1. Presenter: Siddharth Krishna Sinha Instructor: Jing Gao
Crowdsourcing on Wall Street: Extracting Value from Collaborative Investing Platforms
Presenter:
Siddharth Krishna Sinha
Instructor:
Jing Gao

2. AGENDA Introduction SeekingAlpha and StockTwits Methodology
Data Collection
Preliminary Data Analysis
Sentiment Extraction
Predicting Stock Price Changes
Practical Sentiment-Based Trading

3. Introduction

7. SeekingAlpha StockTwits Started in 2004
Independent stock analysis platform
3 million registered users
8 million monthly unique users
8000 registered contributors
Editorial Board
$10 per 1000 views on articles
Started in 2009
Financial social network for sharing ideas
300K registered users
Anyone can contribute –peer
Short messages:140 characters
No editorial board

8. Methodology Quality of contents: Expert based versus Peer based.
Longitudinal datasets since inception.
Mixture of keyword processing and machine learning classifiers.
Extracting sentiments.
Analyze correlation between content sentiments from both services.
Subset of ‘top’ or ‘expert’ authors.
Hypothetical investment strategies. (SA outperforms broader markets, ST is slightly better)

9. Analysis

11. Sentiment Extraction Bearish or Bullish Positive or negative sentiment
Partition multistocks
Supervised machine learning.
85.5% validation for SeekingAlpha and 76.2% for StockTwits.

13. Predicting Stock Price Changes
Per-article Sentiment and Stock Performance
We ignore magnitude of price movements and strength of sentiments
Binary metrics
Compute coefficient and Pearson correlation
Rank authors
Top authors more than 75% for SA.

14. Aggregated sentiment for market prediction
Long term performance

15. Impact of time window

16. Practical Sentiment-based Trading
Ranking Authors
By prediction accuracy
By received comments

17. Sentiment based Stock Trading Strategies
Long Strategy
Weekly basis
Evenly purchase N stocks, N=500
Sell if sentiment is negative
Otherwise hold
Long/Short Strategy
More aggressive
‘Short stocks’ with negative sentiments
Risky (Hedge funds)

18. Empirical Evaluation
Q. Do our strategies outperform the broader markets? 108% at the end of the 8-year period, compared to 47.8% of the S&P 500. StockTwits produced 54.5%. (Good in absolute terms but below baseline) Q. What ranking method identifies experts most effectively? All comment obtained highest level of investment returns. A reasonably knowledgeable crowd can serve to pinpoint valuable content in an otherwise mixed collection of content. Q. Do more aggressive strategies improve performance? The high level observation is that adding shorts improves performance for strategies based on SeekingAlpha, but not significantly enough to justify the added risk

19. Q. Does this approach support “real-time” trading?
For StockTwits, daily trading leads to incrementally higher returns compared to those using weekly windows. The opposite is true on SeekingAlpha, where the lower frequency of articles mean that weekly trading outperforms daily trading strategies. This confirms the real-time nature of StockTwits message.

20. Points to be noted
We ignore strength of sentiments, and reduce strategies to a binary “buy all in” or “sell all shares.”
Whether our system will impact the overall health of the market?
Unlikely! Movements on the stock market are dominated by large hedge funds and pension funds.
The wisdom of a “smaller, smarter crowd” can outperform a larger, more general crowd, especially for crowdsourcing tasks that require specific domain knowledge (financial knowledge in our case). Moreover, the larger crowd is still helpful as members collaboratively discover domain experts through their interactions.

21. Thank You