CISC 849 : Applications in Fintech ARUNPRABHU SADAIYAPPAN Dept of Computer & Information Sciences University of Delaware BIG DATA APPROACH TO ANALYZING.

CISC 849 : Applications in Fintech

ARUNPRABHU SADAIYAPPANDept of Computer & Information Sciences

University of Delaware

BIG DATA APPROACH TO ANALYZING MARKET VOLATILITY


WHY BIG DATA TECHNOLOGY?


BIG DATA IN FINANCE SERVICES• Markets

• Customers

• Channels

• Products

• Regulations

• Competitors

• Suppliers

• Employees


MARKET LIQUIDITY

Market's ability to facilitate the purchase or sale of

an asset without causing drastic change in the asset's

price


LIQUIDITY INDICATOR• VOLUME SYNCHRONISED PROBABILITY OF

INFORMATION TRADING


TECHNIQUES FOR BIG DATA ANALYSIS

• EFFICIENT FILE ORGANISATION FOR STORING TRADING RECORDS

• EFFICIENT ALGORITHM FOR COMPUTING VPIN

• PARALLELIZATION OF COMPUTATIONAL TASKS


FILE ORGANIZATION

• 67 Months worth of data (Liquid futures trade information)

• File Format : CSV (Comma separated value)

• File Size : 140 GB

Processing time for computing VPIN values : 142 seconds

CONVENTIONAL METHOD:


FILE ORGANIZATION

• Same amount of data (for 67 months)

• File Format : HDF5 (Hierarchical Data Format)

• File Size : 41 GB

Processing time for computing VPIN values : 0.4 second

PROPOSED METHOD:


FILE ORGANIZATION

• 29% gain in memory storage

• Less time in reading/writing the file

• Better organization for data

Reason behind the Efficiency: HDF5 files store data in binary form

IMPROVEMENT:


ALGORITHM FOR COMPUTING VPINInitially, Data arrives at irregular frequency

Volume BarsVolume Bars

Bulk Volume Bulk Volume ClassificationClassification

BucketsBuckets

VPINVPIN


VOLUME BAR• SHELL SORT – To order trades in a volume

bar

• To compute Median Prices


SHELL SORT ANALYSIS

• BEST CASE (LOWER BOUND): O(N(log N )^2) • WORST CASE (UPPER BOUND): Θ(N^2)

In this paper, the runtime is mentioned as O(log N )

In-place Sorting (Space complexity is O(N) )


ALGORITHM FOR COMPUTING VPIN

Time Taken to construct Volume bars with different nominal prices


BULK VOLUME CLASSIFICATION• Trades are classified as “Buyer-initiated” or

“Seller-initiated”

• BVC assigns a fraction of volume as buys

The remainder as sells based on normalized sequential price change


BUCKETS

• When forming buckets Each volume bar is considered as a single trade with the nominal price

• 30 Volume bars in a buckets (maximum upto 50)

• Most recent buckets with buy and sell volumes Kept in Fixed Memory


BUCKETS

Pseudocode for storing Volume in Buckets


CALCULATING VPIN

FORMULA:

PSEUDOCODE:


EXPERIMENTAL RESULTS

Statistics on prices for volume bars and the resulting VPIN:

(Overall trades of ES)


EXPERIMENTAL RESULTS


PARALLELIZATION

PREFERRED IMPLEMENTATION: POSIX THREADS

Instrument with large number of trades assigned first


HEDGE ANALYTICS – SWISS BASED STARTUP

•RISK FACTOR ANALYSIS

•PERFORMANCE ANALYSIS

•NEW CUSTOMER WIN RATE ANALYSIS

Universal Meta data Search Engine

Incorporated NoSQL Technology to avoid inconsistency


THANK YOU!

CISC 849 : Applications in Fintech ARUNPRABHU SADAIYAPPAN Dept of Computer & Information Sciences University of Delaware BIG DATA APPROACH TO ANALYZING.

Documents