Transcript Slide 1

ONE TICK
®
Accelerating Quant Research and Trading
Principal Component Analysis &
Multi-Factor Modeling Tests
with OneTick & R
Historical & Real-Time
Maria Belianina, Director of Pre-Sales Engineering
Jianwen Luo, Director, Trading System Engineering
7 Minute Crash Course
ONE TICK
®
Accelerating Quant Research and Trading
What is OneTick?
ONETICK time series database & analytics
Tick data management and super fast analytics for Finance.
Capture, store, retrieve and analyze real-time and historical tick data
for any asset class, any size & period of time, any granulairty
ONETICK CEP real-time analytics
Low latency Complex Event Processing seamlessly integrating the
analysis of real-time streaming and historical market data
ONETICK reference data file + OneQuantData
Smooth historical time series data with Corporate actions, take into
account symbol name changes, look at earnings, daily prices and
more
Markets
Markets
Real-time market feeds
Collectors
In-memory
Database:
History
Archives:
Today’s
TAQ
TAQ
TAQ
Loaders
History
History:
Archives:
Reference History
Data
Data
CEP
Server
Analytics
Tick
Server
Analytics
ONE TICK
®
Accelerating Quant Research and Trading
Who is using OneTick and why?
Our clients:
 Hedge Funds & Proprietary
Trading Firms
 Market Makers
 Large Asset Managers
 Banks & Brokers
 Marketplaces / Exchanges
 Technology & Information
Providers
 Universities
Business Cases:
 Backtesting & Quantitative
Research
 High frequency trading signal
generation
 Pre- & Post- Trade TCA
 Venue Analysis
 Backbone for
Charting / Time and Sales
 Compliance & Regulatory
Reporting
 Risk, Portfolio Analytics, PnL
 Generic time series analysis
OneTick GUI: Query Language
Runs
Historical
(for research
& backtesting)
or
Real-Time
(alerts & signal
generation)
Trades
Query Example:
Bollinger Bands
Buy/Sell Signals
A “Nested
query”
for Bollinger
Bands
calculations
NOTE: One of the nodes can be a
custom code in R or C++, C#,
Java, Python, Perl, MatLab
OneTick+R
Sample Use
Case:
Principal Component Analysis (PCA)
Based Multi-Factor Trading
 Based on “Developing High-Frequency Equities Trading
Models” by Leandro Infantino and Savion Itzhaki, 2010
 Claim: By filtering out “noise” in a portoflio of stocks
through PCA, we might be able to “predict the future” and
make some gains
 Testing approach:
OneTick + R
1.
OneTick price and quote history. Portfolio of stocks.
Test 1 day at a time.
2.
Define PCAWindowSize (e.g., 15 min);
Run PCA on the initial PCAWindowSize;
Run regression with given MemoryWindowSize (e.g., 30
sec); Calculate betas
3.
For the rest of the day: Run PCA over sliding
PCAWindowSize buckets; Forecast using calculated betas,
Try to generate trading signals; Update betas per
MemoryWindowSize bucket; keep track of returns and PnL
OneTick+R Testing approach
BBO.. MID.. Log Returns
Use nested query
_get_log_return
as a source
Merge
all portfolio
log returns
and
transpose
into matrix
“Nested” query
Pass sliding
sub-matrices
into R
for PCA, etc
OneTick+R: Built-In R Integration
Generic
OneTick
bucket
aggregation
parameters
Full Strategy in OneTick+R
- Work in Progress 
Sample ETA is next month.
Contact us for details.
R in-process
call
parameters
PCA Based Statistic Multi-Factor Trading
Initial Tests
 Preliminary test results. Testing: 1 day of TAQ for
December 15th 2010 quotes, 25-symbol portfolio.
To be tested further…
7000
6000
5000
4000
3000
14
2000
12
1000
10
0
0
0.5
1
1.5
2
Cumulative PnL assuming
position limit of 1 unit
per stock
2.5
x 10
8
4
6
4
2
0
0
0.5
1
1.5
2
2.5
x 10
4
Cumulative Returns assuming
equal weights
ONE TICK
®
Accelerating Quant Research and Trading
STOP BY OUR STAND
FOR A LIVE DEMO!
THANK YOU
Contacts:
[email protected]