Transcript SciDAC Poster: INCITE

Dynamics of Electronic Markets
J. Siaw, G. Warnecke, P. Jain, C. Kenney, D. Gershman, R. Riedi, K. Ensor
Electronic Markets
1
• Electronic markets (ECNs) are networks that
enable users to place orders for stocks via the
internet to a system that executes trades
automatically (Ex: Island, ArcaBook)
• Difference to Traditional markets:
• Prices determined by users (no market maker)
• Speed Computers used to place orders as well
as for Automated Trading
• Research needed: the dynamics of ECNs are little
studied yet make a large share of trades
Intraday dynamics
4
• 3 trading periods per day:
– Pre-market (7 AM – 9:30 AM)
– Market trading (9:30 AM – 4 PM)
– After hours (4 PM – 7 PM)
• Pre-market and after hours trading is very sparse
Example day:
– Pre-market: 79 trades
– Market trading: 33961 trades
– After hours: 4091 trades
• Trading activity correlates with activity on order
book
• Main period of interest: Market trading
• Issue of interest: Stationarity
6
•Survival Time – time between the
addition and removal of an order
from the queue
• Hazard rate - instantaneous
probability that the order will be
traded or cancelled during the next
instant (before t+t)
• Survival rate - probability an
added order will survive beyond a
certain time
• Competing risks – orders may be
fully or partially traded or
cancelled
Challenges
 Large / Complex Data sets:
Innovative Data processing required
Extremely large files
Stocks accumulated in same file
Impossible to use traditional software
 Existing Market Models outdated?
Statistical analysis required
Ultra – High data frequency
Possibility to cancel order -> Intent unclear:
 actual trade vs
 influencing the market
Not all orders lead to price changes
Previously unseen microstructure detail
3
Exploratory Data Analysis
• Summary statistics
• high volume and liquidity
• 125 Million book entries per day
• 85% of orders placed are cancelled
• orders cancelled within the second
• volume steadily increasing over years
• Features of interest
• spread (recall absence of market maker)
• survival times of orders (recall the short life
span of the majority)
5
•
•
•
•
•
Spread
price –price of last trade (unique to stock)
Limit Orders: queued to be matched or cancelled
best bid - the highest buy order in the queue
best ask – the lowest sell order in the queue
spread = (best ask) – (best bid)
• Highly traded stocks (ex. Microsoft):
•spread usually $0.01; deviates only shortly
• Less traded stocks: spread usually larger
• Issues of interest:
• absence of market maker shows in 1c-spread
• requires new models
Heat Plot of Sell Orders
Heat Plot of Sell Orders
Heat Plot of Buy Orders
• Issue of interest: mixture of risk
and complexity of data require new
models
7
2
Survival of Orders
Research questions
•What makes the price
•Identify orders impacting the price
•Effect of order attributes
•Volume
•Timing
•What drives the market
• Depending on the “state” of the market,
what are the order dynamics
• Identify trader “states” (motivations)
•Effect of exogenous events
•Incorporation of additional microstructure
information into existing models
8
Future Work
• Survival analysis
• Competing risk models
• Cox proportional hazard model
• Self-excitation (ACD models)
• Spread
• Correlation
• ARIMA time series
• Self-excitation
• Conditional / Hidden parameter model
• Arbitrage
• Sensitivity to networks