Sentiment Indicators

Can computer software extract exploitable sentiments about individual stocks as conveyed by news articles? In their June 2016 paper entitled “News Versus Sentiment: Predicting Stock Returns from News Stories”, Steven Heston and Nitish Sinha test whether firm news sentiment as interpreted by Thomson Reuters NewScope reliably predicts stock returns. Input data include article publication time, firm mentioned, headline, relevance to the firm, staleness and sentiment as generated a trained neural network. They exclude articles that: are duplicates; mention firms that do not match ticker symbols; and, have firm relevance scores below 35%. They train the neural network with 3,000 randomly selected articles from December 2004 to January 2006. They specify firm net sentiment as average positive sentiment minus average negative sentiment during the measurement interval (one day or one week). They assess predictive power of net sentiment via a hedge portfolio that is long (short) the equally weighted returns of the fifth, or quintile, of stocks with the highest (lowest) net daily or weekly sentiment. They also run a regression that controls for neutral news to isolate the effects of positive and negative news. Using firm sentiment outputs from the Thomson-Reuters news analytics engine for 900,754 articles published during 2003 through 2010, and associated daily stock returns, they find that: Keep Reading

Can systematic, contrarian sports betting usefully diversify conventional investments? In their June 2016 paper entitled “Sports Betting As a New Asset Class: Can a Sports Trader Beat Hedge Fund Managers from 2010-2016?”, Lovjit Thukral and Pedro Vergel investigate whether a specific sports betting strategy outperforms and diversifies the Credit Suisse Hedge Fund Index and the S&P 500 Index. The strategy hypothesizes that horse racing favorites are consistently overrated by betting 1% of a hypothetical portfolio against the top four (lowest odds) horses in each regulated race in the UK. Using historical data from Betfair Exchange for 57,000 horse races and contemporaneous annual returns for the Credit Suisse Hedge Fund Index and the S&P 500 Total Return Index during January 2010 through early January 2016, they find that: Keep Reading

Do volume spikes in specific equity options signal abnormal returns for underlying stocks? In his June 2016 paper entitled “Investor Attention Strategy”, Xuewu Wang examines the motivation, construction and profitability of a strategy that selects stocks based on sudden attention to associated options. He defines sudden attention as volume spikes of at least 10 contracts after a week of no trades for equity options with minimum bid $0.10 and maturity no longer than 120 days. He further discriminates among option spikes by ranking them into thirds (terciles) based on either: (1) ratio of call volume to total volume; or, (2) call implied volatility minus put implied volatility (matched by strike price and maturity and averaged across all pairs for a given stock). He forms stock portfolios by buying stocks immediately after volume spikes of associated options and holding them for 30 calendar days. Using daily returns and associated option volumes for all U.S. stocks having options during January 1996 through  December 2013, he finds that: Keep Reading

Do data-intensive, high-frequency investor sentiment measurements usefully predict stock index performance? In his May 2016 paper entitled “Can Sentiment Indicators Signal Market Reversals?”, Arnaud Lagarde applies a random forest machine learning algorithm to test the power of Amareos sentiment indications to predict stock index reversals. Algorithm training data relates sentiment to known stock index return for the next 182 days (six months). If this return is -20% or lower (+10% or higher), he designates the condition at the time of forecast as a market top (bottom). Otherwise, he designates the condition as neutral. He starts with 20 global equity indexes. He holds out four indexes (CAC40, CSI300, Nikkei and S&P500) for out-of-sample testing. He then randomly selects 80% of daily observations on the other 16 indexes for algorithm training, with the remaining 20% reserved for additional out-of-sample testing. Out-of-sample testing includes tabulation of raw top/bottom identification accuracy and a simple trading strategy that is long (in cash) after a bottom (top) indication and does not react to a neutral indications. He focuses trading strategy testing on: (1) the four hold-out indexes over the entire sample period; and, (2) the last six weeks of data for all indexes, which cannot be used for training. Using daily Amareos market sentiment readings and returns for the 20 equity indexes during January 2005 through mid-April 2016, he finds that: Keep Reading

Are prominent stock market bloggers in aggregate able to predict the market’s direction? The Ticker Sense Blogger Sentiment Poll “is a survey of the web’s most prominent investment bloggers, asking ‘What is your outlook on the U.S. stock market for the next 30 days?'” (bullish, bearish or neutral) on a weekly basis. The site currently lists 29 participating bloggers. Participation has varied over time. Because Ticker Sense collects data weekly, we look at weekly measurements and changes in weekly measurements. Because the poll question asks for a 30-day outlook, we test the forecasts against stock market behavior four weeks into the future. Because polling generally takes place Thursday-Sunday, we use coincident Friday stock market close to represent the state of the stock market for each poll. We use [% Bullish] minus [% Bearish] as the net sentiment statistic for each poll. Using poll results from inception in July 2006 through May 2016 (496 polls) and contemporaneous weekly closes of the S&P 500 Index as representative of the broad stock market, we find that: Keep Reading

Do peaks in the S&P 500 Implied Volatility Index (VIX) signal positive abnormal U.S. stock market returns? If so, can investors exploit these returns? In the May 2016 version of his paper entitled “Abnormal Stock Market Returns Around Peaks in VIX: The Evidence of Investor Overreaction?”, Valeriy Zakamulin analyzes U.S. stock market returns around VIX peaks. He employs two formal methods to detect peaks:

1. When a local maximum (minimum) is at least 20% higher (30% lower) than the last local minimum (maximum), it is a peak (trough).
2. First, identify all local maximums (peaks) and minimums (troughs) within 8-day windows. Then winnow peaks and troughs and systematize alternation by: excluding peaks and troughs in the first and last 20 days; eliminating cycles (peak-to-peak or trough-to-trough) shorter than 22 days; and, excluding phases (trough-to-peak or peak-to-trough) shorter than 10 days, unless daily percentage change exceeds 30%.

