Financial Forecastings using Neural Networks ppt

advertisement
FINANCIAL FORECASTING
USING NEURAL
NETWORKS
What is Financial Forecasting

Prediction of prices of instruments of
speculation
 Stocks
 Commodity futures
 Exchange Rates
 Interest Rates .

Problem : Non linear and non stationary
data
Methods Used

Fundamental Analysis
 Understanding the supply demand curve
 Involves studying of news and economic
factors

Technical Analysis
 Understanding historical price patterns
 Tools like moving average, learning systems

Latest Approach: Combine Technical
and Fundamental Analysis
NEURAL NETWORKS
Map some type of input stream of information to an
output stream of data.
 They derive non-linear models that can be trained to
map past and future values of the input output
relationship .It extracts relationships governing the data
that was not obvious using other analytical tools.
 Capability to recognize pattern and the speed of
techniques to accurately solve complex processes,
exploited exhaustively in financial forecasting.
 Trained without the restriction of a model to derive
parameters and discover relationships, driven and
shaped solely by the nature of the data.

NEURAL NETWORKS V/S CONVENTIONAL
COMPUTERS



Neural networks have the unique capability of
learning thus are adaptive .This problem solving
tool creates a unique likeness to the human brain .
Use the interconnectedness of the elements of the
model rather than follow a set of sequential steps,
that may or may not solve the problem like
computers do.
A different aspect of model building, where the
unique relationships between the variables creates
the model, rather than trying to force variables to
conform to a theoretical abstract that may or may
not exist.
NEURAL NETWORKS IN FINANCE




Neural networks are trained without the restriction of a model to
derive parameters and discover relationships, driven and shaped
solely by the nature of the data. Thus it has profound implications
and applicability to the finance field.
Some of the fields where it is applied are:
Financial forecasting
Capital budgeting and risk management
Stock market analysis
Used to analyze and verify Economic hypothesis and theories
which were not possible otherwise.
Govt. predicts interest rates to gauge the future inflationary situation
of its economy .
Neural Networking and Similarities with the Workings of the Human Brain
A SIMPLE NEURON
VECTOR INPUT TO NEURON
LAYER OF NEURONS
LAYER OF NEURONS …..
MULTIPLE LAYERS
MULTIPLE LAYERS …..
NARX MODEL
TRANSFER FUNCTIONS
TRAINING ALGORITHMS

trainlm : fastest and better for non-linear
cases , default for feed-forwardnet .
BACK-PROPOGATION

Numerous such input/target pairs are
used to train the Neural Network.
TIME SERIES FORECASTING


Time series forecasting or time series prediction,
takes an existing series of data and forecasts the
data values. The goal is to observe or model the
existing data series to enable future unknown data
values to be forecasted accurately.
Done using the NARX model or NAR model .
DIFFICULTIES
Limited quantity of data .
 Noise in data – It obscures the
underlying pattern of the data .
 Non-stationarity - data that do not have
the same statistical properties (e.g.,
mean and variance) at each point in
time
 Appropriate Forecasting Technique
Selection .

Preprocessing of Training Data
Reason: Need to understand underlying
patterns.
 Tools:

 Moving Average
 Fast Fourier Transform (FFT)
 Hilbert Huang Transform (HHT)
Types Of Data Worked Upon



Interest Rates (RBI 91 day Govt. Of India Treasury Bills)
Sensex Data ( 2005-2010)
Exchange Rates (Daily Exchange Rates of INR-Dollars
2004-2011)

All the Data are divided into Three Sets
1.
Training Set
Testing Set
Validation Set
2.
3.
Types Of Preprocessing



No Pre-Processing (Simple NN)
Using FFT (FFT NN)
Using HHT (HHT NN)
All the types of data are used on all the types of
preprocessing techniques , therefore generating 9
cases.
Now, we Compare all of them Data-Wise.
1. Interest Rates


The interest rate data is applied on all three kinds of preprocessing.
The Error Graphs are as:
Simple NN

FFT NN

HHT NN
2. Sensex Data


The sensex data is applied on all three kinds of
preprocessing. The Error Graphs are as:
Simple NN

FFT NN

HHT NN
3. Exchange Rates

The Exchange Rate data is applied on all three kinds of
preprocessing. The Error Graphs are as:

Simple NN

FFT NN

HHT NN
Conclusion from Results

Pre-processing can boost the Neural
Network Performance

The performance of Neural Network also
depends on the nature of the data series
Thank You
Download