The Penn-Lehman
Automated Trading Project
Michael Kearns
Computer and Information Science
University of Pennsylvania
PLAT development team: Luis Ortiz, Berk
Kapicioglu, Byoungjoon Kim, Rashid Tuweiq
What Is It?
• A project in real-time, automated trading in
financial markets
• A realistic, multi-client market simulator
using live and historical depth-of-market data
• A competitive testbed for trading strategies
• An investigation of order book strategies
• An investigation of AI & ML methods for
automated trading
• A research and educational partnership
between Penn and Lehman Brothers
Market Microstructure and
Depth-of-Market
• Consider an exchange for some specific security (e.g.
MSFT stock)
• Market order: specify number of shares desired, but
leave price to “the market”
• Limit order: specify number of shares and desired
price (usually away from the market)
• Limit orders are placed in a queue (or “book”) on the
buy or sell side, ordered by price
• Market orders are matched with the top of the book
on the opposing side, giving price
• Market orders “guaranteed” transaction but not price;
limit orders guaranteed price but not transaction
Electronic Crossing Networks (ECNs)
• Matching process as old as financial markets
• Previously done manually (NYSE specialists)
• Electronic Crossing Networks (ECNs):
– Automate matching process
– Now publishing order book data
– Can be highly liquid
• Examples: Island (17% of NASDAQ volume), Instinet,
Archipelago,…
• Consolidation efforts (SuperMontage)
• OUCH and ITCH messaging protocols
• Example: island.htm or http://www.island.com/
The Penn Exchange Simulator (PXS)
• A (conceptually) simple, real-time market
simulator exploiting order book data
• Core processes:
– Data Gathering: frequent (live or historical) polling
of any given stock from www.island.com
– Client Order Processing: acceptance of connections
and limit orders from automated clients via API
– Market Simulation: simulation of matching process in
a virtual market merging the Island and client data;
computation of profit and loss of clients
LAST MATCH
Pric
e
23.7790
9:01:55.61
4
BUY ORDERS
TODAY'S ACTIVITY
Orders
1,630
Volume
44,839
PXS
Time
SELL ORDERS
SHA
RES
PRICE
SHA
RES
PRICE
1,00
0
23.760
0
100
23.780
0
3,08
7
23.750
0
800
23.799
0
200
23.750
0
500
23.800
0
100
23.740
0
1,72
0
23.807
0
1,72
0
23.728
0
900
23.819
0
2,00
0
23.720
0
200
23.850
0
1,00
0
23.700
0
1,00
0
23.850
0
100
23.700
0
1,00
0
23.850
0
100
23.700
0
1,00
0
23.860
0
800
23.697
0
200
24.000
0
500
23.650
0
500
24.000
Client 1
BUY
SELL
V 23.80
I 23.78
I 23.80
I 23.80
V 23.80
I 23.81
V 23.82
V 23.83
V 23.84
I 23.85
I 23.85
I 23.86
V 23.89
V 23.90
V 23.97
V 24.00
V 23.77
I 23.76
I 23.75
I 23.75
V 23.74
I 23.74
V 23.74
V 23.73
I 23.73
Client 2
Client 3
early-snap.txt
later-snap.txt
pxsAPI9.htm
Features of PXS
•
•
•
•
•
•
•
Simulation merges real market data with client
No guesses or models for limit order fills
Permits investigation of order-book strategies
Permits high-speed, high volume trading
Forces real-time performance
Sandbox for diverse strategies & interaction
Execute “live” on real-time data or historical
Design Challenges
•
•
•
•
•
Data volume, network latency, performance
Timing issues in live versus historical
Partial observability of Island queues
Client impact on market
Future functionality:
–
–
–
–
Multi-stock simulation
External data feeds
Web interface and GUI
Multiple ECNs; other markets (futures, options)
The Project: Participants
• Penn Development Team:
–
–
–
–
PXS design, development, maintenance
Client strategy design
Education of student researchers/users
Competition design and execution
• Students:
– Approximately 30 students designing 13 strategies
– Many senior projects; many joint CIS-Wharton
– Several external participants (UTexas and CMU)
• Lehman:
– Financial support from proprietary trading
– Student mentorship and competition judging
Goals for the Trading Strategies
and Competitions
• Attempt to recreate Wall Street “biodiversity”
• Market making, pairs trading, VWAP, block
trading, technical strategies, etc.
• Investigate predictive value of order book data
• Investigate application of AI and ML methods
• Create a library of strategies and competition
structures for mix and match experimentation
Strategy Types
• Many order-book strategies:
– order book imbalance and variations
obi.gif or Island Individual Investors
• Market-making strategies
• Traditional technical strategies:
– various “breakout” strategies, trend spotting, etc.
– often with order-book modulation
• Machine learning strategies using order books
– boosting
– case-based learning
– SVMs
The Competitions
• All have share position limits
• Fall 2002 Historical MSFT Competition
hist2002.html
• Fall 2002 Live MSFT Competition (1-day)
LiveComp.htm
red-smooth250.ps
blue-smooth250.ps
• Order-book Face-off
obonly.html
• Currently: 3-round, 3-week MSFT Competition
FebComp.htm
Further Info
•
•
•
•
General: www.cis.upenn.edu/~mkearns/
PLAT: www.cis.upenn.edu/~mkearns/projects/pat.html
mkearns@cis.upenn.edu
New participants welcome!