Business

Fear, Greed And Efficient Market – Evidence From News Sentiment Analytics

Fear, Greed And Efficient Market – Evidence From News Sentiment Analytics

Tongli Zhang

Johns Hopkins University

May 6, 2015

Abstract:

I analyze the relationship between natural gas prices and the news sentiment data obtained from the Thomson Reuters News Analytics system. I conduct studies on two different horizons: the daily basis study and the intraday basis study. For the daily study, our results show that any correlation we observed between the daily news sentiments and the daily returns of natural gas can be largely attributed to the news relevant to historical natural gas price movements, or “price-related news”. This discovery also emphasizes the failure of price-related news sentiment in predicting future natural gas returns, which in turn supports the weak form of the Efficient Market Hypothesis. For the intraday study, I label news items with high positive and negative sentiment scores as extreme positive news and extreme negative news. I then conduct event study to analyze the price-related extreme new items’ impact on natural gas prices. I found natural gas has two different types of price movement in reaction to price-related extreme negative news. I also show that these two types of abnormal return can serve as a signal for the investors’ sentiment. I made a connection between these investors’ sentiment signals and the S&P 500 index to show that these signals can be used to predict long term S&P 500 returns. The capability of these signals to predict the future S&P 500 index and the fact that the signals are derived solely from past price reactions and past price-related news items suggests that the current natural gas prices and equity market prices do not fully reflect all the information related to historical prices.

Fear, Greed And Efficient Market – Evidence From News Sentiment Analytics – Introduction

Ever since the Efficient Market Hypothesis (EMH) was formally introduced by Fama in 1970 (Fama, 1970), the debate about the market efficiency has continued for several decades. Fama divided the EMH into three forms: weak, semi-strong, and strong, based on the different information set being considered. Many theoretical and empirical evidences are introduced to examine whether the market is efficient or to which extent the market is efficient (Cont, 2001) (Malkiel, 2003).

Historically, most of the literature focused on the efficiency of the stock market (Sewell, 2011). Testing the efficiency of commodity futures market became a popular topic in financial journals in recent years. Taylor (1995) first argued that in its simplest form, the Efficient Market Hypothesis can be reduced to a joint hypothesis that agents, in an aggregated sense endowed with rational expectation and are risk neutral, so that the futures price is an unbiased estimator of the future spot price. Kellard et al. (1999) compared the inefficiency of different futures markets by estimate accuracy of future price on future spot price. Kristoufek & Vosvrda (2013) analyzed the market efficiency of stock indices at 41 different geographic locations by introducing Efficient Index as measurement of the efficiency. They also utilized this Efficiency Index method to study the market efficiency of 25 commodity futures from five groups: energy, metals, soft commodities, grains, and other agriculture commodity. Their results showed that energy is the most efficient market among five groups while the natural gas future (which is the topic of our paper) is the 7th most efficient market among all 25 commodities (Kristoufek & Vosvrda, 2014). Kim et al. (2011) studied the dependence structure between stock market and commodity market. The result showed that generally commodity market and stock market should be seen as completely two separate markets except for crude oil and gold. However, their research results did not include natural gas futures, which is a reason motivated us to examining natural gas futures price reactions to news sentiments in order to examine the efficiency of natural gas futures market and its relationship with stock market.

Much of the extant literature relating EMH use historical prices or data from financial statements as the information set while testing the market efficiency. Such information set focuses on the quantitative information of the news while ignoring the information conveyed by languages of the news articles. Tetlock (2007) started to try to capture the non quantitative information of news by assigning numerical scores to the sentiment of the news articles. Tetlock (2008) find out that the proportion of negative words in a news article can be used to predict earnings. Devitt & Ahmad (2007) and Engelberg et al. (2012) also find that the text of news have a profound impact on the market. In recent years, research interest on the relationship between language sentiment and the market has extended to social media such as blog and Tweets. Bollen et al. (2009), Forbergskog and Blom (2013), and Dennis, and Yuan (2014) studied the stock prices and Tweet sentiment. Zhang & Skiena (2010) build trading strategies based on sentiment of the text on blogs. The discovery of the impact of news on the market and the advance of technology motivate the development of the news analytical systems such as the Thomson Reuters News Analytics (TRNA) system. TRNA system utilizes programmed algorithms to read news text and assign sentiment scores to newswire releases (Smales, 2014). The development of such system allows researchers to process massive amount of news articles to study the impact of news on the market. Feuerriegel et al. investigated the impact of number of terrorist attack on the crude oil prices by using data from the TRNA system. Smales (2015)& (2014) studied the impact of news sentiment on the gold market on a daily basis. He found that news releases with negative news sentiment would invoke contemporaneous response in return of gold futures.

Borovkova (2011) conducted the event study on the natural gas future prices around time when the news with extreme sentiment being released. The result showed that the movement of the natural gas prices happened before the news being released instead of after the news being released. This result and the contemporaneous relationship observed by Smales (2014) inspired our paper. We suspect that the contemporaneous relationship we often observed between news sentiment and commodity prices are actually due to the fact that news articles are about past price movements. That can also explain why the price movements happened before the release of the news. Since Borovkova and Smales did not study the contemporaneous relationship between natural gas future prices and news sentiment, we will first investigate that if the contemporaneous relationship observed on gold market also exist on natural gas market. Our first two hypotheses are:

H1: There is no significant correlation between all three average sentiment scores (positive, neutral and negative) with the return of natural gas future on daily basis

H2: The source of the observed correlation between daily average news sentiment and daily return of natural gas is the news that talking about past prices

Fear, Greed And Efficient Market - Evidence From News Sentiment Analytics

Our analysis was conducted separately in two different settings, a daily basis and an intraday basis. For the daily basis study, we first aggregate the sentiment scores of all news released on a certain day into daily average news sentiments. Then we tested the correlation between these daily average news sentiments with the daily returns of natural gas futures on the same day, the daily return of natural gas futures on the previous day and the daily return of natural gas futures on the next day. Throughout this paper, for convenience, for a given current day we will refer to the current day as “today”, the previous day as “yesterday” and the next day as “tomorrow”. A simple example of our daily analysis method is presented below in Figure 1.

Fear, Greed And Efficient Market - Evidence From News Sentiment Analytics

See full PDF below.