Is All That Talk Just Noise? The Information Content of Internet Stock Message Boards Authored by Werner Antweiler and Murray Z. Frank Presented by: Suisui Huang Agenda Research Results Preview Paper Structure Message Board Data and Classification (Data pre-processing) Data Description Relating Message Boards and Stock Returns Volatility
Trading Volume My Opinions Research Results Preview Three specific issues considered Does the number of messages posted or the bullishness of these messages help to predict returns? Is disagreement among the messages associated with trading volume? Does the number of messages posted or the bullishness of the messages help to predict volatility?
Research Results Preview The First Issue Question: Does the number of messages posted or the bullishness of these messages help to predict returns? Result: An increase in the number of messages posted has a statistically significant but economically small negative return on the next day Research Results Preview The Second Issue Question:
Is disagreement among the messages associated with trading volume? Historical Researches (Opposite opinions): Traditional Hypothesis: Disagreement induces trading No-trade Theorem: Disagreement leads to a revision of market prices and beliefs Result: Disagreement induces contemporaneous trading but reduces the trading volume on the next day Research Results Preview The Third Issue Question: Does the number of messages posted or the bullishness of the messages help to
predict volatility? Model for volatility: Fractionally integrated realized volatility news response function Result: Message posting and trading volume help predict volatility Message Board Data and Classification Message Data collection (Data Pre-processing Part 1) Source: Yahoo! Finance & Raging Bull message boards Object: 45 firms that together made up the Dow Jones Industrial Average (DIA) and the Dow Jones Internet Commerce Index (XLK).
Format: 1.5 million text messages Bullish? Bearish? Neither? Message Board Data and Classification Nave Bayes Classification Method Key Assumption: Occurrences of words are independent of each other (Nave) Procedure:
Consider a stream of words Wi that are found either in a message of type T or itsanti-type T-prime Updating Rule: With mi: the number of occurrences of this word Wi in type T mi-prime: the number of occurrences of this word W i in anti-type Tprime ni: the total number of words in type T
n -prime: the total number of words in type T-prime Message Board Data and Classification Nave Bayes Classification Method Procedures: 1. Manually split 1000 messages into buy, sell and hold 2. Run a software named rainbow, set Nave Bayes for method, get 3 probabilities P(T|W k for each message k and each categories T (Buy, hold or sell) 3. Choose the classification with the highest probability for message k In-sample Accuracy:
Message Board Data and Classification Aggregation of the Coded Messages (Data Pre-processing Part 2) Bullishness signal Bt: MtBuy : total number of messages in category BUY during the time interval t. MtSell : total number of messages in category SELL during the time interval t. Agreement Index At: Data Description Messages & Financial Data
Figure2 shows the weekly message posting over the full year 2000 message posting activity was reasonably stable over the year Figure3 shows the weekly trading volume over the full year 2000 There was a decline in the volume of stock trading over the year The trading volume is often elevated during the same week that message posting is elevated Data Description
Messages & Financial Data Yahoo! Finance vs Raging Bull: More messages More bullishness Shorter messages DIA vs XLK: Lower losses Lower volatility Lower activity levels Less bullishness Higher level of coverage by WSJ
Relating Message Boards and Stock Returns Time sequencing tests Significant negative relationship between the number of messages on day t and stock returns on day (t+1). Significant positive relationship between the number of messages on day t and stock returns on day (t+2). Results: The return predictability is statistically significant but very small in magnitude Difficult to take advantage of because potential gains
would likely be offset by small transaction costs Relating Message Boards and Volatility Fractionally integrated realized volatility news response function 1. What is Vi,t Rt,i,d: Return of company i on day t and intra-day period d (15 mins) Vi,t: Volatility of company i on day t
2. What is d (fractional integration parameter) *Log-periodogram method (Geweke and PorterHudak(1983)) Using estimates of the periodogram for company i at frequency points k=0,1,...,T^0.6, we get d=0.3097 Relating Message Boards and Volatility Fractionally integrated realized volatility news response function 3. What is I(ri,t-1<0)
4. What is L,A, M and N: An indicator variable for yesterdays return to be negative L: Lag operator Leverage effect: When the coefficient is not 0, then there exists leverage effect. The sign of yesterdays return,
negative or positive, will have an impact on todays volatility M: Number of messages A: Agreement Index N: Number of trades Relating Message Boards and Volatility Time sequencing tests
We can predict volatility using the message posting data More messages today imply significantly greater market volatility tomorrow Message posting activity has a more significant effect on market volatility than market volatility has on message posting Relating Message Boards and Volatility Volatility Panel Regressions The number of messages is a predictive factor of volatility for both DIA and XLK firms The number of trades is a significant factor for the DIA firms, but not significant for the XLK firms
Agreement among the posted messages is not significant Relating Message Boards and Trading Volume Volume regressions WSJs effect If there is an article in todays WSJ, trading is elevated. If there was an article in yesterdays WSJ, then todays trading is depressed Message posting
The message board posting volume remains a predictor of trading volume. The sign is positive and it is statistically significant Relating Message Boards and Trading Volume Contemporaneous Regressions The Contemporaneous Regressions support for the hypothesis that
disagreement produces trading Relating Message Boards and Trading Volume Time sequencing tests Greater agreement on a given day is followed by more trades on the next day Sluggish traders hypothesis: It takes a while for people to get around to completing their trades. Cao(2002) suggests that when a person learns that others have received the same signal that they have received, this will make them more willing to trade My Opinions
Agreement Disagreement Innovative model for volatility Test error for Nave Bayes Method? Interesting results found between message posting and return
R square for volume regressions relatively small Granger causality test Thanks for your time! Any questions?
24. From Petroleum to Pharmaceuticals (chemistry of organic molecules, the organic world). Exam 3: Wednesday, November 30 (6 lectures) _____ Period before final (3 lectures) 25. Synthetic and Biological Polymers (chemistry of giant molecules and life).
European demand for more and cheaper products. Search for New Routes to the East- Ottomans had a monopoly on trade routes. The 3 G's- "Gold, Glory, Gospel" (Explorers wanted Gold and Glory/ missionaries wanted to convert Natives Reasons for Exploration...
Forgiving Each Other As Christ Forgave Us:Defeating Injustice in the Church Through Forgiveness. ... "Who is a God like you who pardons sin and forgives the transgression of the remnant of his inheritance? You do not stay angry forever but...
MasterCard International Credit Card Security & Risk IS6800 Group Presentation Mike Cornish Kathleen Delpha Mary Erslon November 2004 Agenda MasterCard Organization Credit Card 101 Credit Card Fraud Case Studies Card Not Present Fraud Identity Theft Fraud Best Practices for Credit...
Ready to download the document? Go ahead and hit continue!