Sentiment Indicators

Larry Connors introduces his 2018 book, Buy the Fear, Sell the Greed: 7 Behavioral Quant Strategies for Traders, by stating in Chapter 1 that the book shows when, where and how: “…to trade directly against traders and investors who are having…feelings of going crazy and impending doom. …The goal of this book is to make you aware of when and why short-term market edges exist in stocks and in ETFs, and then give you the quantified strategies to trade them. …Thirty years ago, when a news event would occur, it could take days to assimilate it. …The only thing that’s changed is the timing of their emotion; today it occurs faster and at times is more extreme primarily due to the role the media (and especially social media) plays in disseminating the news that triggers this behavior.” Based on analyses of specific trading setups using data through 2017, he finds that: Keep Reading

Does sentiment on StockTwits and Twitter social media platforms usefully predict returns for individual stocks? In their June 2018 paper entitled “Momentum, Mean-Reversion and Social Media: Evidence from StockTwits and Twitter”, Shreyash Argarwal, Pablo Azar, Andrew Lo and Taranjit Singh analyze relationships between stock price behaviors and real-time measures of sentiment uniquely attributable to StockTwits and Twitter in three ways:

1. Linear regressions for a sample of 4,544 stocks that each day relate volume and liquidity metrics for each stock to aggregate news and social media sentiments for that stock measured either during the same trading day (9:30AM to 4:00PM, for coincident relationships) or during preceding non-trading hours (4:00AM to 9:30AM, for predictive relationships).
2. An intraday event study for a subsample of 500 large-capitalization stocks that examines stock trading behaviors when associated bullish and bearish social media sentiment reaches extreme levels.
3. A backtest of an intraday mean reversion strategy applied to the 500 companies with the highest average volumes over the previous 200 days (with no more than 30% from a single sector) that exploits the power of social media sentiment to predict mean reversion. Every 30 minutes, this strategy buys (sells) stocks with negative (positive) returns over the preceding 30 minutes, with weights elevated for stocks with high StockTwits and Twitter message volume over the preceding 30 minutes.

Using the RavenPack Composite Sentiment Score to measure conventional stock sentiment, minute-by-minute StockTwits and Twitter-with-retweets data from PsychSignal to measure social media sentiment, and trade/quote data for 4,544 stocks during 2011 through 2014, they find that: Keep Reading

What firm/asset/market conditions signal mispricing? In the November 2017 version of their paper entitled “Bonds, Stocks, and Sources of Mispricing”, Doron Avramov, Tarun Chordia, Gergana Jostova and Alexander Philipov investigate drivers of U.S. corporate stock and bond mispricing based on interactions among asset prices, financial distress of associated firms and investor sentiment. They measure financial distress via Standard & Poor’s long term issuer credit rating downgrades. They measure investor sentiment primarily with the multi-input Baker-Wurgler Sentiment Index, but they also consider the University of Michigan Consumer Sentiment index and the Consumer Confidence Index. They each month measure asset mispricing by:

1. Ranking firms into tenths (deciles) based on each of 12 anomalies: price momentum, earnings momentum, idiosyncratic volatility, analyst forecast dispersion, asset growth, investments, net operating assets, accruals, gross profitability, return on assets and two measures of net share issuance.
2. Computing for each firm the equally weighted average of its anomaly rankings, such that a high (low) average ranking indicates the firms’s assets are relatively overpriced (underpriced).

Using monthly firm, stock and bond data for a sample of U.S. firms with sufficient data and investor sentiment during January 1986 through December 2016, they find that: Keep Reading

Should investors view stock returns around recurring firm events in aggregate as an exploitable anomaly? In their October 2017 paper entitled “Recurring Firm Events and Predictable Returns: The Within-Firm Time-Series”, Samuel Hartzmark and David Solomon review the body of research on relationships between recurring firm events and future stock returns. They classify events as predictable (1) releases of information or (2) corporate distributions, with some overlap. Information releases include earnings announcements, dividend announcements, earnings seasonality and predictable increases in dividends. Corporate distributions cover dividend ex-days, stock splits and stock dividends. They specify a general trading strategy to exploit these events that is long (short) stocks of applicable firms during months with (without) predictable events. They use market capitalization weighting but, since there are often more stocks in the short side, they scale short side weights downward so that overall long and short sides are equal in dollar value. Based on the body of research and updated analyses based on firm event data and associated stock prices from initial availabilities through December 2016, they conclude that:

Keep Reading

How effective is investor sentiment in predicting stock market returns? In his October 2017 paper entitled “Measuring Investor Sentiment”, Guofu Zhou reviews various measures of equity-oriented investor sentiment based on U.S. market, survey and media data. He highlights the Baker-Wurgler Index (the most widely used), which is based on the first principal component of six sentiment inputs: (1) detrended NYSE trading volume; (2) closed-end fund discount relative to net asset value; (3) number of initial public offerings (IPO); (4) average first-day return on IPOs; (5) ratio of equity issues to total market equity/debt; and, (6) dividend premium (difference between average market-to-book ratios of dividend payers and non-dividend payers). Based on the body of research and using monthly inputs for the Baker-Wurgler Index during July 1965 through December 2016, three sets of investor sentiment survey data since inceptions (between Dec 1969 and July 1987) through December 2016 and two sets of textual analysis data spanning Jan 2003 through December 2014 and Jul 2004 through Dec 2011, he finds that: Keep Reading