He then tests for abnormal stock market returns around VIX peaks and during preceding and following intervals of rising and falling VIX. Abnormal means relative to the average market return for the sample period. Finally, he investigates whether abnormal returns around peaks are due to investor overreaction. Using daily closes for VIX and daily returns of the broad capitalization-weighted U.S. stock market during January 1990 through December 2015, he finds that: Keep Reading

Is there a best measure of investor sentiment for predicting stock market returns? In his March 2016 paper entitled “Investor Sentiment and Stock Market Returns”, Lee Smales updates relationships between stock market/portfolio returns and five sentiment measures:

1. CBOE Implied Volatility Index (VIX).
2. Baker-Wurgler composite sentiment index (readily available only through 2012).
3. American Association of Individual Investors (AAII) investor sentiment.
4. University of Michigan Consumer Sentiment Index.
5. Commitment of Traders (COT) reports from the Commodity Futures Trading Commission.

He controls for multiple economic and financial variables likely to be related to stock market returns (gross domestic product, industrial production, unemployment rate, consumer price index, Federal Funds target rate, term spread, credit spread and dividend yield). He also investigates economic recessions separately. Principal tests relate sentiment levels and changes in sentiment levels to S&P 500 Index and style/industry portfolio returns (from Kenneth French’s data library) at horizons of 1, 3, 6 and 12 months. Using monthly values of sentiment measures as available and monthly index/portfolio returns during January 1990 through December 2015, he finds that: Keep Reading

A subscriber suggested evaluating Sentdex Sentiment Trading Signals. These signals attempt to derive the emotion of a current body of text (over 20 sources, mainly Reuters, Bloomberg, WSJ, LA Times, CNBC, Forbes, Business Insider and Yahoo Finance) regarding financial assets such as individual stocks and stock indexes. Signal values are 24-hour averages, ranging from -3 (strongest negative) to +6 (strongest positive), available daily (if there is any relevant news) 30 minutes before market open. The offeror’s backtest buys when sentiment is +6 and sells when it turns negative, with a -0.5% stop-loss. To evaluate, we extract from offered sample data Sentdex sentiment series for Apple (AAPL) and Bank of America (BAC). We apply the backtest rules, except the stop-loss, to these series using daily opening prices for these stocks, adjusted for dividends and splits. We do not use the stop-loss rule because: (1) it may obscure sentiment informativeness; and, (2) research on stop-losses is at best equivocal on their effectiveness. When considering strategy frictions, we use a 0.1% stock-cash switching fee and a $10 monthly data fee. We ignore return on cash, which is practically zero over the sample periods. We use buy-and-hold as a benchmark. Using the specified Sentdex sentiment series and contemporaneous daily adjusted opening prices for AAPL (mid-October 2012 through mid-June 2015) and BAC (mid-November 2012 through mid-June 2015), we find that: Keep Reading

Are moody investors prone to avoid risk on Monday and accept it on Friday? In his January 2016 paper entitled “Day of the Week and the Cross-Section of Returns”, Justin Birru examines how long-short U.S. stock anomaly portfolio returns vary by day of the week. His hypothesis is that pessimistic (optimistic) mood on Monday (Friday) leads to relatively low (high) returns for speculative stocks. His analysis focuses on 14 anomalies arguably tied to investor sentiment, with one side (short or long) speculative and the other side non-speculative, based on idiosyncratic volatility, lottery-like, firm age, distress, profitability, payouts, size or illiquidity. He also tests anomalies arguably unrelated to investor sentiment based on momentum, book-to-market, and asset growth. Using anomaly variable and return data for a broad sample of U.S. common stocks during July 1963 through December 2013, he finds that: Keep Reading

Are high-frequency sentiment feeds useful in predicting stock market behavior? In the November 2015 version of their paper entitled “Stock Return Predictability and Investor Sentiment: A High-Frequency Perspective”, Licheng Sun, Mohammad Najand and Jiancheng Shen measure the predictive power of half-hour changes in investor sentiment for subsequent half-hour U.S. stock market returns during the trading day. Their intraday sentiment is based on the Thomson Reuters MarketPsych Indices (TRMI), which provide textual analysis of news wires, internet news sources and social media. They test exploitability via a strategy that buys (sells) SPDR S&P 500 (SPY) during each of the last four half-hours of the trading day when the preceding change in sentiment predicts a positive (negative) return. Using intraday TRMI data aggregated in half-hours and intraday half-hour returns for SPY during 1998 to 2011, they find that: Keep Reading