A Machine Learning based Pairs Trading Investment Strategy

Contents
Acronyms
1 Introduction
	1.1 Topic Overview
	1.2 Objectives
	1.3 Outline
	References
2 Pairs Trading—Background and Related Work
	2.1 Mean-Reversion and Stationarity
		2.1.1 Augmented Dickey-Fuller Test
		2.1.2 Hurst Exponent
		2.1.3 Half-Life of Mean-Reversion
		2.1.4 Cointegration
	2.2 Pairs Selection
		2.2.1 The Minimum Distance Approach
		2.2.2 The Correlation Approach
		2.2.3 The Cointegration Approach
		2.2.4 Other Approaches
	2.3 Trading Execution
		2.3.1 Threshold-Based Trading Model
		2.3.2 Other Trading Models in the Literature
	2.4 The Application of Machine Learning in Pairs Trading
	2.5 Conclusion
	References
3 Proposed Pairs Selection Framework
	3.1 Problem Statement
	3.2 Proposed Framework
	3.3 Dimensionality Reduction
	3.4 Unsupervised Learning
		3.4.1 Problem Requisites
		3.4.2 Clustering Methodologies
		3.4.3 DBSCAN
		3.4.4 OPTICS
	3.5 Pairs Selection Criteria
	3.6 Framework Diagram
	3.7 Conclusion
	References
4 Proposed Trading Model
	4.1 Problem Statement
	4.2 Proposed Model
	4.3 Model Diagram
	4.4 Time Series Forecasting
	4.5 Autoregressive Moving Average
	4.6 Artificial Neural Network Models
		4.6.1 Long Short-Term Memory
		4.6.2 LSTM Encoder-Decoder
	4.7 Artificial Neural Networks Design
		4.7.1 Hyperparameter Optimization
		4.7.2 Weight Initialization
		4.7.3 Regularization Techniques
	4.8 Conclusion
	References
5 Implementation
	5.1 Research Design
	5.2 Dataset
		5.2.1 Exchange-Traded Funds
		5.2.2 Data Description
		5.2.3 Data Preparation
		5.2.4 Data Partition
	5.3 Research Stage 1
		5.3.1 Development of the Pairs Selection Techniques
		5.3.2 Trading Setup
		5.3.3 Test Portfolios
	5.4 Research Stage 2
		5.4.1 Building the Forecasting Algorithms
		5.4.2 Test Conditions
	5.5 Trading Simulation
		5.5.1 Portfolio Construction
		5.5.2 Transaction Costs
		5.5.3 Entry and Exit Points
	5.6 Evaluation Metrics
		5.6.1 Return on Investment
		5.6.2 Sharpe Ratio
		5.6.3 Maximum Drawdown
	5.7 Implementation Environment
	5.8 Conclusion
	References
6 Results
	6.1 Data Cleaning
	6.2 Pairs Selection Performance
		6.2.1 Eligible Pairs
		6.2.2 Pairs Selection Rules
		6.2.3 OPTICS Application Results
		6.2.4 Trading Performance
	6.3 Forecasting-Based Trading Model Performance
		6.3.1 Eligible Pairs
		6.3.2 Forecasting Algorithms Training
		6.3.3 Forecasting Performance
		6.3.4 Trading Performance
		6.3.5 Implementation Remarks
		6.3.6 Alternative Methods
	References
7 Conclusions and Future Work
	7.1 Conclusions
	7.2 Future Work