國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
3
1. Introduction
Savor and Wilson (2012) document that the returns are extremely different between the days macroeconomic news scheduled for release (the following we use term “announcement-day” or “announcement days” ) and other trading days (the following we use term “non-announcement-day” or “non-announcement days”). Their study reveals that during period from 1963 to 2011 announcement-day earns 11 bps average daily excess returns whereas non-announcement-day earns only 1.1bps
average daily excess returns, and that the sharp ratio in announcement-day is ten times to the ratio in non-announcement-day.
Stocks returns should perform poorly when the news suggest that the market will turn to be bear, making the announcement days is risky than other days. For example, recent studies document that interest rate, unemployment rate, and inflation rate strongly predict stock returns (Bernanke & Kuttner 2005; Ioannidis & Kontonikas 2008; Bjørnland & Leitemo 2009; Birz & Lott 2011). When those indicators show good (bad) news, the market turns to be bull (bear). Also, when the indicators shows weak (strong) signal about fundamental economics, investors tend to require higher (lower) compensation of systematic risk. While economic announcements presumably release important information about economy, the risk of such announcement days should be higher than that of other days so that higher returns on announcement days should be compensation for investors who bear macroeconomic risks.
While several studies (Black 1972; Fama & French 1992; Black 1995; Polk et al. 2006) suggest no direct relation between the beta and the excess returns in most
stocks, Savor and Wilson (2013), following intuition of macroeconomic risk on announcement days, explore the risk-returns trade off relation on such days. That is, they employ conditional variance to forecast the next period returns, showing that in announcement days beta successfully predicts the excess returns and that excess
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
4
returns holds traditional asset pricing theories. But as SW(2013) present, risk factor is possibly not the only factor to explain the difference excess returns between
announcement-day and non-announcement-day because there exist much difference between these two types of days. For example, both individual investors and
institutional investors extremely focus on the announcement of macroeconomics news, and the composition of those investors will be different between announcement days and other days (Pastor & Veronesi 2012). Once macroeconomic news announced monthly, analysts will write the analysis, comments, and expectation about the future, and then these contexts distribute by TV, magazine and several social media to lots of investors, including professional investor and noise traders (Kumar & Lee 2006). That is, unlike in other days, in announcement days not only informative investor but also noise traders buy and sell in the market to hold their own portfolio (Black 1986).
Tetlock (2007) finds that the future market is less informative when Wall Street Journal releases news, this fact account for an increase in noise traders in that time.
Podolski–Boczar et al. (2009) reveal that noise traders will affect not the returns but the volatility. Those evidences suggest that the composition of investors is different between announcement days and non-announcement days. The different level of noise trader, who affected by investor sentiment, should account for excess returns
difference between two days.
Follow this intuition, we propose two hypotheses. First, in the announcement days both sentiment driven noise traders and rational investors participate in the stock market, whereas less noise traders participate in non-announcement days. The
different level of sentiment driven noise traders leads to different influence on excess returns (De Long et al. 1990; Shleifer & Summers 1990). We anticipate that investor sentiment will predict difference excess returns between two days because these noise traders are strongly affected by investor sentiment. Also, Brown and Cliff (2004)
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
5
reveal that changes in investor sentiment predicts future returns. Shleifer (2012) and Figlewski (1981) documents that investors usually follow “System 1” thinking, which means they decide to buy or to sell simply depends on first expression, to decide their strategy. This action links to the nearest feeling, changes in sentiment (Lee et al.
2002). That is, returns of announcement days will positively relate to changes in sentiment. On the other hand, the noise traders were also affected by status, weak or strong, of the market (Stambaugh et al. 2012). We define high and sentiment periods that depend on using sentiment index. During high sentiment period, noise traders will much strongly mispricing than in low sentiment. We expect that in high sentiment investor sentiment positively related to returns of announcement days whereas it does not exist in low sentiment.
In our research, we first present summary statistics of each announcement days and non-announcement days. We use both equal-weighted portfolio and value-weighted portfolios. During 1966 to 2010 period, the monthly average excess daily returns of value-weighted portfolio is 1.46% in announcement days versus only 0.28% in non-announcement days. The difference excess returns is significant with t-statistics 2.10 of value-weighted portfolio. Then we regress the different excess
returns on the changes in investor sentiment index, reporting that investor sentiment is a significant factor both in controlling and non-controlling equations with at least t-statistics 2.19. Also, we test different proxies of investor sentiment, such as VIX and closed-end fund (Lee et al. 1991) discount, revealing strong significance as well as above. Besides, we use time series regression to test predictability of investor sentiment on long-short strategy, a strategy that long announcement day and short non-announcement day, and we show that investor sentiment successfully predicts the returns of strategy in time t+1. These results suggest investor sentiment is positively correlated to difference excess returns between announcement days and
non-‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
6
announcement days, and it also predict future returns.
We then examine the second hypothesis: investor sentiment works during high sentiment period rather than does during low sentiment. We classified high- and low- investor sentiment periods based on the median level of the index of Baker and Wurgler (2006), and we regress equations during each high- and low- sentiment periods. Our reports show that each of t-statistics of sentiment estimate is significant during high investor sentiment, and that none of t-statistics of estimate of sentiment is significant during low investor sentiment. To show that our result is strongly robust, we use various portfolios, which include portfolios formed by industry, size, book-to-market(Fama & French 1992), momentum, to test, suggesting the results still hold on most of such different portfolio in high sentiment, especially in small firms (Lemmon
& Portniaguina 2006). These results suggest investor sentiment is positively correlated to difference excess returns between announcement days and
non-announcement days during high sentiment period rather than in low sentiment period.
“A tale of two days” leads different patterns, the patterns account not only for the risk story that SW proposed but, significantly, for the sentiment driven factors such as noise trader and investor sentiment. As Lakonishok et al. (1994) suggests, we
consider the factors other than risk as explanation when we find that risk-returns trade off relation does not always hold. We adapt the methodology that proposed in Savor and Wilson (2012). In our methodology, we regress –by each high- and low-
sentiment – announcement-day aggregate excess returns in quarter t+1 on expected variance, which is in quarter t, and on investor sentiment, which is also in quarter t.
Our report suggests that the expected variance predicts the returns only during low sentiment period, and instead investor sentiment predicts the returns during high sentiment. This fact implies investor sentiment plays an important role in the
difference excess returns between announcement days and non-announcement days.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
7
In this papers, we provide investor sentiment explanation toward a tale of two days puzzle. We show that investor sentiment empirically accounts for difference excess returns between announcement days and predicts returns on long-short strategy portfolio, and that different excess returns interact with investor sentiment index during high sentiment period. Finally, we reconcile the relation between risk factors and sentiment factors.
The rest of the paper is organized as follows. Section 2 describes data and introduces methodology. Section 3 reports the main empirical results and Section 4 gives the result robust test. Section 5 concludes.