Transcript group1(AI_in_finance)

Applicatons of AI in finance

Amit

Anshum

Pratyush

Siddharth
The AI view of money

''Money is just a type of information, a pattern
that, once digitized, becomes subject to
persistent programmatic hacking by the
mathematically skilled. As the information of
money swishes around the planet, it leaves in
its wake a history of its flow, and if any of that
complex flow can be anticipated, then the
hacker who cracks the pattern will become a
rich hacker." -- from Cracking Wall Street
Why Computers?




Computers can process lot more information
per unit time than we can, without getting tired
Computers can recognize patterns in data
easily
Computers can do calculations for you, so that
you can work at a higher level of abstraction
You don't have to pay a computer on an yearly
basis
Areas where AI is applied

Financial Data mining

Arbitrage Opportunities

Hedging and Trading Strategies

Financial Time Series Forecasting

Supply Chain Management

Fraud Detection
Arbitrage

Arbitrage: is an investment, where there is no
chance of loss in any case (state), and a positive
cash inflow in atleast one case.

Liquid market: Minimal Arbitrage opportunities

For example: In India: 1 Euro = 65Rs, 1$ = 50 Rs

In US: 1 Euro= 1.5$

Purchase Euros from India, sell them in US to get
$, sell them back in India. Sure Profit!!
How can we make MONEY



Arbitrage opportunities are mostly present after
following a long chain of relationships
In an efficient market, arbitrage opportunities
exist for very small periods of time
Can be taken advantage of, using fast
computers, and launching automatic trades
Statistical Arbitrage -- Casinos



The arbitrage opportunity, which are true in
expectations, i.e. In the long run, repeating a
trading strategy
In financial markets, wherever statistical
arbitrage is used, it involves hundreds and
thousands of transactions of various securities
over short holding periods, days to seconds.
Clearly, we need intelligent systems to gain
from them.
Online Auctions



Various bidding strategies possible: Bid
shading, Chandelier binding
Data needs to be processed on the fly
Complicated models to select a good Opening
Bid

Probabalistic models

Need for intelligent systems

False-name bids possible: Leveled division set
protocol
Genetic Algorithms for our aid




Genetic Algorithms : Good for optimization
problems.
Provide quick acceptable solution
Particularly good for noisy and discontinuous
functions appearing so frequently in market
modelling and asset allocation
Also very good for combinatorial optimisation
Genetic Algorithms

GAs work with a population of ”individuals”

Fitness Score of Individuals


”Fit individuals” are given opportunities to
reproduce by ”cross breeding”. Least fit
members ”die out”
A well designed GA, converged to optimal
solution of the problem
Genetic Algorithms: Method Overview



Evaluation Function: Provides a measure of
performance wrt the set of parameters
Fitness Fuction: Provides a relative measure
of fitness using the evaluation function.
Generally it is the ratio of my evaluation function
to the avg of evaluation function
Each individual gets to place number of copies
in the population depending upon the ratio.
Higher your ratio, more you represent.
Genetic Algorithms: Method Overview



Recombination & Mutation:Take any two
parent strings, choose a 1 point crossover.
Swap the strings on either side & mutate with
some low probability.
The recombination probabilities depend on the
type of coding which you choose for the
problem.
Mutation is done so that no point in the search
space has zero probability of being examined.
AI in Financial Data Mining and
Manufacturing



What is the role of AI in data mining?
What is the nature of its contribution towards
Business?
What is the role of an intelligent machines in
manufacturing?
AI in Data Mining



Data mining is the process of extracting hidden
patterns and useful knowledge from a set of raw data.
Computers come into picture when the data is too
large to be analysed manually and when greater
speed and accuracy is required.
Modern computers have largely enhanced data mining
by use of sophisticated tools and complex algorithms.
An important part of this is performing complex
calculations in feasible time.
Automated data mining in Finance




The need for data mining in finance arises due
to the following (and many others) :
Benefit from short-term subtle patterns.
Read the impact of market players on market
regularities.
Make coordinated multi resolution forecast
(minutes,days,weeks,months,and years).
AI in manufacturing



AI provides the edge required to stay in
competition in today's highly competitive
market.
On the factory floor, Artificial Intelligence will
enable machines of automated reasoning thus
providing solutions to manufacturing problems
during the production process.
Automatic scheduling of manufacturing
operations helps in better utilization of
resources.
Practical applications of AI in
manufacturing.


