Predicting Forex Currency
Exchange Rate
using Machine Learning
Project Proposal
Group 14
TEAM MEMBERS
E/18/375 Vindula K.P.A.
E/18/330 Sewwandi H.R.
E/18/323 Seekkubadu H.D.
SUPERVISOR
Dr. Suneth Namal Karunarathna
2
Introduction
Introduction
Our Research aims to use Machine Learning techniques to predict
currency fluctuation patterns in the forex market over a relatively short
period of time.
Downtrend Uptrend
3
FOREX Market
● The forex market/ foreign exchange market, is a global
decentralized marketplace where currencies are traded.
● The largest and most liquid financial market in the world, with a
daily trading volume exceeding $6 trillion
● Trading involves the simultaneous buying of one currency and
selling another.
● Foreign exchange rates represent the value of one currency in
terms of another and determine the exchange ratio between two
currencies
Ex : Currency Pairs - EUR/USD : 1.0787 At close: February 9 10:29PM GMT
4
● Factors Influencing Exchange Rates:
➔ Interest Rates
➔ Economic Indicators
➔ Political Stability
➔ Market Sentiment
➔ Trade Balances
All of these factors fall under two main categories:
- Macroeconomic/fundamental indicators
- Technical indicators
5
Introduction
Motivation
Why Exchange Rate Prediction is important ?
● Risk Management
● Cost Reduction
● Strategic Decision Making
● Investment Opportunities
● Forecasting Economic Trends
● Trade Balance Management
6
Methodology
Various models are utilized in methodologies for forex currency exchange
rate prediction.
Key Models:
● Convolutional Neural Networks (CNN)
● Recurrent Neural Networks (RNN)
● Long Short-Term Memory (LSTM)
7
Convolutional Neural Network (CNN)
8
Example:
Financial Time Series Forecasting – A Deep Learning Approach
Making use of CNNs to forecast the next period price direction with respect to
the current price by Alexiei Dingli and Karl Sant Fournier xie
Forecasting Model: Develop a CNN-based model using TensorFlow, with
adjustments for optimal performance.
● Two convolutional and pooling layers :
○ same padding with stride of 1
○ Max-pooling
○ ReLu activation function
Calculating Accuracy: Argmax function provided by the TensorFlow library
9
Fine tuning the network by adjusting:
● the depth of the network
● the Learning rate
● the size of the local receptive area
● the number of features for each convolution
Achieved an accuracy of 65% when forecasting the next month price direction
and 60% for the next week price direction forecast.
10
Recurrent Neural Network (RNN)
RNN architecture
11
Example:
Using Recurrent Neural Networks To Forecasting of Forex
Trained Recurrent Neural Networks forecast the exchange rates between
American Dollar and four other major currencies, Japanese Yen, Swiss Frank,
British Pound and EURO.
Data: Taken from www.gelium.net and www.forexite.com
Method: Elman-Jordan neural network has been used. Since it is the recurrent
neural net which learns the rules in the time series, this method was
successfully used for prediction of financial markets.
Attempts were taken to improve the achieved results in different directions.
● Model with two added technical indicators
● Model with only RSI and %R in input
● Model with RSI, %R and moving average in input
● Model with indicators, returns and exponential moving average
12
Result:
● The best results have been shown by the network, which had two indicators,
moving average and returns in input.
● Replaced moving average by exponential moving average, since it
approximates time series more accurate and contains information on longer
time period.
Conclusion: statistical estimates of Forex forecast indicates, that neural
network can predict the increments sign with relatively high probability –
approximately 80%, which is quite sufficient for practical use.
13
Long Short-Term Memory (LSTM)
● Cell: The cell stores the state of a
sequence, so it has the ability to
either keep or forget certain
information.
● Input Gate: It decides the extent of
information to be stored in the cell.
● Output Gate: It determines what
the next hidden state should be.
● Forget Gate: It decides what
information should be thrown away
or kept.
14
Example:
Using Long Short-Term Memory To Forecasting of Forex
1. Macroeconomic LSTM model
○ Interest rates of Germany and the EU
○ FED funds rate (for the US)
○ Inflation rates in the EU and the US
○ Close value of the S&P 500 market index
○ Close value of the DAX market index
2. Technical LSTM model
○ Moving Average with a period of 10
○ Moving Average Convergence Divergence with short- and long-term periods of 12 and 26.
○ Rate of Change with a period of 2
○ Momentum with a period of 4
○ RSI with a period of 10
○ BB with period of 20
○ CCI with a period of 20
3. Hybrid LSTM Model
15
Results:
16
Results:
17
Key insights from the research
1. Can major trends in price fluctuations of a currency pair be predicted in
advance?
2. Can the start point of that price fluctuation be predicted with acceptable
accuracy?
3. Can the end point of that price fluctuation be predicted with acceptable
accuracy?
18
Use of AI Tools
To :
● Extract important points from the complex contents
● Explore new topics and contents
● Spellings and grammar checking
19
Use of AI Tools
Pros: Cons:
● Efficient Content Extraction ● Risk of Dependency
● Facilitates Topic Exploration ● Limited Interpretation
● Improves Communication ● Potential Inaccuracy
● Accuracy Assurance ● Privacy and Accessibility
Concerns
20
Demonstration
21
Data sources and visualization
Alpha Vantage API - financial data provider that offers a wide range of
financial data and services primarily targeted at investors, traders, and
developers
22
Demo - Resources
Colab Files
https://colab.research.google.com/drive/1q-kW3mBJP8HPbWOE-ZIM5EJ
ZJcICS7bL?usp=sharing
https://colab.research.google.com/drive/126F2ax3BBxFgVXSIsgjXicLo6jm
OX1lC?usp=sharing
Data File
https://drive.google.com/file/d/1ipsRBgv8aChxObX0liPDapDsxsC_LNdB/v
iew?usp=sharing
23
Resources
GitHub Repository
cepdnaclk/e18-4yp-Predicting-Forex-Currency-Exchange-Rate-using-M
achine-Learning
Project Page
Projects | Department of Computer Engineering (cepdnaclk.github.io)
24
Q&A
25
Thank You!
10 26