Does margin debt serve as an intermediate-term stock market sentiment indicator based on either momentum (with an increase/decrease in margin debt signaling a continuing stock market advance/decline) or reversion (with change in margin debt signaling a pending reversal)? To investigate, we relate the behavior of NYSE end-of-month margin debt, published with a delay of about a month, with the monthly behavior of the S&P 500 Index as a proxy for the U.S. stock market. Using monthly data during January 1959 through August 2017 (703 months), we find that: Keep Reading

Do certain types of news about specific stocks reliably predict risk-adjusted returns of those stocks? In their March 2017 paper entitled “Using Natural Language Processing Techniques for Stock Return Predictions”, Ming Li Chew, Sahil Puri, Arsh Sood and Adam Wearne investigate relationships between financial news headlines and stock returns stripped of non-news risks. They use natural language processing to classify corporate events by firm, illustrating via five types: dividend declaration; oversold conditions; receipt of approval; signing an agreement; and, hiring an advisor. They isolate each type by segmenting headlines into 10, 20, 50 or 100 clusters of similar headlines. They then form portfolios for the most relevant clusters that are long (short) stocks for which events have occurred (same-industry stocks for which there are no events), with positions weighted to eliminate exposures to market, size and value factors. Outputs include factor-adjusted cumulative and daily average returns. They focus on stocks in the S&P 500 as it evolves and divide the sample into 2006-2014 to identify event clusters in-sample and 2015-2016 to test cluster portfolio performance out-of-sample. Using 60,949 active voice financial news headlines that relate to specific S&P 500 firms and associated daily/quarterly stock price and firm characteristics data during 2006 through early 2017, they find that: Keep Reading

Should investors buy or sell stocks experiencing unique (idiosyncratic) volatility spikes? In their August 2016 paper entitled “Unusual News Flow and the Cross-Section of Stock Returns”, Turan Bali, Andriy Bodnaruk, Anna Scherbina and Yi Tang investigate relationships among sudden increases in stock idiosyncratic volatility, unusual firm news, changes in analyst earnings forecast dispersion, short selling and future returns. They identify idiosyncratic volatility shocks as large deviations from the volatility predicted out-of-sample by a regression model that accounts for market, size and book-to-market effects. They identify unusual news flow using Thomson-Reuters News Analytics data (covering 41 media) by comparing the number of stories about a firm in the current month to the average monthly coverage the prior four months, measured overall and separately for positive, negative and neutral stories. They measure changes in analyst earnings forecast dispersion (standard deviation divided by mean) based on data from I/B/E/S as the difference between current dispersion and dispersion two months ago. They measure data on shorting demand and utilization (shares borrowed divided by shares available for lending) using data from Markit. Using monthly values of the specified data from various inceptions through December 2012, they find that: Keep Reading

Do measures of investor attention to specific firms/stocks indicate how the stocks react to earnings surprises? In their July 2016 paper entitled “Yahoo Finance Search and Earnings Announcements”, Alastair Lawrence, James Ryans,  Estelle Sun and Nikolay Laptev investigate the interaction of investor attention and earnings surprises. They focus on abnormal Yahoo Finance search activity as the measure of attention. They define abnormal search activity on a certain day as the total number of searches that day minus average number of searches on the same day of the week during the prior 10 weeks, divided by average number of searches on the same day of the week during the prior 10 weeks. They examine the interaction of abnormal search activity and standardized unexpected earnings (earnings surprises). For comparison they also consider interaction of earnings surprises with three other commonly used measures of investor attention: abnormal trading volume, EDGAR search and Google Trends search. Using daily search and quarterly earnings announcement data during July 2014 through June 2015 (14,172 firm-earnings announcement observations) and associated daily stock returns during July 2014 through June 2016, they find that: Keep Reading

What variables best predict U.S. stock market returns? In his June 2016 paper entitled “Which Variables Predict and Forecast Stock Market Returns?”, David McMillan examines the power of 25 variables to predict excess return (relative to the 3-month U.S. Treasury bill yield) of Shiller’s S&P Composite Index both in-sample and out-of-sample. He chooses variables based on connectedness to expected cash flow/dividends and risk and assigns them to five groups:

1. Financial ratios: dividend-price, price-to-earnings, cyclically adjusted price-to-earnings (CAPE or P/E10), Tobin’s Q and market capitalization-to-Gross Domestic Product (GDP).
2. Economic:  GDP cycle, GDP acceleration (rate of change in GDP growth), consumption growth, 10-year to 3-month Treasuries term spread and inflation.
3. Labor: wage growth, unemployment, natural rate of unemployment, productivity growth and labor market conditions.
4. Housing: house price growth, house affordability, home ownership, housing supply and new house sales.
5. Other: University of Michigan Consumer Sentiment, Purchasing Managers Index, National Financial Conditions Index, leverage and non-financial leverage.

He employs regressions to test in-sample predictive power. He then tests out-of-sample forecasts starting in 2000 using various forecast methods and accuracy measures and considering both single-variable and multi-variable models. Using the specified data series as available during 1973 through 2014, he finds that: Keep Reading