Sentiment Analysis

MORE THAN WORDS:
QUANTIFYING LANGUAGE TO MEASURE
FIRMS’ FUNDAMENTALS
JOURNAL OF FINANCE, 2008
CONVERSATIONS ON FINANCE
PRESENTED BY:
SAHITHI GADDAM | UDIT GUPTA | JOHN LIU | BEEJAL SHAH

SHOULD THIS IMPACT STOCK PRICE?
Source: Wall Street Journal (Oct 23, 2014)

SHOULD THIS IMPACT STOCK PRICE?
Source: Factiva.

AGENDA
Part I.
 Motivation For The Study
Part II.
 Base Paper - Overview
 Case Study
 Principal Idea Explored
 Testing For Predictability Power
 Conclusion
Part III.
 Discussion on Present Scenario

MOTIVATION FOR THE STUDY
(1/2)
 Efficient Markets claim
 Firm’s Value = Expected [Present Value (Cash Flows)]
 Conditional ‘Expectation’ based on Investor’s Information Set
 Investor’s Information Set = Quantitative + Qualitative
 Abundant literature studying Quantitative information
 However, substantial stock price movements are not explained
by quantitative measures (of firm’s fundamentals)
 Qualitative information may help explain stock returns
 Firm’s business environment, operations and prospects etc.

MOTIVATION FOR THE STUDY
(2/2)
 Possible advantages from quantifying language
1) Allows researchers to study the impact of limitless variety of
events (e.g. the Microsoft case)
2) May have incremental explanatory power for future earnings
and returns
 If analysts’ forecasts and accounting variables are
incomplete or biased
 Using Negative vs. Positive words
 Literature in psychology
 ‘Negative’ words, best summarize the cross-sectional variation
in the word list, as compared to other categories
 In the following study - primary focus is negative news

BASE PAPER - OVERVIEW
 Does ‘language’ predict firms’ ‘accounting earnings’ and
‘stock returns’
 Major findings:
1) Negative words in firm-specific news stories forecasts low
earnings
2) Stock prices briefly underreact to the information embedded
in negative words, but incorporate fully with a slight delay
3) Negative words in stories that focus on fundamentals – have
highest predictability power (on earning and return)
 Findings suggest: Investors quickly incorporate information
on firms’ fundamentals available in linguistic media, into
stock prices

PRINCIPAL IDEA EXPLORED IN THE
PAPER
 Principal idea explored:
 Can a simple quantitative measure of language be used to
predict individual firm’s earnings and stock returns
 If yes, then how to quantify the language used in financial new
stories
 Unit of Measure (defined in the paper):
 Raw metric: 𝑁 =
𝑛𝑜.𝑜𝑓 𝑁𝑒𝑔𝑎𝑡𝑖𝑣𝑒 𝑊𝑜𝑟𝑑𝑠
𝑛𝑜. 𝑜𝑓 𝑇𝑜𝑡𝑎𝑙 𝑊𝑜𝑟𝑑𝑠
 Standardized metric: 𝑛𝑒𝑔 =
𝑁 − 𝜇 𝑁
𝜎 𝑁
 𝐌𝐞𝐭𝐫𝐢𝐜 𝐮𝐬𝐞𝐝 𝐚𝐬 𝐢𝐧𝐝𝐞𝐩𝐞𝐧𝐝𝐞𝐧𝐭 𝐯𝐚𝐫𝐢𝐚𝐛𝐥𝐞: 𝒏𝒆𝒈(−𝟑𝟎,−𝟑)
(i.e. treat all news stories in the [-30,-3] trading day period, prior
to an earnings announcement, as one composite story)

CONTENT ANALYSIS METHODOLOGY
 Research area:
 Qualitative analysis; Natural language processing
 Content analysis: Two-step process
Word Category Freq. Value
Alleged Negativ 1/29 1
Abuse Negativ 1/29 1
Worse Negativ 1/29 1
Happy Pstv 0/29 1
Neutral Passive 0/29 1
Step 1:
Mapping
Step 2:
Summarizing
Number of negative words
> 99% of all news articles
Example:

