Transcript Detecting Insider Trading
June 2, 2008
Detecting Insider Trading
MS&E444 Final Presentation Manabu Kishimoto Xu Tian Li Xu
Overview
• Motivation & Focus • Litigation Case Study (CNS Inc.) • Detecting Strategy • Automation and Optimization • Performance Evaluation • Conclusion
Motivation & Focus
• If we can detect insider trading
before the news release
, we can generate excess returns.
• In our project, we focus on the
option market
because – It gives leveraged return for insiders; – It is more thinly traded than the stock market; – It is more informative than the stock market.
• We also focus on
good news
(e.g. Acquisition).
$ 40 35 30 25 20 15 10 5 0 8/9/2006
Daily Stock Price (CNS Inc.)
36.72
28.5% increase
28.56
GlaxoSmithKline would acquire CNS for $37.50 per share
10/9/2006
Acquisition News Release
11/9/2006
2000 1800 1600 1400 1200 1000 800 600 400 200 0 9/11/2006
Daily Option Volume (CNS Inc.) SEC claims that there was illegal insider trading on these four trading days.
9/27 10/2 call put 10/9
News
1600 1400 1200 1000 800 600 400 200 0
Aggregated Call Option Volume (CNS Inc.) Sep 27 - Oct 2, 2006
Stock Price: $28 17.5
20 22.5
25 Strike Price ($) 30 35
900 800 700 600 500 400 300 200 100 0
Aggregated Call Option Volume (CNS Inc.) Sep 27 - Oct 2, 2006
3 weeks (Oct 21) 7 weeks (Nov 18) 11 weeks (Dec 16) Time to Expiration 24 weeks (Mar 17, 2007)
Salient Statistical Patterns
1. Call-put imbalance is large; 2. Total option volume is high; 3. Insiders prefer slightly in-the-money or out-of-the-money option; 4. Near-term option is preferred.
Detecting Strategy (1)
100
days
10
days News?
• • Insider?
Background Signal
Use moving windows
: take
100
background data and
10
trading days as days as the signal
Filter the data
: focus on the data which satisfy the following two conditions:
1. Strike Price Filter Criterion
Stock price – Strike price Stock price <
+0.15
2. Expiration Date Filter Criterion
Expiration date – Current date <
6
months
Detecting Strategy (2)
100
days
10
days News?
• Insider?
Background Signal
Apply the following criteria
:
1. Call Ratio Criterion
Call volume Call volume + Put volume >
2. Total Volume Criterion
Signal daily average volume Background daily average volume
75%
>
1
Automation
• Automatic processing script (PERL) • Optimize detection criteria • Use several benchmarks to evaluate the effectiveness of detection strategy # of Tickers Year # of events Litigation Database Training Database CNXS, DJ, INVN 3 2002/01-2004/06 2005/01-2007/06 15 Event Database 99 2005/01/01 2007/06/30 474 Testing Database 2007 First Half OptionMetrics Database 3068 2007/01/01 2007/06/30 1902
Optimize Detection Criteria
• Define: – Right Detection: stock price
rallies ≥ 10%
– Wrong Detection: stock price
sinks ≥ 10%
• Optimization on Training database – Optimize to maximize
Right/Total Ratio
– Optimize the criteria to maximize
Right/Wrong Ratio
• Change only
one
parameter at a time
Performance Evaluation Benchmark #1: Histogram of Stock Return
• If we buy 1 share of stock when the signal suggests insider events, and sell it after holding it for
10
days, we obtained the histogram of the percentage return for all tickers in the database.
Litigation Database Training Database Testing Database
Performance Evaluation Benchmark #2: Percentage Return of Non-leveraged Simple Trading Strategy
• Non-leveraged Simple Trading Strategy ( NSTS ): – Allocate
$1
for every ticker in the database – Check whether there is possible insider trading just before the market closes
Yes
: Use all balance allocated to buy shares of stocks and sell it after
10
days.
No
: Do nothing.
– Calculate annualized percentage returns for all the funds allocated at the end of the period – Compare the return with the
Buy-and-Hold
strategy NSTS Return Buy and hold Return Litigation Database +15% +39% ( Acquisition rich ) Training Database +5.7% +28% ( Acquisition rich ) Testing Database
+7.47% +2.82%
Performance Evaluation Benchmark #3: Histogram of S ignal’s Lead Time before the News Announcement Training Database Testing Database
Performance Evaluation Benchmark #4: Prediction Errors
# of events Detected?
Yes No
# of events Detected?
Yes No
# of events Detected?
Yes No
Stock jump more than 5%?
Yes No
4 55 11 Stock jump more than 5%?
Yes No
114 360 828 Stock jump more than 5%?
Yes No
417 18108 1485
Conclusion
• There are salient statistical patterns of insider trading in the option market.
1. Call-put imbalance is large; 2. Total option volume is high; 3. Slightly in-the-money or out-of-the-money is preferred; 4. Near-term option is preferred.
• By detecting insider trading before the news release, excess returns can be generated.
- Based on 2007 data,
Market return = + 2.82% Our return = + 7.47%