demand more stocks. Greater demand drives the stock prices to higher levels and expected returns are lower. Consequently, expected returns are negatively related to investor sentiment. This is called price pressure effect. At the same time, bullish irrational investors hold more risky assets than rational arbitrageurs during the period of high sentiment, higher expected returns is a form of reward for noise traders bearing the risk. I expect a positive relationship between sentiment and returns. This is called hold more effect. These two effects interact to determine the impact of sentiment on returns. If the hold-more effect dominates the price pressure effect, one would expect bullish sentiment to lead to higher returns. When the price pressure effect outweighs (weaker than) the hold-more effect, bullish sentiment predicts lower (higher) returns. The impact of the price pressure effect and hold more effect is always negative when investors are bearish. If the 𝜶𝜶𝟐𝟐in the Equation 14a and 14bis significant,

the required return for each unit of risk increases or decreases at a fixed amount at all levels of risk, depending on the magnitude of price pressure and hold more effect.

Hypothesis 10

H0: Sentiment does not change the sensitivity of return to risk.

H1a: Investors are less responsive to risk in the high sentiment regime.

H1b: Investors are more responsive to risk in the low sentiment regime.

Alternatively, investor sentiment can interact with risk and change the sensitivity of return to risk as suggested by Yu and Yuan (2011). This study proposes that investors are less responsive to risk during the high sentiment period. High sentiment following market run-up

weakens the risk-return relationship. I extend this thought to the low sentiment period. Sentiment investors overreact to bad trading-induced news and take the risk cautiously. Low news sentiment interacts with risk and makes the risk-return slope steeper. The investors become more sensitive to risk during the low sentiment period. I expect 𝛽𝛽2 in the Equation

14a and 14b to be negative in a high sentiment regime and to be positive in a low sentiment regime.

I test the Hypothesis 8,9, and 10 based on the following equations in line with Yu and Yuan (2011):

𝑅𝑅𝑡𝑡+1− 𝑅𝑅𝐹𝐹𝑡𝑡+1=𝛼𝛼1+𝛽𝛽1𝛾𝛾𝑉𝑉𝛽𝛽𝑡𝑡(𝑅𝑅𝑡𝑡+1) +𝛼𝛼2𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑡𝑡+𝛽𝛽2𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑡𝑡𝛾𝛾𝑉𝑉𝛽𝛽𝑡𝑡(𝑅𝑅𝑡𝑡+1) +𝜀𝜀𝑡𝑡+1

(3.14a)

𝑅𝑅𝑡𝑡+1− 𝑅𝑅𝐹𝐹𝑡𝑡+1=𝛼𝛼1+𝛽𝛽1𝛾𝛾𝑡𝑡+𝛼𝛼2𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑡𝑡+𝛽𝛽2𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑡𝑡𝛾𝛾𝑡𝑡+𝜀𝜀𝑡𝑡+1 (3.14b)

where the dependent variable is the daily excess index futures returns of the HSIF, KLCIF and SiMSCIF.

Excess returns are defined as the daily returns less the risk free rate. 𝑅𝑅𝑡𝑡is the daily

return of index futures returns. 𝑅𝑅𝐹𝐹𝑡𝑡 is the risk-free rate defined as the three-month treasury

bill discount rate i for the HSIF and SiMSCIF and the one-month Kuala Lumpur interbank offer rate for the KLCIF. 𝛾𝛾𝑉𝑉𝛽𝛽𝑡𝑡 is the conditional variance while 𝛾𝛾𝑡𝑡 is the realised variance.

Sentiment is a dummy variable that refers to sentiment measures namely Bad, Newlow, Good, Newhigh. Bad = 1 if the daily routine news reports the market fell on the prior day; Newlow=1 if the market dipped to a new low; Good=1 if the market rose; Newhigh=1 if the market hit a new high; otherwise, 0. These sentiment measures apply to the HSIF, KLCIF and SiMSCIF with two exceptions. For the case of the HSIF, Lowbench is used instead of Newlow while Highbench is used instead of Newhigh. Lowbench =1 if the HSI fell to a lower benchmark and

Highbench=1 if HSI rose and hit a higher benchmark. Section 3.3.5 explains how to derive sentiment regime from daily news.