Sentiment-Driven Financial Intelligence
Download as PPTX, PDF3 likes416 views
This document discusses using sentiment analysis and behavioral finance to generate investment returns. It outlines how natural language processing can extract sentiment and behavioral biases from public and private investor communications. These extracted features are then used to construct factor portfolios targeting specific market anomalies related to biases. Preliminary results show some behavioral factors have significant premiums and improve upon traditional models like Fama-French in explaining returns. Overall the approach aims to take advantage of cognitive and emotional biases to identify new sources of excess return, or "alpha".
![Quantitative Behavioral Finance
• Behavioral asset pricing model:
• This is a linear factor model that includes behavioral
bias sentiment factors
• Exposure to behavioral biases given by weights gi
2/9/2017 14
E r[ ] = rf + bj
j=1
k
å (rj -rf )+ gi
i=1
m
å bi
Expected
Return
Risk-free
Rate
Factor
Beta
Bias
Sentiment
Premium
Risk
Factor
Premiu
m](https://image.slidesharecdn.com/johnliusentimentsymposiumpresentation-170209041017/85/Sentiment-Driven-Financial-Intelligence-14-320.jpg)
Sentiment-Driven Financial Intelligence
- 1. Sentiment-Driven Financial Intelligence By John Liu, Ph.D, CFA 1
- 2. • Introduction • Financial Intelligence • Behavioral Approach • Key Takeaways • Questions 2/9/2017 2
- 3. 2/9/2017 3 Machine Learning platform that understands human communication Nashville Washington New York London Goldman Sachs Credit Suisse In-Q-Tel Silver Lake (individuals) National Security Financial Services Health Analytics Data Science CONFIDENTIAL 3 DIGITAL REASONING | CONFIDENTIAL Prepared Exclusively for Intel Capital Digital Reasoning Trusted Cognitive Computing For A Better World
- 4. DR Platform 2/9/2017 4 • Scalable platform designed to analyze human communications
- 5. What is Financial Intelligence? 2/9/2017 5 Customer Insight Market Intelligence Investment Research Risk & Compliance Predict client interests and behavior from communications and interactions Uncover investment opportunities through insights derived from public and private data Extract key market themes and monitor risk metrics from structured and unstructured data Monitor AML, SAR, misconduct, manipulation and unauthorized trading, trade reconstruction
- 6. Customer Insight • Goal: leverage client communications to predict client preferences and behavior, extract client holdings • CRM data, emails, documents, phone calls, chat • Example: • Banking chat mining: 2/9/2017 6 * * Bid/Offer * * Financial Product Financial Product * Bid/Offer * * * Financial Entity Financial Benchmark Financial Benchmark Financial Benchmark Financial Model Financial Benchmark Bid/Offer * Financial Level Financial Benchmark ** * *
- 7. Risk & Compliance • Goal: Monitor AML, SAR, misconduct, manipulation and unauthorized trading, trade reconstruction 2/9/2017 7 Electronic Communications Surveillance FX Rate Manipulation • Reveal collusion and market manipulation • "Quid-pro- quo” exchanges • Clustered Bid/Ask quotes • Order execution around the benchmark • Indentify trading abuse • Deal related language • Abusive/dirty references • Improper placement and execution Wall Cross Violations • Uncover insider trading • M & A terminology • Personal trade execution • Discussing Companies on Restricted list • Email from/to sensitive countries • Bribery & gift language • Presence of attachments • Deal related language & order execution • Use public information to enhance KYC profiles • Email to/from economically sanctioned countries • Deal related language & order execution • Analyze wire instruction fields for hidden risks Conduct Risk / Anti-Bribery Anti-Money LaunderingUnauthorized Trading
- 8. Useful Analyst Sentiment? “The stable outlook reflects our expectations that Apple will continue to maintain and defend its strong market position in mobile devices and tablets despite seeing an contraction in its addressable market.” 2/9/2017 8 Source: Yahoo
- 9. Investment Intelligence • Goal: identify persistent sources of alpha through sentiment analysis • Key is “persistent” as sentiment is widely viewed as transitory and unstable over long-term • Current approaches incorporate emotional, affect psychology and intent models that aggregate individual sentiment to identify short-term group trends • Incorporate models of investor behavior? 2/9/2017 9 Alpha: the active return above that of a passive benchmark
- 10. 2/9/2017 10 Investments & Finance Behavioral Science Computer Science Sentiment Analysis Algorithmic Trading Behavioral Economics Cognitive Alpha Investment Intelligence
- 11. Traditional vs. Behavioral Economics Traditional • Perfect rationality • Expected utility theory • Risk aversion • Efficient market hypothesis • CAPM and APT Behavioral • Bounded rationality • Prospect theory • Loss aversion • Adaptive market hypothesis • Behavioral Asset Pricing • Market anomalies 2/9/2017 11 Empirical evidence diverges greatly from classic theory Why do market anomalies exist?