Nissan and Toyota, for example, are modeling
material flow throughout the production floor
that a manufacturing execution system applies
rules to in sequencing and coordinating
manufacturing operations.
Many automotive plants use rules-based
technologies to optimize the flow of parts
through a paint cell based on colors and
sequencing, thus minimizing spray-paint
changeovers.
Benefits of AI in Manufacturing

Production Scheduling

Advanced Planning and Scheduling

Production Reporting

Inventory Management

Accounting

Capacity Planning

Materials Requirements Planning

Process Control.
How AI has fared so far


Abundance of data in financial market and
diversity of the requirements provide a suitable
environment for testing the data mining
techniques and models.
Since 1990 there has been a huge revolution in
application of AI in business and
manufacturing.AI has become a mainstream
phenomenon and has largely benefited those
who have adopted it.
Fraud detection



Fraud cases has a severe impact on company
profit and reputation.
Number of fraud cases are increasing day by
day.
Fraud detection might need to be done at real
time,For example:Consider the case of credit
card company.In this case fraud must be
detected while transaction going on.
Expert system in fraud detection.


Although a given case may look
legal,Experienced expert may tell that it is the
case of fraud
We can Extract the experience of the expert
and put them into the system.
Rule Based Expert System


Rule Based Expert System work on set of rules
given to it(fraud rule),Based on experts
experience.
For example:If pin for ATM card is entered
wrongly for more than three times,An expert
system might detect the possibility of fraud.
Share and Confidence of Rule


We define the share of fraud rule as the
percentage of fraud cases which is covered by
the rule.
Share of fraud rule does say about acurracy of
the rule.
Confidence of fraud rule



Some non-fraud cases may also be flagged as
a case of fraud,which may lead to wrong
diagnosis.
We define the confidence of the fraud rule
as:number of misused cases covered by the
rule/total number of cases covered by the rule
More confidence means greater accuracy and
less false alarm.
Problem with rule based System


Number of rules increases substantially over
the years,slowing the process of fault detection
Rules valid few years ago might not be valid
now or may be of very little use,Which might still
be there in the system.
Fraud Detection using neural
Networks


fraud detection in many operation falls neatly in
principle within the scope of pattern recognition
procedures.Hence neural network as fraud
detection technique is a good option
Neural Networks can even detect new types of
fraud
Problems with Neural Networks




Number of fraud cases as compared to legal
cases is very low.
Difficult to collect data and training set for the
network.
Data set are given in different ratio of fraud
cases to legal cases,then it occur in practice.
neural network will start flagging legal cases as
the case of fraud
Market Forecasting

What is forecasting?

Need for forecasting?

What is the role of AI in forecasting?

Applications of forecasting in various domain

What all things Intelligent System still can’t
capture?
Need for forecasting

High incentives

Strategic decision and Policy making

Manage risk

Capture the dynamics of market and complex
patterns in data
Where does AI fit in?


Sum up the experience of seasoned investor
Indicators for different phases of business life
cycle.
Recession consolidation/ fiscal recovery  growth  fiscal decline

Efficient market hypothesis

Different methods of forecast eg. GARCH, ARCH, ARIMA,
Neural Networks.
Flow Diagram and basic model of
Neural Network
Data Collection
Data Preprocessing
Extract Test Data Set
Select Network Architecture
Training
Forecasting
Result Analysis
Uses of Intelligent System

Manage Risk eg. Currency market average daily turnover
is $ 3.2 trillion as reported in April 2007.

Building up portfolio eg. Hedge funds, mutual funds,
fund managers use intelligent system to build up portfolio from
different asset classes.

Forecast future returns.

Analyze “risk-reward” ratio.

Trend analysis and pattern recognition.

Trading strategies and economic indicators
eg.Projecting
Inflation and GDP figures.
What all things Intelligent Systems
still can’t capture?



Market sentiments eg. War situations, natural
calamities etc.
Emotional attachment to an investment. eg.
Gold in india people are attached.
Market reaction to scams and scandals eg.
Satyam fraud.
Questions and Answers
QUESTIONS ARE GAURANTEED IN LIFE
ANSWERS AREN’T
Bibliography



Data Mining For Financial Applications. Boris
Kovalerchuk , Central Washington University
USA ; Evgenii Vityaev , Institute of Mathematics
Russian Academy of Sciences Russia
Artificial Intelligence in Manufacturing improving the bottom line. Dawn Tupciauskas,
Tuppas Software Corporation ,2008.
Financial forecasting using neural networks, Ed.
Gately 1996.

Genetic algorithms overview, Franco Busset.

Wikipedia for most of the other references.

R. Brause, T. Langsdorf, M.Hepp: Neural Data
Mining for Credit Card Fraud Detection,IEEE
Int. Conf on Tools with Art. Intell. ICTAI-99,
IEEE Press 1999, pp.103-106