DATASET
MEASURING NEGATIVITY
 Harvard-IV-4 psychological dictionary to categorize
positive and negative words
 Around 12,000 words (rows) and 180 categories
(columns)
 Measure negativity by negative word frequency
 Standardized fraction of negative words per story
 Combine all stories per firm for each trading day to
measure frequency
Source: General Inquirer Website

DATASET
FIRMS AND STORIES
 1980 to 2004
 S&P 500 firms
 Represent ¾ of U.S. market capitalization
 DJNS and WSJ stories
 350,000 stories
 100,000,000 words
 Stories for 95.8% of S&P 500 firms
 Center for Research on Security Prices for stock price data
 Institutional Brokers’ Estimates System for analyst forecast
data
 Compustat for accounting data
 Factiva database for news stories

MICROSOFT’S CASE STUDY (1/2)
 Second sentence: “The alleged ‘pricing abuse will only get
worse if Microsoft is not disciplined sternly by the antitrust
court,’ said Mark Cooper, director of research for Consumer
Federal of America.”
 Hypothesis: fraction of negative words relates to effect of
news on market value
Source: Factiva
 1999 DJNS article headline:

MICROSOFT’S CASE STUDY (2/2)
Source: Google Finance
 Fraction of negative words is in 99th percentile of negative
sentences
 Microsoft had irregularly low stock returns around news
story
 Cumulative abnormal stock return of -42, -141 and -194
bps for the 3 trading days surrounding the news event

TESTING FOR PREDICTABILITY POWER
 In order to impact stock returns, at least one relationship
must hold:
1) Negative words predict Earnings (proxy for cash flows)
2) Negative words predict Discount Rates (proxy by returns)
 OLS regression tests performed
 using different dependent variables and control variables

TEST 1 - EARNINGS PREDICTABILITY
 Dependent Variable: Two measures of quarterly earnings
used
 Standardized Unexpected Earnings (SUE)(1)
- Raw metric: 𝑈𝐸𝑡 = 𝐸𝑡 − 𝐸𝑥𝑝𝑒𝑐𝑡𝑒𝑑 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝐸𝑡
- Standardized metric: 𝑆𝑈𝐸𝑡 =
𝑈𝐸𝑡−µ 𝑈𝐸 𝑡
𝜎 𝑈𝐸 𝑡
 Standardized Analysts Forecast Errors (SAFE)
- SAFE =
Median analyst forecast error
σUEt
 Control Variables:
 lagged earnings, neg-30,-3, size, B/M, trading volume, recent
stock returns (3 measures), analyst forecast revision(2) etc.
 Winsorize SUE & all analysts forecast variables at 1% level
 Similar results from both ‘SUE’ and ‘SAFE’
Note: (1) Based on Bernard and Thomas (1989), who use a seasonal random walk with trend model for
each firm’s earnings. (2) Using Chan et. al. (1996) methodology.

MAIN RESULT
 All 6 estimates are significant at 99% level
 Several control variables also exhibit strong explanatory power, as expected
 Predictability is robust even when using ‘before forecasts’ news stories
Note: Table presented above has been truncated, and does not include all control variables. Please
refer to the appendix for complete details.

TEST 2 - RETURN PREDICTABILITY
 Following two ideas are tested
 Return predictability in daily returns
 Is there a trading strategy possible off this under-reaction?
 Considerations:
 Data at daily frequency
 Dependent variable: return based on closing price (t=0 & t=1)
 Cut-off time: DJNS (up to 3:30pm), WSJ (same day)
 Control Variables - Earnings, size, B/M, trading volume, recent
stock returns (5 measures)

DAILY RETURNS, MAIN RESULT
 neg robustly predicts slightly lower returns on the following trading day
 neg coeff. is significant in 4 cases (where DJNS data in included)
 Coeff. for DJNS source is higher, as compared to WSJ
 Low R2, as expected in efficient markets (< 0.0026)
Note: Table presented above has been truncated, and does not include all control variables. Please

TRADING STRATEGY
 Two equal weighted portfolios – constructed by ranking firms on the
basis of positive/ negative news
 Long-short strategy, with daily rebalancing
 Cumulative raw returns would be 21.1% per year (no trading costs)
 Strategy will not be profitable if trading cost is considered
Note: Standard errors calculated using White (1980) heteroskedasticity-consistent covariance matrix
approach.