- 12. Drivers of Anomalies Cognitive Biases • Conservatism • Overweight Bayesian priors • Confirmation • Selection bias • Gambler’s Fallacy • Mean reversion • Representativeness • Overweight recent evidence • Availability • Familiarity premium Emotional Biases • Loss aversion • Pain > Gain • Overconfidence • Underestimate risk • Endowment • Sentimental value premium • Status-Quo • Regret-aversion • Herd behavior 2/9/2017 12 Anomalies are drivers of excess return (alpha)
- 13. Studies of Market Anomalies 2/9/2017 13 Year Authors Anomaly 1968 Ball & Brown Earnings announcement effect 1975 Haugen & Heins Low volatility effect 1976 Rozeff &Kinney January effect 1977 Basu Low P/E effect 1981 Gibbons &Hess Monday effect 1981 Shiller Excess volatility effect 1981 Banz Size effect 1985 Rosenberg, Reid, Lanstein Value effect 1987 DeBondt & Thaler Over-reaction to bad news 1993 Jegadeesh & Titman Momentum investing
- 14. Quantitative Behavioral Finance • Behavioral asset pricing model: • This is a linear factor model that includes behavioral bias sentiment factors • Exposure to behavioral biases given by weights gi 2/9/2017 14 E r[ ] = rf + bj j=1 k å (rj -rf )+ gi i=1 m å bi Expected Return Risk-free Rate Factor Beta Bias Sentiment Premium Risk Factor Premiu m
- 15. In Search of Alpha 2/9/2017 15 NLP Sentiment Analysis Behavioral Bias Factors Factor Portfolios Alpha 1. Context-based sentiment analysis to extract features of behavioral biases observed in investor commentary 2. Combine these features with structured data to formulate behavioral bias factors 3. Use the factors to fit regression models for asset returns 4. Construct factor portfolios to monetize market anomalies by overweight/underweight exposure to specific behavioral biases
- 16. Deep-Learning Sentiment Pipeline NLP Pipeline Our Methodology at DR • Data: public and private investor commentary 2/9/2017 16 TOK docs POS Shallow Parse Extract or NER W2V LRDBOWDM Agg Synthesys Studio Factor Constructio n Facts Context
- 17. Some Results Standard Fama-French model (MKT, SMB, HML) regression with two behavioral factors (UMO, ITF): 2/9/2017 17 Factor Premium Standard Deviation P-value Sharpe Ratio Corr. Vs. MKT MKT 0.47 4.66 0.0262 0.10 1.00 SMB 0.19 3.16 0.1942 0.06 0.28 HML 0.42 3.04 0.0022 0.14 -0.33 UMO 0.32 1.57 0.0001 0.20 -0.12 ITF 0.91 3.06 0.0001 0.30 -0.50 Sun 2014
- 18. Key Takeaways • Intersection between sentiment analysis and behavioral economics is opening up new opportunities for investment alpha generation • DR’s context-based sentiment analysis methodology is useful for extracting behavioral bias features from public and private data • Behavioral bias feature extraction yields factor portfolios that can take advantage of anomalies manifested from these biases • Open research field 2/9/2017 18
- 19. Final Words • We at DR are actively exploring neural embedding and deep learning methods to enhance feature extraction of behavioral biases • Toolbox: Synthesys Studio • Text analytics • Model training • Open platform • Q4 2015 release • Freeware 2/9/2017 19 Investments & Finance Behavioral Science Computer Science Sentiment Analysis Algorithmic Trading Behavioral Economic s Cognitive Alpha