IS THERE A SUBSET OF NEWS
WITH BETTER PREDICTABILITY
 Hypothesis:
 Negative words in news stories containing word-stem ‘earn’
have better predictability
 Results Expected:
 Better earnings predictability
 Stronger contemporaneous relationship with returns
 Magnitude of under-reaction should be greater

ADDING NEW INDEPENDENT VARIABLES
 Regression (similar to previous case):
 Add 2 new independent variables to capture specific effects
 ‘Fund-30,-3’: words in stories containing word-stem ‘earn’,
divided by total words across all stories
 Interaction term: neg-30,-3 * Fund-30,-3
Not
News Stories
Not “About” Firm
Fundamentals
“About” Firm
Fundamentals
neg-30,-3
Fund-30,-3

REPEATING REGRESSIONS WITH
TWO ADDITIONAL VARIABLES
 Coeff. of both new terms is strongly negative and significant
 Interaction: negative words in earnings-related stories are much better predictors
Note: Tables presented above has been truncated, and does not include all control variables. Please
 Strong contemporaneous relationship exists
 5x larger response from negative words in earnings-related stories

CONCLUSION
 More than ‘negative’ words
 Contain valuable information
 Forecast low earnings
 Return Predictability
 Slight delay in reaction to negative news
 Predictability in t+1 day return
 ‘Simple’ trading strategy does not exist
 Words from specific type of news carry more information
 Negative words from earnings related stories are better
predictors

POWER OF SOCIAL MEDIA:
THE HASH CRASH INCIDENT
 Demonstrated social media’s potential to move markets
 Dow Jones fell more than 150 points, the price of crude oil
plummeted, and US bond yields dropped, briefly wiping $121
billion off the value of companies in the S&P 500 index, before
recovering minutes later

1. ‘SNTMNT’- OVERVIEW
 Launched an API to monitor Twitter-based stock sentiment
 World’s first API that makes predictions based about future stock
price movement for all stocks in the S&P 500
 Accuracy as high as 60%, averaging at 54% (company estimates)
 Should traders rely on Twitter sentiment alone for their trades?
 Signal-to-noise ratio on social media channels is too low to provide
standalone trading signals, but definitely high enough to provide
an innovative trading indicator
 Based on work of Professor Johan Bollen, an academic who found
correlations between the stock market and activity on Twitter

2. SOCIAL MARKET ANALYTICS -
METHODOLOGY
 Social Market Analytics produces a family of metrics, called
S-Factors – designed to capture the signature of financial market
sentiment
 SMA applies these metrics to data captured from social media
sources to estimate sentiment for indices, sectors, and individual
securities

SOME USEFUL ONLINE RESOURCES
 IBM’s Watson and the Jeopardy! Challenge:
https://www.youtube.com/watch?v=P18EdAKuC1U
 Free online course on ‘Natural Language Processing’,
offered on Coursera by Stanford University:
https://www.coursera.org/course/nlp
 General Inquirer website:
http://www.wjh.harvard.edu/~inquirer/
 IBM Publications:
http://researcher.watson.ibm.com/researcher/view_group.
php?id=147
 List of words in spreadsheet format:
http://www.wjh.harvard.edu/~inquirer/spreadsheet_guide.
htm
 Princeton’s WordNet: http://wordnet.princeton.edu/

REFERENCES
Tetlock, Paul C., Maytal Saar-Tsechansky, and Sofus Mackassy.
2008. “More than words: Quantifying language to
measure firms’ fundamentals.” The Journal of Finance 63
Issue 3 p. 1437-1467.
“Welcome to the General Inquirer Home Page.” Web. 6 April
2015. < http://www.wjh.harvard.edu/~inquirer/
spreadsheet_guide.htm>.

QUESTIONS AND DISCUSSION
 Can you think of other examples in which any news had an
impact on returns?
 Do you think big firms are using such analysis for alpha
generation?

APPENDIX

APPENDIX
SUBSET - EARNINGS PREDICTABILITY

APPENDIX
SUBSET - RETURNS PREDICTABILITY

Sentiment Analysis

More Related Content

What's hot (20)

Viewers also liked (11)

Similar to Sentiment Analysis (20)

Recently uploaded (20)

Sentiment Analysis